Health economic evaluation of a serum-based blood test for brain tumour diagnosis: exploration of two clinical scenarios

Objectives To determine the potential costs and health benefits of a serum-based spectroscopic triage tool for brain tumours, which could be developed to reduce diagnostic delays in the current clinical pathway. Design A model-based health pre-trial economic assessment. Decision tree models were constructed based on simplified diagnostic pathways. Models were populated with parameters identified from rapid reviews of the literature and clinical expert opinion. Setting Explored as a test in both primary and secondary care (neuroimaging) in the UK health service, as well as application to the USA. Participants Calculations based on an initial cohort of 10 000 patients. In primary care, it is estimated that the volume of tests would approach 75 000 per annum. The volume of tests in secondary care is estimated at 53 000 per annum. Main outcome measures The primary outcome measure was quality-adjusted life-years (QALY), which were employed to derive incremental cost-effectiveness ratios (ICER) in a cost-effectiveness analysis. Results Results indicate that using a blood-based spectroscopic test in both scenarios has the potential to be highly cost-effective in a health technology assessment agency decision-making process, as ICERs were well below standard threshold values of £20 000–£30 000 per QALY. This test may be cost-effective in both scenarios with test sensitivities and specificities as low as 80%; however, the price of the test would need to be lower (less than approximately £40). Conclusion Use of this test as triage tool in primary care has the potential to be both more effective and cost saving for the health service. In secondary care, this test would also be deemed more effective than the current diagnostic pathway.

At an average of 20 years, patients with malignant brain tumours have the highest number of years of life lost, compared to all other primary cancers. 1 This, at least in part, may relate to diagnostic delays, reflecting the non-specific early symptoms, such as headache and dizziness, from which general practitioners (GPs) must identify patients at risk for further investigation. The perceived lack of a low-cost diagnostic and/or screening tools available within the health service also impacts on this delay. We have recently demonstrated that a spectroscopic test using blood-serum is able to effectively identify brain tumours in patients with sensitivities and specificities as high as 91·5% and 83% respectively. 2,3 This approach is based upon Fourier-transform infrared (FTIR) spectroscopy and can detect disease-specific signatures, which are extracted mathematically using pattern recognition and machine learning algorithms.

Current Diagnostic Pathway
Currently, patients with symptomatic brain tumour visit their GP on average five times before being referred to secondary care. 4 Partly as a result of this diagnostic delay up to 61% of brain tumour diagnoses occur in an emergency setting, often following a seizure. 5 Patients diagnosed by the emergency route have a poorer prognosis. 6,7 For some patients this may be because the disease is often of a more advanced stage at diagnosis, and the complication precipitating the emergency brings addition mortality per se.
Screening programmes for breast, prostate, cervical, and colorectal cancer have proved effective for diagnosing patients at an earlier stage, which can result in a better prognosis. [8][9][10][11] These screening programmes have had a significant impact in reducing the number of patients presenting as an emergency. There is no accessible and economically viable diagnostic tool for early detection of asymptomatic and symptomatic brain tumours. 12 The addition of a rapid and accurate blood test for patients with suspected primary brain tumours therefore has the potential to improve outcomes by allowing prioritisation of patients most at risk for further investigation. Under the current patient pathway it is not feasible to provide fast-track diagnostic imaging because the number of patients with non-specific headache symptoms is very large and the positive predictive value (PPV) on the basis of symptoms alone is less than 3% for all symptoms other than a new-onset seizure. 13,14 Magnetic resonance imaging (MRI) and computed tomography (CT) are the current gold standard for identifying structural brain lesions including tumours. Treatment decisions made at the neuro-oncology multidisciplinary team (MDT) meeting are often based on the imaging alone, and following surgical resection or biopsy with histopathology and molecular analysis, definitive treatment can be planned. 15 Surgery to secure the tissue diagnosis has a small risk to the patient of neurological deterioration and death. 16 In additional to the patient risk, the diagnostic pathway also represents a significant cost burden to the health service, with a single MRI and CT scan in the UK costing around £164 and £85, respectively (National Schedule of Reference Costs , Medicare Physician Fee Schedule 2016). A typical timeframe of the diagnostic pathway for brain tumours, specifically primary gliomas, is  Figure 1, and effectively highlights the significant wait that a symptomatic patient may have before receiving brain imaging. Even from this stage, regardless of the time to GP referral or emergency presentation, full diagnosis may take a further 5 weeks.

Serum Spectroscopic Diagnostics
This novel blood test for early brain tumour detection is based upon the interaction of infrared (IR) light with biological components of blood serum (Figure 2). Specifically, using attenuated total reflectance (ATR)-FTIR spectroscopy, specific bond vibrations of given molecules can be elucidated from serum samples, thus providing a unique insight into the composition via an absorbance spectrum. 17 Benefits of an ATR-FTIR based approach include a robust, user-friendly methodology without extensive sample preparation, that would readily fit into a clinical setting. 18 In short, serum is obtained according to standard protocols and can be snap frozen and stored at -80° until the point of analysis. A small volume of serum is required for analysis (1-5 µL), which is pipetted onto a crystal, known as an internal reflection element (IRE), where IR non-destructively interacts with the sample and produces an IR spectrum, with peaks representative of known bond vibrations and hence biomolecular constituents.  Blood serum is a complex medium that contains a variety of biomolecules, including around 20,000 proteins, that may be employed as diagnostic biomarkers. 19 In the case of brain tumours, such blood based technologies are limited, due to a lack of an established brain tumour specific diagnostic biomarker. 20 With our spectroscopic approach, rather than derive single biomarker specific information, a global signature is obtained, which encompasses the entire biomolecular makeup. This is epitomised as an equally complex biological absorbance spectrum, which contains a wealth of diagnostic information (an example may be seen in Figure 3).  Pattern recognition and machine learning algorithms using spectral features from FTIR data have been demonstrated as rapid and accurate for separating primary brain cancer and noncancer cases. 2,21 When these algorithms are applied to this information rich dataset, the relationship between all biological components of the sample is ascertained, providing a multi-dimension analysis of the sample. This approach has been able to not only detect between cancer and non-cancer, but also stratify based upon cancer pathology. 2 The ability to triage patients likely to have a brain tumour based on serum sample alone raises the possibility of systematic triaging prior to investigation with more expensive (MRI/CT imaging) and invasive (biopsy) tests. One major impact of having a serum test available would be a clear reduction in the number of unnecessary brain scans; however, as this test is also able to differentiate between primary and secondary tumours, there could also be a knock-on reduction of chest and abdomen scans which are conducted to rule out primary disease elsewhere. Ultimately, it is expected that this could allow earlier and potentially more effective treatment of brain tumour.

Aims & Objectives
The aim of the economic evaluation is to assess the potential cost-effectiveness of this spectroscopic technology, in advance of any prospective study results being available. There are three main objectives for the evaluation. The first objective is to create a map of where the test could be used in the clinical pathway. The second is to assess the potential costeffectiveness of the technology, if the performance shown in the case-control study is replicated prospectively. This will give an indication of whether the technology would meet the criteria for acceptance for use in the National Health Service that are applied by Health Technology Assessment agencies (HTA). Related to this, the third objective is to define the level of performance in prospective trials, and any additional evidence that would be needed to meet the cost-effectiveness criteria of an HTA decision-making process. This can include diagnostic performance and also effects on long-term outcomes such as survival and resource use. To achieve these objectives a simplified economic model of two important clinical scenarios is used to explore cost effectiveness. In order to appreciate the current clinical pathway and determine an appropriate entry point for a serum spectroscopic test, a pan-UK clinical focus group was established. This cohort included neurosurgeons, clinical and medical oncologists, neuropathologists, neuroradiologists, academic GPs with special interests, and experts in primary care diagnostics (see Appendix 1).

Cost Effectiveness Analysis
A cost-effectiveness analysis (CEA) was conducted to calculate the effects on health outcomes and health service costs of introducing spectroscopic testing in each of two scenarios. The health outcomes considered were life-years and quality-adjusted life-years (QALYs).
A decision tree was used to model the pathway for patients presenting with symptoms warranting a referral for MRI/CT imaging for suspected brain tumour (Figure 4). Separate models were considered for primary and secondary care. The time horizon of the model is 2 years and the perspective is that of the health care service. A two-year horizon was selected because of the short duration of survival in this patient group: median survival is approximately 1 year for high grade gliomas, which are the commonest primary brain tumour. In all scenarios the comparator is the current diagnostic pathway (i.e. imaging alone). Further details regarding the node probabilities, simplifications and assumptions of the model can be found in Appendix 2.  . A decision tree model describing the integration of a serum spectroscopy test in the current diagnostic pathway, and the effect on MRI/CT imaging for suspected brain tumour.

Diagnostic Performance
The sensitivity and specificity of the test for detecting brain tumours has been demonstrated in a series of case-control studies using historical samples from the Brain Tumour North West and Walton Centre NHS bio-banks. 2,3 In the key case-control study nine FTIR spectra were collected from each of 433 patients. 3 Of these 134 were from patients with primary brain tumours (64 high grade glioma), 177 were from patients with cerebral metastases and 122 were from non-cancer controls. FTIR spectra were analysed using the random forest method to fit a classification model. Classification performance was estimated by applying the fitted model to a test set containing 20% of the patients from the original data set that were not used in the model fitting step (hold-out test set). Classification statistics are computed as averages of this process, iterated 96 times using random training and test sets. Under the best available model, sensitivity estimated by this method was 92·8% and specificity was 91·5% for the analysis of cancerous vs. non-cancerous serum. 3  Establishing whether the performance demonstrated case-control data from historical samples translates to equivalent performance when applied prospectively in clinical practice is the subject of a planned clinical trial. This is critical to demonstrating the clinical and costeffectiveness of a serum spectroscopy test as part of the diagnostic pathway for brain tumours.

Sensitivity Analysis
Incremental cost effectiveness ratios (ICER) were calculated for scenarios 1 and 2 using the base case parameter estimates. Base case analysis was repeated for UK and USA settings. Additional sensitivity analyses are reported for the UK setting only.
Sensitivity analyses included one-way sensitivity analysis (OWSA), systematically varying a single parameter in the model, and scenario analysis in which specific model assumptions were altered. OWSA were conducted for test sensitivity, specificity and test cost. Scenario analyses included assuming an additional consultation cost for discussion of test results, F o r p e e r r e v i e w o n l y assuming a higher proportion of patients continue to imaging following a negative spectroscopy result in primary care, and using mean survival rather than median survival.
A probabilistic sensitivity analysis (PSA) was conducted to explore the effects of joint uncertainty in the parameter estimates on the model results. 30

Mapping the clinical pathway
For the first study objective, initial discussion with clinical experts indicated that there are potential uses for the test in both primary and secondary care. The main advantage of employing a cost-effective spectroscopic blood test in the diagnostic pathway is to use it as a triage test. This prioritises more urgent cases for access to services, and acts as a gate-keeper (requiring a positive result in some conditions to give access to services) for imaging studies.
Two clinical scenarios are mapped out below and subsequently explored in this early economic evaluation of the serum spectroscopy test for aiding diagnosis of brain tumours:

Triage tool in primary care
The primary care scenario explores a population of patients with a clinical presentation that warrants further investigation of possible brain tumour. This would include some patients with headaches and some with focal neurological deficits. This is the group of patients who would be considered for direct access imaging, using MRI or CT, where this is available and referral to neurology where it is not. 31 The blood test is used to provide rapid information, within 24 hours, where a positive result would lead to patients receiving more timely access to imaging. It may also be the case that negative test results, in addition to establishing the low probability of a brain tumour, could also provide some reassurance for those patients that must wait for imaging. The total volume of tests would be approximately 75,000 per year in the UK (see Appendix 4 for further details).

Triage tool in secondary care
In this scenario the population is the group of patients that are currently referred for imaging studies from secondary care for suspected brain tumour, typically via neurology clinics. This is the patient group for whom the clinical presentation has the highest positive predictive value (PPV). However, even in this high risk group, the odds of a brain tumour being present are approximately 1:33. [22][23][24] Again, the spectroscopy test is used to provide rapid information to allow a subset of these patients to access immediate imaging and provide reassurance to other patients who may have to wait longer for definitive imaging studies and diagnosis. The extent of the benefits of triage in this scenario is likely to vary by locality depending on the capacity constraints on imaging and pathology services. This evaluation uses estimates of the delays in diagnosis, and potential improvements in the speed of diagnosis, from a consecutive patient case series in London, UK. 5 The total volume of tests if this scenario occurred would be approximately 53,000 per year in the UK (see Appendix 4 for further details).

Cost Effectiveness Assessment
The standard threshold value per QALY gained in the UK, is considered to be between £20,000 to £30,000. Below this value, a healthcare intervention may be considered cost effective, whereas a negative ICER value would be deemed cost saving. Base case results for primary care (scenario 1) and secondary care (scenario 2) are presented in Table 2. Note results are reported for cohorts of 10,000 patients. Incremental cost-effectiveness ratios were well below standard threshold values of £20,000 to £30,000 per quality adjusted life year (QALY) gained used in the UK, and similar thresholds used internationally, provided the test cost did not exceed £100. The base case results demonstrate that the serum spectroscopy test dominates (more effective and less costly) standard care at the lower bound of test cost in the primary care setting in both the UK and USA. At the upper bound of test cost the ICERs may be within commonly used thresholds or, in the case of the USA, remain dominant to standard care. In the secondary care setting ICERs of £9,982 and $10,153 at the lower bounds of test cost indicate that this test is potentially cost-effective in this setting. At the upper bounds ICERs may still be within commonly used thresholds for cost-effectiveness.

Sensitivity Analysis Results
The performance of the test with regards to levels of sensitivity and specificity are addressed using sensitivity analysis. OWSA results for a range of test specificities are displayed in  Figure 4, for primary and secondary care respectively, displaying the ICER with varying test specificity. Note that the estimated QALYs do not change with specificity in the model therefore changes in the ICER are due solely to changes in incremental costs. Varying sensitivity changes both estimated QALYs and estimated costs therefore results of the OWSA for test sensitivity are presented on the cost-effectiveness plane (Appendix 5). In primary care, using the upper cost limit of £100 it is evident that the test is deemed cost effective at specificities of around 0·9 and above, where the ICER is below standard thresholds. In contrast, at the lower cost limit, the test is cost effective at specificities above 0·8. Although the serum spectroscopy test is not cost saving at low cost (or near perfect specificities), the test is still considered cost effective at specificity levels around 0·7 and 0·8 for £50 and £100 pricing respectively.   The OWSA results highlight how ICERs in scenario 2 (secondary care) are strongly influenced by test sensitivity while ICERs in scenario 1 (primary care) are more strongly influenced by test specificity. These features are a result of the varying prevalence of disease and the assumption in scenario 2 that 100% of patients are referred for imaging following negative result. It should be noted that test specificity is important in both scenarios.
Relatively small improvements in test specificity can substantially change the ICER, while larger improvements in test sensitivity are required to substantially alter the ICERs.
Additional scenario analyses are also reported in Appendix 5. These demonstrate that results are robust to using mean survival estimates rather than median survival estimates and including additional consultation costs for positive test results. If the prevalence of brain tumours in Scenario 1 is 1% rather than 0·5% the incremental QALYs increase substantially and the ICERs are reduced.

Discussion
This economic evaluation establishes the potential for serum spectroscopy to have a role in the diagnosis of both benign and malignant brain tumours in both primary and secondary care. The potential costs and health benefits of testing using a spectroscopic method prior to CT/MRI tests (or in some scenarios to avoid imaging) have been estimated based on a mathematical model with parameter values taken from published studies and expert opinion. This diagnostic tool is sensitive to all brain tumours (benign or malignant), however, this assessment is closely aligned with the diagnosis of primary gliomas, where there is a maximum potential benefit to the health service.
When used as a triage tool in primary care, this novel test has the potential to deliver improvements in health outcomes and also to reduce costs. At the lower end of test costs, the technology would be cost-saving for the health service. At higher test costs the technology is still likely to be considered cost-effective in HTA agency decision processes.
In Scenario 2, in which serum spectroscopy is used as a triage tool in secondary care, the technology will create additional costs but also produce sizable health benefits. At test costs of under £100 ($200) the technology would be likely to be considered as a cost-effective use of resources in HTA agency decision processes in the UK (and USA).
Sensitivity analyses have demonstrated the importance of diagnostic performance on the costeffectiveness of the test. In particular, test specificity is important in the primary care setting. If test specificity is 87·5% or worse the technology may not be considered cost-effective at higher values of assumed test cost. This is due to the increased number of false positive results in this low prevalence population, generating a greatly increased proportion of fasttrack imaging studies which increases costs.
To strengthen the case that this approach represents a cost-effective use of healthcare resources it is necessary to establish the diagnostic performance of the test prospectively. This can be accomplished by a suitably large cohort study in which serum spectroscopy is used alongside current clinical practice in one of the patient groups included in this evaluation. It would be appropriate to initially target the secondary care patient population, because the higher prevalence of disease in this group will reduce the sample size needed to accurately estimate diagnostic performance. Several results in this analysis suggest cost savings through reduced use of imaging for patients with a negative test result. To make the case that a serum spectroscopic test can improve the efficiency of the diagnostic pathway prospective studies will also need to explore the impact of these test results on clinician and patient imaging study decisions. The possibility remains that the test may triage patients, but may not reduce the number of scans being conducted, and could potentially increase the demand on imaging. However, this triaging of patients would still benefit each patient that is provided with an early diagnosis.
This evaluation has explored the potential for serum spectroscopy to be a cost-effective addition to the diagnostic pathway for brain tumours. It has demonstrated that in specific scenarios this novel test may be an effective and cost-effective technology in reducing the delay to diagnosis for patients with brain tumours. Prospective trials are required to provide definitive evidence.

Appendix 2 -Decision tree model
The decision node probabilities that populate the decision tree model are reported in Table S1. The model is simplified by assuming that the reference test (MRI/CT) has perfect accuracy (100% sensitivity and 100% specificity). This is desirable if we wish to evaluate the test against an MRI/CT only diagnostic pathway in which these imaging studies are considered as the 'gold standard'. It is also difficult to obtain valid estimates of MRI/CT sensitivity and specificity for this indication as these are not reported in a comparable manner in the literature. This simplification will not substantially alter the economic results compared to using more realistic values for the scenarios under consideration. This is because it is assumed confirmation of the diagnosis always requires imaging and therefore outcomes for patients that would hypothetically be classified correctly using this test but incorrectly by MRI/CT will remain the same whether or not the test is used. The effect of less than perfect accuracy of the reference test is therefore the same as a (small) decrease in prevalence as a proportion of patients with the disease cannot benefit from the additional test due to incorrect subsequent testing. Outcomes are calculated by 'rolling-back' the tree for both cancer cases and non-cases and taking an average of the two groups weighted by the prevalence of brain tumours in the scenario population.

Effect of testing on time-to-diagnosis and time-to-treatment
To estimate the expected time-to-diagnosis under a fast-track pathway it is assumed that this would match the current median time-to-diagnosis for patients presenting with brain tumour in emergency care as observed in Aggarwal et al 5 . This study reported a cohort of high grade glioma patients from a single hospital in London, UK. No other suitable estimates of time-to-treatment or time-to-diagnosis were identified in the literature. This is the most common primary brain tumour in the indications considered in this evaluation. Median time-todiagnosis for the standard pathway is also taken from the same source.

Effect of testing on use of imaging studies
The effect of spectroscopic testing on imaging study decisions of clinicians and patients is uncertain. Based on clinical expert input it was assumed, in the base case, that in secondary care all patients would continue to imaging, while in primary care 50% would continue to imaging following a negative spectroscopy result.

Effect of early diagnosis on patient outcomes
A systematic review of the literature failed to identify any studies that directly estimated the effect of earlier (or later) diagnosis on outcomes for primary brain tumours. Instead, based clinical expert guidance, estimates of the effects of time-to-treatment were sourced from the literature. The estimated effects are based on fitting natural history models to observational datasets of high grade glioma patients. 27,28 It is necessary to use a natural history model to produce survival estimates rather than using the survival data stratified by time-to-treatment directly, in order to adjust for potential confounding factors influencing both time-to-treatment and survival.
A one-to-one correspondence is assumed for the effects of an additional day added to the time-to-diagnosis and an additional day added to the time-to-treatment. Median survival by weeks between diagnosis and treatment are displayed in Table 2. The hazard ratio per additional day delay calculated in Do et al of 1·015 was applied to each day between diagnosis and treatment up to 28 days to calculate median survival time. Time between diagnosis and treatment beyond 28 days were fixed at the same median survival as 28 days.

Resource use and cost
It is assumed that there is no additional cost for converting a small proportion of standard referral to additional urgent referrals. This could be justified on the basis that patients receiving negative results allows more flexibility in the scheduling for these patients. This could create more opportunity to schedule patients with positive spectroscopy results sooner. As a larger proportion of patients are referred for immediate imaging it was assumed costs would increase because achieving this would require additional capacity. It was assumed that unit costs for MRI and CT grow exponentially with increasing proportion of immediate referrals such that costs double if 50% of patients are referred for immediate imaging.
Treatment costs are assumed to be the same for both fast-track and standard referrals that are diagnosed as cases. This may be a conservative assumption as treatment costs are likely to increase as disease progresses and therefore earlier treatment may, on average, be less costly than later treatment. However, it is also possible that  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   treatment costs could be increased if some cases are converted from palliative care to more aggressive treatment. The balance of these effects is unknown given the lack of appropriate data to estimate the effects. Secondary treatment and end-of-life costs are similarly assumed to be equal between fast track and standard referrals. This reflects the model assumption that all patients ultimately progress.
The time intervals considered (0-4 weeks) are too short for discounting to have an important impact on cost estimates therefore the effects of moving forward treatment on present value is not considered.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  The Glasgow estimate is particularly low and may result from limited uptake in the early part of the 8 year period for which data were reported. As it was not possible to determine if uptake increased over the period or was stable this estimate was not considered reliable. The Nottingham and Lothian estimates report potentially more stable and established services and are therefore more likely to reliably estimate demand. The likely demand for this novel test, if the indication is considered identical to neuroimaging, is therefore between 54,000 and 93,000. This is likely to be conservative as the test may well have a slightly broader indication than current neuroimaging. A reasonable point estimate may be approximately 75,000.

Secondary Care
An estimate of the total demand for serum spectroscopic tests in the UK can be made based on the reported incidence of brain tumour in the UK and the proportion of cases currently diagnosed through secondary (nonemergency) care. Annual incidence of malignant brain tumours in the UK is approximately 4,700. 35 Approximately one third of these are likely to present through secondary, non-emergency care. Using the positive predictive value for patients currently referred to neuroimaging in this setting of 3% suggests that for each case there will be 33 non-case patients receiving neuroimaging. In total, in the UK, this would imply 53,000 patients per year in secondary care may be suitable for spectroscopic testing.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y Appendix 5 -Sensitivity Analysis

Appendix 2 -Decision tree model
The decision node probabilities that populate the decision tree model are reported in Table S1. The model is simplified by assuming that the reference test (MRI/CT) has perfect accuracy (100% sensitivity and 100% specificity). This is desirable if we wish to evaluate the test against an MRI/CT only diagnostic pathway in which these imaging studies are considered as the 'gold standard'. It is also difficult to obtain valid estimates of MRI/CT sensitivity and specificity for this indication as these are not reported in a comparable manner in the literature. This simplification will not substantially alter the economic results compared to using more realistic values for the scenarios under consideration. This is because it is assumed confirmation of the diagnosis always requires imaging and therefore outcomes for patients that would hypothetically be classified correctly using this test but incorrectly by MRI/CT will remain the same whether or not the test is used. The effect of less than perfect accuracy of the reference test is therefore the same as a (small) decrease in prevalence as a proportion of patients with the disease cannot benefit from the additional test due to incorrect subsequent testing. Outcomes are calculated by 'rolling-back' the tree for both cancer cases and non-cases and taking an average of the two groups weighted by the prevalence of brain tumours in the scenario population.

Effect of testing on time-to-diagnosis and time-to-treatment
To estimate the expected time-to-diagnosis under a fast-track pathway it is assumed that this would match the current median time-to-diagnosis for patients presenting with brain tumour in emergency care as observed in Aggarwal et al 5 . This study reported a cohort of high grade glioma patients from a single hospital in London, UK. No other suitable estimates of time-to-treatment or time-to-diagnosis were identified in the literature. This is the most common primary brain tumour in the indications considered in this evaluation. Median time-todiagnosis for the standard pathway is also taken from the same source.

Effect of testing on use of imaging studies
The effect of spectroscopic testing on imaging study decisions of clinicians and patients is uncertain. Based on clinical expert input it was assumed, in the base case, that in secondary care all patients would continue to imaging, while in primary care 50% would continue to imaging following a negative spectroscopy result.

Effect of early diagnosis on patient outcomes
A systematic review of the literature failed to identify any studies that directly estimated the effect of earlier (or later) diagnosis on outcomes for primary brain tumours. Instead, based clinical expert guidance, estimates of the effects of time-to-treatment were sourced from the literature. The estimated effects are based on fitting natural history models to observational datasets of high grade glioma patients. 27,28 It is necessary to use a natural history model to produce survival estimates rather than using the survival data stratified by time-to-treatment directly, in order to adjust for potential confounding factors influencing both time-to-treatment and survival.
A one-to-one correspondence is assumed for the effects of an additional day added to the time-to-diagnosis and an additional day added to the time-to-treatment. Median survival by weeks between diagnosis and treatment are displayed in Table 2. The hazard ratio per additional day delay calculated in Do et al of 1·015 was applied to each day between diagnosis and treatment up to 28 days to calculate median survival time. Time between diagnosis and treatment beyond 28 days were fixed at the same median survival as 28 days.

Resource use and cost
It is assumed that there is no additional cost for converting a small proportion of standard referral to additional urgent referrals. This could be justified on the basis that patients receiving negative results allows more flexibility in the scheduling for these patients. This could create more opportunity to schedule patients with positive spectroscopy results sooner. As a larger proportion of patients are referred for immediate imaging it was assumed costs would increase because achieving this would require additional capacity. It was assumed that unit costs for MRI and CT grow exponentially with increasing proportion of immediate referrals such that costs double if 50% of patients are referred for immediate imaging.
Treatment costs are assumed to be the same for both fast-track and standard referrals that are diagnosed as cases. This may be a conservative assumption as treatment costs are likely to increase as disease progresses and therefore earlier treatment may, on average, be less costly than later treatment. However, it is also possible that  34 Assuming this was representative of the UK generally then the demand for this type of service would be approximately 54,000 for the UK. The Glasgow estimate is particularly low and may result from limited uptake in the early part of the 8 year period for which data were reported. As it was not possible to determine if uptake increased over the period or was stable this estimate was not considered reliable. The Nottingham and Lothian estimates report potentially more stable and established services and are therefore more likely to reliably estimate demand. The likely demand for this novel test, if the indication is considered identical to neuroimaging, is therefore between 54,000 and 93,000. This is likely to be conservative as the test may well have a slightly broader indication than current neuroimaging. A reasonable point estimate may be approximately 75,000.

Secondary Care
An estimate of the total demand for serum spectroscopic tests in the UK can be made based on the reported incidence of brain tumour in the UK and the proportion of cases currently diagnosed through secondary (nonemergency) care. Annual incidence of malignant brain tumours in the UK is approximately 4,700. 35 Approximately one third of these are likely to present through secondary, non-emergency care. Using the positive predictive value for patients currently referred to neuroimaging in this setting of 3% suggests that for each case there will be 33 non-case patients receiving neuroimaging. In total, in the UK, this would imply 53,000 patients per year in secondary care may be suitable for spectroscopic testing.

Appendix 5 -Sensitivity Analysis
One-way sensitivity analysis (OWSA) OWSA results for a range of test sensitivity are displayed in S1 and S2 as a series of points on a costeffectiveness (CE) plane with incremental QALYs (intervention-control) on the x-axis and incremental costs (intervention -control) on the y-axis. Corresponding OWSA results for a range of test specificities in primary and secondary care are shown in Figure 3 and 4 in the main article text respectively. These are displayed with the ICER on the y-axis and the test specificity on the x-axis. Note that the estimated QALYs do not change with specificity in the model therefore changes in the ICER are due solely to changes in incremental costs.Error! Reference source not found. All OWSA analyses show results assuming the spectroscopy based test costs of £50 and £100.  of methodological assumptions (such as discount rate, study perspective). 20b Model-based economic evaluation: Describe the effects on the results of uncertainty for all input parameters, and uncertainty related to the structure of the model and assumptions. Characterising heterogeneity 21 If applicable, report differences in costs, outcomes, or costeffectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.

Copyright/Licence for publication
The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, a worldwide licence to the Publishers and its licensees in perpetuity, in all forms, formats and media (whether known now or created in the future), to i) publish, reproduce, distribute, display and store the Contribution, ii) translate the Contribution into other languages, create adaptations, reprints, include within collections and create summaries, extracts and/or, abstracts of the Contribution, iii) create any other derivative work(s) based on the Contribution, iv) to exploit all subsidiary rights in the Contribution, v) the inclusion of electronic links from the Contribution to third party material where-ever it may be located; and, vi) licence any third party to do any or all of the above

Authors' Contributions
EG conducted the health economic assessment including data analysis, data interpretation, and writing. HJB contributed to data interpretation, writing and providing details regarding clinical spectroscopy. MJB is the principal investigator of this research, and initiated this interdisciplinary project and team.

Role of the Funding Source
Scottish Enterprise is Scotland's main economic development agency and non-departmental public body of the Scottish Government. They work with partners in both the private and public sectors to exploit the best opportunities to deliver a significant, lasting effect on the Scottish economy. The High-Growth Spinout Programme helps researchers take their ideas and inventions from the lab to the global marketplace. To determine the costs and health benefits of a serum-based spectroscopic triage tool for brain tumours, that could potentially reduce diagnostic delays in the current clinical pathway.

Design
A model-based health economic assessment. Decision tree models were constructed based on simplified diagnostic pathways. Models were populated with parameters identified from rapid reviews of the literature and clinical expert opinion.

Setting
Explored as a test in both primary and secondary care (neuroimaging) in the UK health service, as well as application to the US.

Participants
Calculations based upon an initial cohort of 10,000 patients. In primary care, it is estimated that the volume of tests would approach 75,000 per annum. The volume of tests in secondary care is estimated at 53,000 per annum.

Main Outcome Measures
The primary outcome measure was quality adjusted life years (QALYs), which were employed to derive incremental cost-effectiveness ratios (ICERs) in a cost-effectiveness analysis.

Results
Results indicate that using a blood-based spectroscopic test in both scenarios may be considered highly cost-effective in a health technology assessment (HTA) agency decision making process, as ICERs were well below standard threshold values of £20,000 to £30,000 per QALY. This test may be cost-effective in both scenarios with test sensitivities and specificities as low as 80%; however, the price of the test would need to be lower (less than approximately £40).

Conclusion
Use of this test as triage tool in primary care has the potential to be both more effective and cost saving for the health service. In secondary care, this test would also be deemed more effective than the current diagnostic pathway.

Strengths and Limitations of this Study
• Simplified models of clinical pathways were mapped with input from, and consensus among, a wide range of clinical experts including neurosurgeons, neurooncologists, neuropathologists, neuroradiologists, and primary care experts.
• The spectroscopic blood test was highly sensitive and specific in retrospective data, with performances of 91.5 and 83.0% respectively. There is potential for this to contribute towards improved prognosis for patients, as well as healthcare savings.
• This study is based upon proof-of-concept studies, in advance of a pending prospective clinical trial. As these samples are retrospective, there is the possibility the diagnostic performance will not be as high in prospective studies.
• A lack of clinical trial evidence necessitates the estimate of long-term benefits of improved diagnostic protocols based on disease natural history models. This creates additional uncertainties.
• The precise patient population for whom the test may be suitable in the primary care setting is difficult to establish at this stage in development. This study considers a limited definition of eligibility that may need revision in light of future evidence.

Introduction
At an average of 20 years, patients with malignant brain tumours have the highest number of years of life lost, compared to all other primary cancers. 1 This, at least in part, may relate to diagnostic delays, reflecting the non-specific early symptoms, such as headache and dizziness, from which general practitioners (GPs) must identify patients at risk for further investigation. The lack of a low-cost diagnostic and/or screening tools available within the health service contributes to this delay. We have recently demonstrated that a spectroscopic test using blood-serum is able to effectively identify brain tumours in patients with sensitivities and specificities as high as 91·5% and 83% respectively. 2,3 This approach is based upon Fourier-transform infrared (FTIR) spectroscopy and can detect disease-specific signatures, which are extracted mathematically using pattern recognition and machine learning algorithms.

Current Diagnostic Pathway
Currently, patients who are symptomatic with a brain tumour visit their GP on average five times before being referred to secondary care. 4 Partly as a result of this diagnostic delay up to 61% of brain tumour diagnoses occur in an emergency setting, often following a seizure. 5 Patients diagnosed by the emergency route have a poorer prognosis. 6,7 For some patients this may be because the disease is at a more advanced stage at diagnosis. The complications precipitating the emergency admission make an additional contribution to mortality.
Screening programmes for breast, prostate, cervical, and colorectal cancer have proved effective for diagnosing patients at an earlier stage, which can result in a better prognosis. [8][9][10][11] These screening programmes have had a significant impact in reducing the number of patients presenting as an emergency. To date there has been no accessible and economically viable diagnostic tool for early detection of asymptomatic and symptomatic brain tumours. 12 The addition of a rapid and accurate blood test for patients with suspected primary brain tumours (symptomatic patients) therefore has the potential to improve outcomes by allowing prioritisation of patients most at risk of a brain tumour for further investigation. Under the current patient pathway it is not feasible to provide fast-track diagnostic imaging because the number of patients with non-specific headache symptoms is very large and the positive predictive value (PPV) on the basis of symptoms alone is less than 3% for all symptoms other than a new-onset seizure. 13,14 Magnetic resonance imaging (MRI) and computed tomography (CT) are the current gold standard for identifying structural brain lesions including tumours. Treatment decisions made at the neuro-oncology multidisciplinary team (MDT) meeting are often based on the imaging alone. Following surgical resection or biopsy with histopathology and molecular analysis, definitive treatment can be planned. 15 Surgery to secure the tissue diagnosis has a small risk to the patient of neurological deterioration and death. 16 The diagnostic pathway also represents a significant cost burden to the health service, with a single MRI and CT scan in the UK costing around £164 and £85, respectively (National . A typical timeframe of the diagnostic pathway for brain tumours, specifically primary gliomas, is illustrated in Figure 1, and effectively highlights the significant wait that a symptomatic patient may have before receiving brain imaging. Even from this stage, regardless of the time to GP referral or emergency presentation, full diagnosis may take a further 5 weeks.

Serum Spectroscopic Diagnostics
This novel blood test for early brain tumour detection is based upon the interaction of infrared (IR) light with biological components of blood serum (Figure 2). Specifically, using attenuated total reflectance (ATR)-FTIR spectroscopy, specific bond vibrations of given molecules can be elucidated from serum samples, thus providing a unique insight into the composition via an absorbance spectrum. 17 Benefits of an ATR-FTIR based approach include a robust, user-friendly methodology without extensive sample preparation, that would readily fit into a clinical setting. 18 In short, serum is obtained according to standard protocols and can be snap frozen and stored at -80° until the point of analysis. A small volume of serum is required for analysis (1-5 µL), which is pipetted onto a crystal, known as an internal reflection element (IRE), where IR non-destructively interacts with the sample and produces an IR spectrum, with peaks representative of known bond vibrations and hence biomolecular constituents.
Blood serum is a complex medium that contains a variety of biomolecules, including around 20,000 proteins, that may be employed as diagnostic biomarkers. 19 In the case of brain tumours, such blood based technologies are limited, due to a lack of an established brain tumour specific diagnostic biomarker. 20 With our spectroscopic approach, rather than derive single biomarker specific information, a global signature is obtained, which encompasses the entire biomolecular makeup. This is epitomised as an equally complex biological absorbance spectrum, which contains a wealth of diagnostic information (an example may be seen in Figure 3).
Pattern recognition and machine learning algorithms using spectral features from FTIR data have been demonstrated as rapid and accurate for separating primary brain cancer and noncancer cases. 2,21 When these algorithms are applied to this information rich dataset, the relationship between all biological components of the sample is ascertained, providing a multi-dimensional analysis of the sample.
This approach has been able to not only detect between cancer and non-cancer, but also stratify based upon cancer pathology. 2 For further information and in-depth description of the methodology, we direct the readers to the following fundamental review and recent research papers. 2,3,17,21 The ability to triage patients likely to have a brain tumour based on serum sample alone raises the possibility of systematic triaging prior to investigation with more expensive (MRI/CT imaging) and invasive (biopsy) tests. One major impact of having a serum test available would be a clear reduction in the number of unnecessary brain scans; however, as this test is also able to differentiate between primary and secondary tumours, there could also be a knock-on reduction of chest and abdomen scans which are conducted to rule out primary disease elsewhere. There is also the possibility that this approach will reduce the incidence of incidental abnormalities which in themselves can cause considerable distress. Ultimately, it is expected that this could allow earlier and potentially more effective treatment of brain tumours.

Aims & Objectives
The aim of the economic evaluation is to assess the potential cost-effectiveness of this spectroscopic technology, in advance of any prospective study results being available. There are three main objectives for the evaluation. The first objective is to create a map of where the test could be used in the clinical pathway. The second is to assess the potential costeffectiveness of the technology, if the performance shown in the case-control study is replicated prospectively. This will give an indication of whether the technology would meet the criteria for acceptance for use in the National Health Service that are applied by Health Technology Assessment agencies (HTA). Related to this, the third objective is to define the level of performance in prospective trials, and any additional evidence that would be needed to meet the cost-effectiveness criteria of an HTA decision-making process. This can include diagnostic performance and also effects on long-term outcomes such as survival and resource use. To achieve these objectives a simplified economic model of two important clinical scenarios is used to explore cost effectiveness.

Mapping the Clinical Pathway
In order to appreciate the current clinical pathway and determine an appropriate entry point for a serum spectroscopic test, a pan-UK clinical focus group was established. This cohort included neurosurgeons, clinical and medical oncologists, neuropathologists, neuroradiologists, academic GPs with special interests, and experts in primary care diagnostics (see Appendix 1).

Cost Effectiveness Analysis
A cost-effectiveness analysis (CEA) was conducted to calculate the effects on health outcomes and health service costs of introducing spectroscopic testing in each of two scenarios. The health outcomes considered were life-years and quality-adjusted life-years (QALYs).
A decision tree was used to model the pathway for patients presenting with symptoms warranting a referral for MRI/CT imaging for suspected brain tumour (Figure 4). Separate models were considered for primary and secondary care. The time horizon of the model is 2 years and the perspective is that of the health care service. A two-year horizon was selected because of the short duration of survival in this patient group: median survival is approximately 1 year for high grade gliomas, which are the commonest malignant primary brain tumour. In all scenarios the comparator is the current diagnostic pathway (i.e. imaging alone). Further details regarding the node probabilities, simplifications and assumptions of the model can be found in Appendix 2.

Diagnostic Performance
The sensitivity and specificity of the test for detecting brain tumours has been demonstrated in a series of case-control studies using historical samples from the Brain Tumour North West and Walton Centre NHS bio-banks. 2,3 In the key case-control study nine FTIR spectra were collected from each of 433 patients. 3 Of these 134 were from patients with primary brain tumours (64 high grade glioma), 177 were from patients with cerebral metastases and 122 were from non-cancer controls. FTIR spectra were analysed using the random forest method to fit a classification model. Classification performance was estimated by applying the fitted model to a test set containing 20% of the patients from the original data set that were not used in the model fitting step (hold-out test set). Classification statistics are computed as averages of this process, iterated 96 times using random training and test sets. Under the best available model, sensitivity estimated by this method was 92·8% and specificity was 91·5% for the analysis of cancerous vs. non-cancerous serum. 3 These classification statistics were established using 96 independent iterations of a Random Forest model, and resulted in standard deviations of 1.1% and 1.9% respectively.
Establishing whether the performance demonstrated case-control data from historical samples translates to equivalent performance when applied prospectively in clinical practice is the subject of a planned clinical trial. This is critical to demonstrating the clinical and costeffectiveness of a serum spectroscopy test as part of the diagnostic pathway for brain tumours.

Prevalence of disease
Prevalence data were sourced from the literature based on clinical expert guidance. Brain tumour prevalence in scenario 2 (secondary care) was assumed to be 3% based on observed rates of primary brain tumour diagnosis among patients referred for brain imaging for suspected cancer in secondary care. [22][23][24] In scenario 1 (primary care) an estimated prevalence of 0·5% is used based on case-control evidence and expert opinion of the prevalence among patients who would be considered for direct access imaging, using MRI or CT, where this is available and referral to neurology where it is not. 25,26 An alternative prevalence of 1% was explored for scenario 1 based on unpublished data from a direct access imaging service in the UK (P. Brennan, personal communication).
The effect of serum spectroscopy testing on the time-to-diagnosis and time-to-treatment is discussed in Appendix 3, additional to the effect of testing on the use of imaging studies, and also on the patient outcomes. The primary assumptions in this model are that first, the expected time-to-diagnosis would match the current median time-to-diagnosis for patients presenting with brain tumour in emergency care as observed in Aggarwal et al. 5 Furthermore, based upon expert opinion, it is assumed that in secondary care all patients would continue to imaging, while in primary care 50% would continue to imaging following a negative spectroscopy result. This is a conservative estimate associated with the possibility that an imaging test will still be required in some cases based on interpretation of a patient's symptoms and the other non-tumour diagnoses being considered by the clinician. Finally, the effects of early diagnosis on the outcome of brain tumours is estimated using literature describing fitting natural history models to observational datasets of high grade glioma patients. 27,28 Utility Weights Health state utility weights are applied to life-years to generate quality adjusted life years (QALYs). A systematic review of health state utility weights for high grade glioma, the most common and aggressive primary brain tumour, was conducted to identify suitable utility weights. Due to heterogeneity it was not considered suitable to pool the estimates. The most appropriate health state utility weight was taken from a previous UK economic evaluation of glioblastoma treatment. 29 A value of 0·89 was used in the base case.

Resource Use and Costs
Resource use includes the application of a spectroscopic serum test to all patients prior to imaging, the imaging studies used in the diagnostic process, outpatient neurology clinic visits and general practitioner visits. In the UK analysis, unit costs for imaging studies are taken from UK NHS reference costs (2014/15), clinic and GP visits from the PSSRU costs schedule (Table 1). In the USA analysis, unit costs are taken from Medicare reimbursement schedules. Unit costs of the test were applied at an upper bound and lower bound rather than a single value as these products have not yet been commercialized. Bounds were set by consultation

Incremental cost effectiveness ratios (ICER)
The comparative cost-effectiveness of spectroscopic testing compared to no testing is summarized by the ICER defined as: and ‫ܥ‬ are the total costs with spectroscopic testing and no testing respectively. Equivalently ‫ܪ‬ ௦ and ‫ܪ‬ are the total QALYs with and without spectroscopic testing. The ICER can be interpreted as the additional cost per QALY gained. Sensitivity analyses included one-way sensitivity analysis (OWSA), systematically varying a single parameter in the model, and scenario analysis in which specific model assumptions were altered. OWSA were conducted for test sensitivity, specificity and test cost. Scenario analyses included assuming an additional consultation cost for discussion of test results, assuming a higher proportion of patients continue to imaging following a negative spectroscopy result in primary care, and using mean survival rather than median survival.
A probabilistic sensitivity analysis (PSA) was conducted to explore the effects of joint uncertainty in the parameter estimates on the model results. 30

Mapping the clinical pathway
For the first study objective, initial discussion with clinical experts indicated that there are potential uses for the test in both primary and secondary care. The main advantage of employing a cost-effective spectroscopic blood test in the diagnostic pathway is to use it as a triage test. This prioritises more urgent cases for access to services, and acts as a gate-keeper (requiring a positive result in some conditions to give access to services) for imaging studies.
Two clinical scenarios are mapped out below and subsequently explored in this early economic evaluation of the serum spectroscopy test for aiding diagnosis of brain tumours:

Triage tool in primary care
The primary care scenario explores a population of patients with a clinical presentation that warrants further investigation of possible brain tumour. This would include some patients with headaches and some with focal neurological deficits. This is the group of patients who would be considered for direct access imaging, using MRI or CT, where this is available and referral to neurology where it is not. 31 The blood test is used to provide rapid information, within 24 hours, where a positive result would lead to patients receiving more timely access to imaging. It may also be the case that negative test results, in addition to establishing the low probability of a brain tumour, could also provide some reassurance for those patients that must wait for imaging. The total volume of tests would be approximately 75,000 per year in the UK (see Appendix 4 for further details).

Triage tool in secondary care
In this scenario the population is the group of patients that are currently referred for imaging studies from secondary care for suspected brain tumour, typically via neurology clinics. This is the patient group for whom the clinical presentation has the highest positive predictive value (PPV). However, even in this high risk group, the odds of a brain tumour being present are approximately 1:33. [22][23][24] Again, the spectroscopy test is used to provide rapid information to allow a subset of these patients to access immediate imaging and provide reassurance to other patients who may have to wait longer for definitive imaging studies and diagnosis. The extent of the benefits of triage in this scenario is likely to vary by locality depending on the capacity constraints on imaging and pathology services. This evaluation uses estimates of the delays in diagnosis, and potential improvements in the speed of diagnosis, from a consecutive patient case series in London, UK. 5 The total volume of tests if this scenario occurred would be approximately 53,000 per year in the UK (see Appendix 4 for further details).

Cost Effectiveness Assessment
The standard threshold value per QALY gained in the UK, is considered to be between £20,000 to £30,000. Below this value, a healthcare intervention may be considered cost effective, whereas a negative ICER value would be deemed cost saving. Base case results for primary care (scenario 1) and secondary care (scenario 2) are presented in Table 2. Note results are reported for cohorts of 10,000 patients.

Sensitivity Analysis Results
The performance of the test with regards to levels of sensitivity and specificity are addressed using sensitivity analysis. OWSA results for a range of test specificities are displayed in Figure 5 and 6, for primary and secondary care respectively, displaying the ICER with varying test specificity. Note that the estimated QALYs do not change with specificity in the model therefore changes in the ICER are due solely to changes in incremental costs. Varying sensitivity changes both estimated QALYs and estimated costs therefore results of the OWSA for test sensitivity are presented on the cost-effectiveness plane (Appendix 5). In primary care, using the upper cost limit of £100 it is evident that the test is deemed cost effective at specificities of around 0·9 and above, where the ICER is below standard thresholds. In contrast, at the lower cost limit, the test is cost effective at specificities above 0·8. Although the serum spectroscopy test is not cost saving at low cost (or near perfect specificities), the test is still considered cost effective at specificity levels around 0·7 and 0·8 for £50 and £100 pricing respectively.
The OWSA results highlight how ICERs in scenario 2 (secondary care) are strongly influenced by test sensitivity while ICERs in scenario 1 (primary care) are more strongly influenced by test specificity. These features are a result of the varying prevalence of disease and the assumption in scenario 2 that 100% of patients are referred for imaging following negative result. It should be noted that test specificity is important in both scenarios.
Relatively small improvements in test specificity can substantially change the ICER, while larger improvements in test sensitivity are required to substantially alter the ICERs.
Additional scenario analyses are also reported in Appendix 5. These demonstrate that results are robust to using mean survival estimates rather than median survival estimates and including additional consultation costs for positive test results. If the prevalence of brain tumours in Scenario 1 is 1% rather than 0·5% the incremental QALYs increase substantially and the ICERs are reduced.
The PSA results reported in Figure 7 and 8 indicate that at a test cost of £50 and an ICER threshold of £30,000 per QALY there is a near 100% probability the serum spectroscopy test is cost-effective in scenario 1 and approximately 90% probability it is cost-effective in scenario 2. The corresponding probabilities at the upper bound cost of £100 are approximately 85% and 75%.

Discussion
This economic evaluation establishes the potential for serum spectroscopy to have a role in the diagnosis of both benign and malignant brain tumours in both primary and secondary care. The potential costs and health benefits of testing using a spectroscopic method prior to CT/MRI tests (or in some scenarios to avoid imaging) have been estimated based on a mathematical model with parameter values taken from published studies and expert opinion. This diagnostic tool is sensitive to all brain tumours (benign or malignant), however, this assessment is closely aligned with the diagnosis of primary gliomas, where there is a maximum potential benefit to the health service.
When used as a triage tool in primary care, this novel test has the potential to deliver improvements in health outcomes and also to reduce costs. At the lower end of test costs, the technology would be cost-saving for the health service. At higher test costs the technology is still likely to be considered cost-effective in HTA agency decision processes.
In Scenario 2, in which serum spectroscopy is used as a triage tool in secondary care, the technology will create additional costs but also produce sizable health benefits. At test costs of under £100 ($200) the technology would be likely to be considered as a cost-effective use of resources in HTA agency decision processes in the UK (and USA).
Sensitivity analyses have demonstrated the importance of diagnostic performance on the costeffectiveness of the test. In particular, test specificity is important in the primary care setting. If test specificity is 87·5% or worse the technology may not be considered cost-effective at higher values of assumed test cost. This is due to the increased number of false positive results in this low prevalence population, generating a greatly increased proportion of fasttrack imaging studies which increases costs.
To strengthen the case that this approach represents a cost-effective use of healthcare resources it is necessary to establish the diagnostic performance of the test prospectively. This can be accomplished by a suitably large cohort study in which serum spectroscopy is used alongside current clinical practice in one of the patient groups included in this evaluation. It would be appropriate to initially target the secondary care patient population, because the higher prevalence of disease in this group will reduce the sample size needed to accurately estimate diagnostic performance.
Decision makers are often most interested in patient outcomes, such as survival, rather than intermediate outcomes, such as accuracy or speed of diagnosis (although this latter point is vital for treatment of high grade gliomas). From this perspective, a randomized trial, or a prospective cohort study with extended follow-up, may be required to fully establish the size of survival and quality-of-life benefits of including a serum spectroscopy test in the diagnostic pathway. A trial with primary outcomes relating to survival and quality-of-life would be specific to either the primary or secondary care setting (rather than generalisable to both), would need a large sample size, and would also require a follow-up period to capture survival benefits. In the case of malignant glioma this would require a period of at least 24 Future developments beyond trials such as emerging epidemiological evidence and new technologies should also be included in any future evaluations. It was not possible to foresee and include all such possible scenarios in this early evaluation but that should not preclude assessment in the light of new evidence. Updated analysis should inform any decisions about system wide implementation.
Several results in this analysis suggest cost savings through reduced use of imaging for patients with a negative test result. To make the case that a serum spectroscopic test can improve the efficiency of the diagnostic pathway prospective studies will also need to explore the impact of these test results on clinician and patient imaging study decisions. The possibility remains that the test may triage patients, but may not reduce the number of scans being conducted, and could potentially increase the demand on imaging. However, this triaging of patients would still benefit each patient that is provided with an early diagnosis.
This evaluation has explored the potential for serum spectroscopy to be a cost-effective addition to the diagnostic pathway for brain tumours. It has demonstrated that in specific scenarios this novel test may be an effective and cost-effective technology in reducing the delay to diagnosis for patients with brain tumours. Prospective trials are required to provide definitive evidence.

Appendix 2 -Decision tree model
The decision node probabilities that populate the decision tree model are reported in Table S1. The model is simplified by assuming that the reference test (MRI/CT) has perfect accuracy (100% sensitivity and 100% specificity). This is desirable if we wish to evaluate the test against an MRI/CT only diagnostic pathway in which these imaging studies are considered as the 'gold standard'. It is also difficult to obtain valid estimates of MRI/CT sensitivity and specificity for this indication as these are not reported in a comparable manner in the literature. This simplification will not substantially alter the economic results compared to using more realistic values for the scenarios under consideration. This is because it is assumed confirmation of the diagnosis always requires imaging and therefore outcomes for patients that would hypothetically be classified correctly using this test but incorrectly by MRI/CT will remain the same whether or not the test is used. The effect of less than perfect accuracy of the reference test is therefore the same as a (small) decrease in prevalence as a proportion of patients with the disease cannot benefit from the additional test due to incorrect subsequent testing. Outcomes are calculated by 'rolling-back' the tree for both cancer cases and non-cases and taking an average of the two groups weighted by the prevalence of brain tumours in the scenario population.

Effect of testing on time-to-diagnosis and time-to-treatment
To estimate the expected time-to-diagnosis under a fast-track pathway it is assumed that this would match the current median time-to-diagnosis for patients presenting with brain tumour in emergency care as observed in Aggarwal et al 5 . This study reported a cohort of high grade glioma patients from a single hospital in London, UK. No other suitable estimates of time-to-treatment or time-to-diagnosis were identified in the literature. This is the most common primary brain tumour in the indications considered in this evaluation. Median time-todiagnosis for the standard pathway is also taken from the same source.

Effect of testing on use of imaging studies
The effect of spectroscopic testing on imaging study decisions of clinicians and patients is uncertain. Based on clinical expert input it was assumed, in the base case, that in secondary care all patients would continue to imaging, while in primary care 50% would continue to imaging following a negative spectroscopy result.

Effect of early diagnosis on patient outcomes
A systematic review of the literature failed to identify any studies that directly estimated the effect of earlier (or later) diagnosis on outcomes for primary brain tumours. Instead, based clinical expert guidance, estimates of the effects of time-to-treatment were sourced from the literature. The estimated effects are based on fitting natural history models to observational datasets of high grade glioma patients. 27,28 It is necessary to use a natural history model to produce survival estimates rather than using the survival data stratified by time-to-treatment directly, in order to adjust for potential confounding factors influencing both time-to-treatment and survival.
A one-to-one correspondence is assumed for the effects of an additional day added to the time-to-diagnosis and an additional day added to the time-to-treatment. Median survival by weeks between diagnosis and treatment are displayed in Table 2. The hazard ratio per additional day delay calculated in Do et al of 1·015 was applied to each day between diagnosis and treatment up to 28 days to calculate median survival time. Time between diagnosis and treatment beyond 28 days were fixed at the same median survival as 28 days.

Resource use and cost
It is assumed that there is no additional cost for converting a small proportion of standard referral to additional urgent referrals. This could be justified on the basis that patients receiving negative results allows more flexibility in the scheduling for these patients. This could create more opportunity to schedule patients with positive spectroscopy results sooner. As a larger proportion of patients are referred for immediate imaging it was assumed costs would increase because achieving this would require additional capacity. It was assumed that unit costs for MRI and CT grow exponentially with increasing proportion of immediate referrals such that costs double if 50% of patients are referred for immediate imaging.
Treatment costs are assumed to be the same for both fast-track and standard referrals that are diagnosed as cases. This may be a conservative assumption as treatment costs are likely to increase as disease progresses and therefore earlier treatment may, on average, be less costly than later treatment. However, it is also possible that The Glasgow estimate is particularly low and may result from limited uptake in the early part of the 8 year period for which data were reported. As it was not possible to determine if uptake increased over the period or was stable this estimate was not considered reliable. The Nottingham and Lothian estimates report potentially more stable and established services and are therefore more likely to reliably estimate demand. The likely demand for this novel test, if the indication is considered identical to neuroimaging, is therefore between 54,000 and 93,000. This is likely to be conservative as the test may well have a slightly broader indication than current neuroimaging. A reasonable point estimate may be approximately 75,000.

Secondary Care
An estimate of the total demand for serum spectroscopic tests in the UK can be made based on the reported incidence of brain tumour in the UK and the proportion of cases currently diagnosed through secondary (nonemergency) care. Annual incidence of malignant brain tumours in the UK is approximately 4,700. 35 Approximately one third of these are likely to present through secondary, non-emergency care. Using the positive predictive value for patients currently referred to neuroimaging in this setting of 3% suggests that for each case there will be 33 non-case patients receiving neuroimaging. In total, in the UK, this would imply 53,000 patients per year in secondary care may be suitable for spectroscopic testing.

One-way sensitivity analysis (OWSA)
OWSA results for a range of test sensitivity are displayed in S1 and S2 as a series of points on a costeffectiveness (CE) plane with incremental QALYs (intervention-control) on the x-axis and incremental costs (intervention -control) on the y-axis. Corresponding OWSA results for a range of test specificities in primary and secondary care are shown in Figure 3 and 4 in the main article text respectively. These are displayed with the ICER on the y-axis and the test specificity on the x-axis. Note that the estimated QALYs do not change with specificity in the model therefore changes in the ICER are due solely to changes in incremental costs.Error! Reference source not found. All OWSA analyses show results assuming the spectroscopy based test costs of £50 and £100.  of methodological assumptions (such as discount rate, study perspective). 20b Model-based economic evaluation: Describe the effects on the results of uncertainty for all input parameters, and uncertainty related to the structure of the model and assumptions. Characterising heterogeneity 21 If applicable, report differences in costs, outcomes, or costeffectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.

Copyright/Licence for publication
The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, a worldwide licence to the Publishers and its licensees in perpetuity, in all forms, formats and media (whether known now or created in the future), to i) publish, reproduce, distribute, display and store the Contribution, ii) translate the Contribution into other languages, create adaptations, reprints, include within collections and create summaries, extracts and/or, abstracts of the Contribution, iii) create any other derivative work(s) based on the Contribution, iv) to exploit all subsidiary rights in the Contribution, v) the inclusion of electronic links from the Contribution to third party material where-ever it may be located; and, vi) licence any third party to do any or all of the above

Role of the Funding Source
Scottish Enterprise is Scotland's main economic development agency and non-departmental public body of the Scottish Government. They work with partners in both the private and public sectors to exploit the best opportunities to deliver a significant, lasting effect on the Scottish economy. The High-Growth Spinout Programme helps researchers take their ideas and inventions from the lab to the global marketplace.  To determine the costs and health benefits of a serum-based spectroscopic triage tool for brain tumours, that could potentially reduce diagnostic delays in the current clinical pathway.

Design
A model-based health economic assessment. Decision tree models were constructed based on simplified diagnostic pathways. Models were populated with parameters identified from rapid reviews of the literature and clinical expert opinion.

Setting
Explored as a test in both primary and secondary care (neuroimaging) in the UK health service, as well as application to the US.

Participants
Calculations based upon an initial cohort of 10,000 patients. In primary care, it is estimated that the volume of tests would approach 75,000 per annum. The volume of tests in secondary care is estimated at 53,000 per annum.

Main Outcome Measures
The primary outcome measure was quality adjusted life years (QALYs), which were employed to derive incremental cost-effectiveness ratios (ICERs) in a cost-effectiveness analysis.

Results
Results indicate that using a blood-based spectroscopic test in both scenarios may be considered highly cost-effective in a health technology assessment (HTA) agency decision making process, as ICERs were well below standard threshold values of £20,000 to £30,000 per QALY. This test may be cost-effective in both scenarios with test sensitivities and specificities as low as 80%; however, the price of the test would need to be lower (less than approximately £40).

Conclusion
Use of this test as triage tool in primary care has the potential to be both more effective and cost saving for the health service. In secondary care, this test would also be deemed more effective than the current diagnostic pathway.

Strengths and Limitations of this Study
• Simplified models of clinical pathways were mapped with input from, and consensus among, a wide range of clinical experts including neurosurgeons, neurooncologists, neuropathologists, neuroradiologists, and primary care experts.
• The spectroscopic blood test was highly sensitive and specific in retrospective data, with performances of 91.5 and 83.0% respectively. There is potential for this to contribute towards improved prognosis for patients, as well as healthcare savings.
• This study is based upon proof-of-concept studies, in advance of a pending prospective clinical trial. As these samples are retrospective, there is the possibility the diagnostic performance will not be as high in prospective studies.
• A lack of clinical trial evidence necessitates the estimate of long-term benefits of improved diagnostic protocols based on disease natural history models. This creates additional uncertainties.
• The precise patient population for whom the test may be suitable in the primary care setting is difficult to establish at this stage in development. This study considers a limited definition of eligibility that may need revision in light of future evidence.  1 This, at least in part, may relate to diagnostic delays, reflecting the non-specific early symptoms, such as headache and dizziness, from which general practitioners (GPs) must identify patients at risk for further investigation. The lack of a low-cost diagnostic and/or screening tools available within the health service contributes to this delay. We have recently demonstrated that a spectroscopic test using blood-serum is able to effectively identify brain tumours in patients with sensitivities and specificities as high as 91·5% and 83% respectively. 2,3 This approach is based upon Fourier-transform infrared (FTIR) spectroscopy and can detect disease-specific signatures, which are extracted mathematically using pattern recognition and machine learning algorithms.

Current Diagnostic Pathway
Currently, patients who are symptomatic with a brain tumour visit their GP on average five times before being referred to secondary care. 4 Partly as a result of this diagnostic delay up to 61% of brain tumour diagnoses occur in an emergency setting, often following a seizure. 5 Patients diagnosed by the emergency route have a poorer prognosis. 6,7 For some patients this may be because the disease is at a more advanced stage at diagnosis. The complications precipitating the emergency admission make an additional contribution to mortality.
Screening programmes for breast, prostate, cervical, and colorectal cancer have proved effective for diagnosing patients at an earlier stage, which can result in a better prognosis. [8][9][10][11] These screening programmes have had a significant impact in reducing the number of patients presenting as an emergency. To date there has been no accessible and economically viable diagnostic tool for early detection of asymptomatic and symptomatic brain tumours. 12 The addition of a rapid and accurate blood test for patients with suspected primary brain tumours (symptomatic patients) therefore has the potential to improve outcomes by allowing prioritisation of patients most at risk of a brain tumour for further investigation. Under the current patient pathway it is not feasible to provide fast-track diagnostic imaging because the number of patients with non-specific headache symptoms is very large and the positive predictive value (PPV) on the basis of symptoms alone is less than 3% for all symptoms other than a new-onset seizure. 13,14 Magnetic resonance imaging (MRI) and computed tomography (CT) are the current gold standard for identifying structural brain lesions including tumours. Treatment decisions made at the neuro-oncology multidisciplinary team (MDT) meeting are often based on the imaging alone. Following surgical resection or biopsy with histopathology and molecular analysis, definitive treatment can be planned. 15 Surgery to secure the tissue diagnosis has a small risk to the patient of neurological deterioration and death. 16 The diagnostic pathway also represents a significant cost burden to the health service, with a single MRI and CT scan in the UK costing around £164 and £85, respectively (National timeframe of the diagnostic pathway for brain tumours, specifically primary gliomas, is illustrated in Figure 1, and effectively highlights the significant wait that a symptomatic patient may have before receiving brain imaging. Even from this stage, regardless of the time to GP referral or emergency presentation, full diagnosis may take a further 5 weeks.

Serum Spectroscopic Diagnostics
This novel blood test for early brain tumour detection is based upon the interaction of infrared (IR) light with biological components of blood serum (Figure 2). Specifically, using attenuated total reflectance (ATR)-FTIR spectroscopy, specific bond vibrations of given molecules can be elucidated from serum samples, thus providing a unique insight into the composition via an absorbance spectrum. 17 Benefits of an ATR-FTIR based approach include a robust, user-friendly methodology without extensive sample preparation, that would readily fit into a clinical setting. 18 In short, serum is obtained according to standard protocols and can be snap frozen and stored at -80° until the point of analysis. A small volume of serum is required for analysis (1-5 µL), which is pipetted onto a crystal, known as an internal reflection element (IRE), where IR non-destructively interacts with the sample and produces an IR spectrum, with peaks representative of known bond vibrations and hence biomolecular constituents.
Blood serum is a complex medium that contains a variety of biomolecules, including around 20,000 proteins, that may be employed as diagnostic biomarkers. 19 In the case of brain tumours, such blood based technologies are limited, due to a lack of an established brain tumour specific diagnostic biomarker. 20 With our spectroscopic approach, rather than derive single biomarker specific information, a global signature is obtained, which encompasses the entire biomolecular makeup. This is epitomised as an equally complex biological absorbance spectrum, which contains a wealth of diagnostic information (an example may be seen in Figure 3).
Pattern recognition and machine learning algorithms using spectral features from FTIR data have been demonstrated as rapid and accurate for separating primary brain cancer and noncancer cases. 2,21 When these algorithms are applied to this information rich dataset, the relationship between all biological components of the sample is ascertained, providing a multi-dimensional analysis of the sample.
This approach has been able to not only detect between cancer and non-cancer, but also stratify based upon cancer pathology. 2 For further information and in-depth description of the methodology, we direct the readers to the following fundamental review and recent research papers. 2,3,17,21 The ability to triage patients likely to have a brain tumour based on serum sample alone raises the possibility of systematic triaging prior to investigation with more expensive (MRI/CT imaging) and invasive (biopsy) tests. One major impact of having a serum test available would be a clear reduction in the number of unnecessary brain scans; however, as this test is also able to differentiate between primary and secondary tumours, there could also be a knock-on reduction of chest and abdomen scans which are conducted to rule out primary disease elsewhere. There is also the possibility that this approach will reduce Ultimately, it is expected that this could allow earlier and potentially more effective treatment of brain tumours.
It is important to note that this study is based upon proof-of-concept studies, in advance of a pending clinical trial. As these samples are retrospective, there is the possibility the diagnostic performance will not be as high in prospective studies. Additional to determining the true diagnostic accuracy of the technique, the planned clinical trial held in primary care will also reveal the suitable patient population for the test, as well as the long-term benefits of an improved diagnostic pathway.

Aims & Objectives
The aim of the economic evaluation is to assess the potential cost-effectiveness of this spectroscopic technology, in advance of any prospective study results being available. There are three main objectives for the evaluation. The first objective is to create a map of where the test could be used in the clinical pathway. The second is to assess the potential costeffectiveness of the technology, if the performance shown in the case-control study is replicated prospectively. This will give an indication of whether the technology would meet the criteria for acceptance for use in the National Health Service that are applied by Health Technology Assessment agencies (HTA). Related to this, the third objective is to define the level of performance in prospective trials, and any additional evidence that would be needed to meet the cost-effectiveness criteria of an HTA decision-making process. This can include diagnostic performance and also effects on long-term outcomes such as survival and resource use. To achieve these objectives a simplified economic model of two important clinical scenarios is used to explore cost effectiveness.

Cost Effectiveness Analysis
A cost-effectiveness analysis (CEA) was conducted to calculate the effects on health outcomes and health service costs of introducing spectroscopic testing in each of two scenarios. The health outcomes considered were life-years and quality-adjusted life-years (QALYs).
A decision tree was used to model the pathway for patients presenting with symptoms warranting a referral for MRI/CT imaging for suspected brain tumour (Figure 4). Separate models were considered for primary and secondary care. The time horizon of the model is 2 years and the perspective is that of the health care service. A two-year horizon was selected because of the short duration of survival in this patient group: median survival is approximately 1 year for high grade gliomas, which are the commonest malignant primary brain tumour. In all scenarios the comparator is the current diagnostic pathway (i.e. imaging alone). Further details regarding the node probabilities, simplifications and assumptions of the model can be found in Appendix 2.

Diagnostic Performance
The sensitivity and specificity of the test for detecting brain tumours has been demonstrated in a series of case-control studies using historical samples from the Brain Tumour North West and Walton Centre NHS bio-banks. 2,3 In the key case-control study nine FTIR spectra were collected from each of 433 patients. 3 Of these 134 were from patients with primary brain tumours (64 high grade glioma), 177 were from patients with cerebral metastases and 122 were from non-cancer controls. FTIR spectra were analysed using the random forest method to fit a classification model. Classification performance was estimated by applying the fitted model to a test set containing 20% of the patients from the original data set that were not used in the model fitting step (hold-out test set). Classification statistics are computed as averages of this process, iterated 96 times using random training and test sets. Under the best available model, sensitivity estimated by this method was 92·8% and specificity was 91·5% for the analysis of cancerous vs. non-cancerous serum. 3 These classification statistics were established using 96 independent iterations of a Random Forest model, and resulted in standard deviations of 1.1% and 1.9% respectively.
Establishing whether the performance demonstrated case-control data from historical samples translates to equivalent performance when applied prospectively in clinical practice is the subject of a planned clinical trial. This is critical to demonstrating the clinical and cost-

Prevalence of disease
Prevalence data were sourced from the literature based on clinical expert guidance. Brain tumour prevalence in scenario 2 (secondary care) was assumed to be 3% based on observed rates of primary brain tumour diagnosis among patients referred for brain imaging for suspected cancer in secondary care. [22][23][24] In scenario 1 (primary care) an estimated prevalence of 0·5% is used based on case-control evidence and expert opinion of the prevalence among patients who would be considered for direct access imaging, using MRI or CT, where this is available and referral to neurology where it is not. 25,26 An alternative prevalence of 1% was explored for scenario 1 based on unpublished data from a direct access imaging service in the UK (P. Brennan, personal communication).
The effect of serum spectroscopy testing on the time-to-diagnosis and time-to-treatment is discussed in Appendix 3, additional to the effect of testing on the use of imaging studies, and also on the patient outcomes. The primary assumptions in this model are that first, the expected time-to-diagnosis would match the current median time-to-diagnosis for patients presenting with brain tumour in emergency care as observed in Aggarwal et al. 5 Furthermore, based upon expert opinion, it is assumed that in secondary care all patients would continue to imaging, while in primary care 50% would continue to imaging following a negative spectroscopy result. This is a conservative estimate associated with the possibility that an imaging test will still be required in some cases based on interpretation of a patient's symptoms and the other non-tumour diagnoses being considered by the clinician. Finally, the effects of early diagnosis on the outcome of brain tumours is estimated using literature describing fitting natural history models to observational datasets of high grade glioma patients. 27,28 Utility Weights Health state utility weights are applied to life-years to generate quality adjusted life years (QALYs). A systematic review of health state utility weights for high grade glioma, the most common and aggressive primary brain tumour, was conducted to identify suitable utility weights. Due to heterogeneity it was not considered suitable to pool the estimates. The most appropriate health state utility weight was taken from a previous UK economic evaluation of glioblastoma treatment. 29 A value of 0·89 was used in the base case.

Resource Use and Costs
Resource use includes the application of a spectroscopic serum test to all patients prior to imaging, the imaging studies used in the diagnostic process, outpatient neurology clinic visits and general practitioner visits. In the UK analysis, unit costs for imaging studies are taken from UK NHS reference costs (2014/15), clinic and GP visits from the PSSRU costs schedule (Table 1). In the USA analysis, unit costs are taken from Medicare reimbursement schedules. Unit costs of the test were applied at an upper bound and lower bound rather than a single value as these products have not yet been commercialized. Bounds were set by consultation

Incremental cost effectiveness ratios (ICER)
The comparative cost-effectiveness of spectroscopic testing compared to no testing is summarized by the ICER defined as: and ‫ܥ‬ are the total costs with spectroscopic testing and no testing respectively. Equivalently ‫ܪ‬ ௦ and ‫ܪ‬ are the total QALYs with and without spectroscopic testing. The ICER can be interpreted as the additional cost per QALY gained. Sensitivity analyses included one-way sensitivity analysis (OWSA), systematically varying a single parameter in the model, and scenario analysis in which specific model assumptions were altered. OWSA were conducted for test sensitivity, specificity and test cost. Scenario analyses included assuming an additional consultation cost for discussion of test results, assuming a higher proportion of patients continue to imaging following a negative spectroscopy result in primary care, and using mean survival rather than median survival.
A probabilistic sensitivity analysis (PSA) was conducted to explore the effects of joint uncertainty in the parameter estimates on the model results. 30

Mapping the clinical pathway
For the first study objective, initial discussion with clinical experts indicated that there are potential uses for the test in both primary and secondary care. The main advantage of employing a cost-effective spectroscopic blood test in the diagnostic pathway is to use it as a triage test. This prioritises more urgent cases for access to services, and acts as a gate-keeper (requiring a positive result in some conditions to give access to services) for imaging studies.
Two clinical scenarios are mapped out below and subsequently explored in this early economic evaluation of the serum spectroscopy test for aiding diagnosis of brain tumours:

Triage tool in primary care
The primary care scenario explores a population of patients with a clinical presentation that warrants further investigation of possible brain tumour. This would include some patients with headaches and some with focal neurological deficits. This is the group of patients who would be considered for direct access imaging, using MRI or CT, where this is available and referral to neurology where it is not. 31 The blood test is used to provide rapid information, within 24 hours, where a positive result would lead to patients receiving more timely access to imaging. It may also be the case that negative test results, in addition to establishing the low probability of a brain tumour, could also provide some reassurance for those patients that must wait for imaging. The total volume of tests would be approximately 75,000 per year in the UK (see Appendix 4 for further details).

Triage tool in secondary care
In this scenario the population is the group of patients that are currently referred for imaging studies from secondary care for suspected brain tumour, typically via neurology clinics. This is the patient group for whom the clinical presentation has the highest positive predictive value (PPV). However, even in this high risk group, the odds of a brain tumour being present are approximately 1:33. [22][23][24] Again, the spectroscopy test is used to provide rapid information to allow a subset of these patients to access immediate imaging and provide reassurance to other patients who may have to wait longer for definitive imaging studies and diagnosis. The extent of the benefits of triage in this scenario is likely to vary by locality depending on the capacity constraints on imaging and pathology services. This evaluation uses estimates of the delays in diagnosis, and potential improvements in the speed of diagnosis, from a consecutive patient case series in London, UK. 5 The total volume of tests if this scenario occurred would be approximately 53,000 per year in the UK (see Appendix 4 for further details).

Cost Effectiveness Assessment
The standard threshold value per QALY gained in the UK, is considered to be between £20,000 to £30,000. Below this value, a healthcare intervention may be considered cost effective, whereas a negative ICER value would be deemed cost saving. Base case results for primary care (scenario 1) and secondary care (scenario 2) are presented in Table 2. Note results are reported for cohorts of 10,000 patients.

Sensitivity Analysis Results
The performance of the test with regards to levels of sensitivity and specificity are addressed using sensitivity analysis. OWSA results for a range of test specificities are displayed in Figure 5 and 6, for primary and secondary care respectively, displaying the ICER with varying test specificity. Note that the estimated QALYs do not change with specificity in the model therefore changes in the ICER are due solely to changes in incremental costs. Varying sensitivity changes both estimated QALYs and estimated costs therefore results of the OWSA for test sensitivity are presented on the cost-effectiveness plane (Appendix 5). In primary care, using the upper cost limit of £100 it is evident that the test is deemed cost effective at specificities of around 0·9 and above, where the ICER is below standard thresholds. In contrast, at the lower cost limit, the test is cost effective at specificities above 0·8. Although the serum spectroscopy test is not cost saving at low cost (or near perfect specificities), the test is still considered cost effective at specificity levels around 0·7 and 0·8 for £50 and £100 pricing respectively. Additional scenario analyses are also reported in Appendix 5. These demonstrate that results are robust to using mean survival estimates rather than median survival estimates and including additional consultation costs for positive test results. If the prevalence of brain tumours in Scenario 1 is 1% rather than 0·5% the incremental QALYs increase substantially and the ICERs are reduced.
The PSA results reported in Figure 7 and 8 indicate that at a test cost of £50 and an ICER threshold of £30,000 per QALY there is a near 100% probability the serum spectroscopy test is cost-effective in scenario 1 and approximately 90% probability it is cost-effective in scenario 2. The corresponding probabilities at the upper bound cost of £100 are approximately 85% and 75%.

Discussion
This economic evaluation establishes the potential for serum spectroscopy to have a role in the diagnosis of both benign and malignant brain tumours in both primary and secondary care. The potential costs and health benefits of testing using a spectroscopic method prior to CT/MRI tests (or in some scenarios to avoid imaging) have been estimated based on a mathematical model with parameter values taken from published studies and expert opinion. This diagnostic tool is sensitive to all brain tumours (benign or malignant), however, this assessment is closely aligned with the diagnosis of primary gliomas, where there is a maximum potential benefit to the health service.
The major limitations of this analysis relate to the use of proof-of-concept studies and a disease natural history model rather than direct clinical trial evidence. This creates additional uncertainties. Results should be interpreted as indicative and used primarily to guide future evidence generation. Furthermore, the scenarios explored were limited in scope; future studies should continue to refine understanding of the role of the test in real-world clinical decision making.
When used as a triage tool in primary care, this novel test has the potential to deliver improvements in health outcomes and also to reduce costs. At the lower end of test costs, the technology would be cost-saving for the health service. At higher test costs the technology is still likely to be considered cost-effective in HTA agency decision processes.
In Scenario 2, in which serum spectroscopy is used as a triage tool in secondary care, the technology will create additional costs but also produce sizable health benefits. At test costs of under £100 ($200) the technology would be likely to be considered as a cost-effective use of resources in HTA agency decision processes in the UK (and USA). It is assumed that in both scenarios, the uptake of the test in the USA would mirror that of the UK; however, this would need to be explored further, alongside clinical experts of the USA care pathways.
Sensitivity analyses have demonstrated the importance of diagnostic performance on the costeffectiveness of the test. In particular, test specificity is important in the primary care setting. If test specificity is 87·5% or worse the technology may not be considered cost-effective at higher values of assumed test cost. This is due to the increased number of false positive results in this low prevalence population, generating a greatly increased proportion of fasttrack imaging studies which increases costs.
To strengthen the case that this approach represents a cost-effective use of healthcare resources it is necessary to establish the diagnostic performance of the test prospectively. This can be accomplished by a suitably large cohort study in which serum spectroscopy is used alongside current clinical practice in one of the patient groups included in this evaluation. It would be appropriate to initially target the secondary care patient population, because the higher prevalence of disease in this group will reduce the sample size needed to accurately estimate diagnostic performance. Decision makers are often most interested in patient outcomes, such as survival, rather than intermediate outcomes, such as accuracy or speed of diagnosis (although this latter point is vital for treatment of high grade gliomas). From this perspective, a randomized trial, or a prospective cohort study with extended follow-up, may be required to fully establish the size of survival and quality-of-life benefits of including a serum spectroscopy test in the diagnostic pathway. A trial with primary outcomes relating to survival and quality-of-life would be specific to either the primary or secondary care setting (rather than generalisable to both), would need a large sample size, and would also require a follow-up period to capture survival benefits. In the case of malignant glioma this would require a period of at least 24 months. Such a trial would clearly be expensive, time consuming and may be unfeasible. Decision makers may be willing to make a decision on implementation of the blood test based on the modeled effects of improvements in intermediate outcomes on later patient outcomes. In this situation, the model proposed in this evaluation, populated with diagnostic performance and other data from a prospective trial, could be used to inform decisions about the wider adoption of the technology.
Future developments beyond trials such as emerging epidemiological evidence and new technologies should also be included in any future evaluations. It was not possible to foresee and include all such possible scenarios in this early evaluation but that should not preclude assessment in the light of new evidence. Updated analysis should inform any decisions about system wide implementation.
Several results in this analysis suggest cost savings through reduced use of imaging for patients with a negative test result. To make the case that a serum spectroscopic test can improve the efficiency of the diagnostic pathway prospective studies will also need to explore the impact of these test results on clinician and patient imaging study decisions. The possibility remains that the test may triage patients, but may not reduce the number of scans being conducted, and could potentially increase the demand on imaging. However, this triaging of patients would still benefit each patient that is provided with an early diagnosis.

Appendix 2 -Decision tree model
The decision node probabilities that populate the decision tree model are reported in Table S1. The model is simplified by assuming that the reference test (MRI/CT) has perfect accuracy (100% sensitivity and 100% specificity). This is desirable if we wish to evaluate the test against an MRI/CT only diagnostic pathway in which these imaging studies are considered as the 'gold standard'. It is also difficult to obtain valid estimates of MRI/CT sensitivity and specificity for this indication as these are not reported in a comparable manner in the literature. This simplification will not substantially alter the economic results compared to using more realistic values for the scenarios under consideration. This is because it is assumed confirmation of the diagnosis always requires imaging and therefore outcomes for patients that would hypothetically be classified correctly using this test but incorrectly by MRI/CT will remain the same whether or not the test is used. The effect of less than perfect accuracy of the reference test is therefore the same as a (small) decrease in prevalence as a proportion of patients with the disease cannot benefit from the additional test due to incorrect subsequent testing. Outcomes are calculated by 'rolling-back' the tree for both cancer cases and non-cases and taking an average of the two groups weighted by the prevalence of brain tumours in the scenario population. Referral decision S1:0·5, S2: 1 S1:0·5, S2: 0 S1 & S2: Scenarios 1 &2  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

Effect of testing on time-to-diagnosis and time-to-treatment
To estimate the expected time-to-diagnosis under a fast-track pathway it is assumed that this would match the current median time-to-diagnosis for patients presenting with brain tumour in emergency care as observed in Aggarwal et al 5 . This study reported a cohort of high grade glioma patients from a single hospital in London, UK. No other suitable estimates of time-to-treatment or time-to-diagnosis were identified in the literature. This is the most common primary brain tumour in the indications considered in this evaluation. Median time-todiagnosis for the standard pathway is also taken from the same source.

Effect of testing on use of imaging studies
The effect of spectroscopic testing on imaging study decisions of clinicians and patients is uncertain. Based on clinical expert input it was assumed, in the base case, that in secondary care all patients would continue to imaging, while in primary care 50% would continue to imaging following a negative spectroscopy result.

Effect of early diagnosis on patient outcomes
A systematic review of the literature failed to identify any studies that directly estimated the effect of earlier (or later) diagnosis on outcomes for primary brain tumours. Instead, based clinical expert guidance, estimates of the effects of time-to-treatment were sourced from the literature. The estimated effects are based on fitting natural history models to observational datasets of high grade glioma patients. 27,28 It is necessary to use a natural history model to produce survival estimates rather than using the survival data stratified by time-to-treatment directly, in order to adjust for potential confounding factors influencing both time-to-treatment and survival.
A one-to-one correspondence is assumed for the effects of an additional day added to the time-to-diagnosis and an additional day added to the time-to-treatment. Median survival by weeks between diagnosis and treatment are displayed in Table 2. The hazard ratio per additional day delay calculated in Do et al of 1·015 was applied to each day between diagnosis and treatment up to 28 days to calculate median survival time. Time between diagnosis and treatment beyond 28 days were fixed at the same median survival as 28 days.

Resource use and cost
It is assumed that there is no additional cost for converting a small proportion of standard referral to additional urgent referrals. This could be justified on the basis that patients receiving negative results allows more flexibility in the scheduling for these patients. This could create more opportunity to schedule patients with positive spectroscopy results sooner. As a larger proportion of patients are referred for immediate imaging it was assumed costs would increase because achieving this would require additional capacity. It was assumed that unit costs for MRI and CT grow exponentially with increasing proportion of immediate referrals such that costs double if 50% of patients are referred for immediate imaging.
Treatment costs are assumed to be the same for both fast-track and standard referrals that are diagnosed as cases. This may be a conservative assumption as treatment costs are likely to increase as disease progresses and therefore earlier treatment may, on average, be less costly than later treatment. However, it is also possible that The balance of these effects is unknown given the lack of appropriate data to estimate the effects. Secondary treatment and end-of-life costs are similarly assumed to be equal between fast track and standard referrals. This reflects the model assumption that all patients ultimately progress.

Primary care
An assumption that patients currently referred in direct-access imaging service would be selected for spectroscopic testing can provide a conservative estimate of the total number of patients likely to be tested. Three sources from the literature were considered: 1. A direct-access brain MRI service in Nottingham (UK) reported 130 patients per quarter (520 per year) among a population of 342,000. 32,33 The population of the UK is approximately 61 million therefore, assuming Nottingham is representative of the UK general population demand for brain MRI, there would be approximately 93,000 per year in the UK. 2. A direct access CT service for chronic headache in Glasgow (UK) reported 4404 patients referred over 8 years (551 per year) among a population of approximately 1 million. 31 Assuming this was representative of the UK generally then the demand for this type of service would be approximately 34,000 for the UK. 3. A direct access CT neuroimaging service in Lothian (UK) reported 389 exams per year for a population of approximately 442,000. 34 Assuming this was representative of the UK generally then the demand for this type of service would be approximately 54,000 for the UK.
The Glasgow estimate is particularly low and may result from limited uptake in the early part of the 8 year period for which data were reported. As it was not possible to determine if uptake increased over the period or was stable this estimate was not considered reliable. The Nottingham and Lothian estimates report potentially more stable and established services and are therefore more likely to reliably estimate demand. The likely demand for this novel test, if the indication is considered identical to neuroimaging, is therefore between 54,000 and 93,000. This is likely to be conservative as the test may well have a slightly broader indication than current neuroimaging. A reasonable point estimate may be approximately 75,000.

Secondary Care
An estimate of the total demand for serum spectroscopic tests in the UK can be made based on the reported incidence of brain tumour in the UK and the proportion of cases currently diagnosed through secondary (nonemergency) care. Annual incidence of malignant brain tumours in the UK is approximately 4,700. 35 Approximately one third of these are likely to present through secondary, non-emergency care. Using the positive predictive value for patients currently referred to neuroimaging in this setting of 3% suggests that for each case there will be 33 non-case patients receiving neuroimaging. In total, in the UK, this would imply 53,000 patients per year in secondary care may be suitable for spectroscopic testing.

Target population and subgroups 4
Describe characteristics of the base case population and subgroups analysed, including why they were chosen. Setting and location 5 State relevant aspects of the system(s) in which the decision(s) need(s) to be made. Study perspective 6 Describe the perspective of the study and relate this to the costs being evaluated. Comparators 7 Describe the interventions or strategies being compared and state why they were chosen. Time horizon 8 State the time horizon(s) over which costs and consequences are being evaluated and say why appropriate. Discount rate 9 Report the choice of discount rate(s) used for costs and outcomes and say why appropriate. Choice of health outcomes 10 Describe what outcomes were used as the measure(s) of benefit in the evaluation and their relevance for the type of analysis performed. Measurement of effectiveness 11a Single study-based estimates: Describe fully the design features of the single effectiveness study and why the single study was a sufficient source of clinical effectiveness data. 11b Synthesis-based estimates: Describe fully the methods used for identification of included studies and synthesis of clinical effectiveness data. Measurement and valuation of preference based outcomes 12 If applicable, describe the population and methods used to elicit preferences for outcomes.

Estimating resources and costs
13a Single study-based economic evaluation: Describe approaches used to estimate resource use associated with the alternative interventions. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs. 13b Model-based economic evaluation: Describe approaches and data sources used to estimate resource use associated with model health states. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs. Currency, price date, and conversion 14 Report the dates of the estimated resource quantities and unit costs. Describe methods for adjusting estimated unit costs to the year of reported costs if necessary. Describe methods for converting costs into a common currency base and the exchange rate. Describe all analytical methods supporting the evaluation. This could include methods for dealing with skewed, missing, or censored data; extrapolation methods; methods for pooling data; approaches to validate or make adjustments (such as half cycle corrections) to a model; and methods for handling population heterogeneity and uncertainty.

Study parameters 18
Report the values, ranges, references, and, if used, probability distributions for all parameters. Report reasons or sources for distributions used to represent uncertainty where appropriate. Providing a table to show the input values is strongly recommended. Incremental costs and outcomes 19 For each intervention, report mean values for the main categories of estimated costs and outcomes of interest, as well as mean differences between the comparator groups. If applicable, report incremental cost-effectiveness ratios. Characterising uncertainty 20a Single study-based economic evaluation: Describe the effects of sampling uncertainty for the estimated incremental cost and incremental effectiveness parameters, together with the impact If applicable, report differences in costs, outcomes, or costeffectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information. To determine the potential costs and health benefits of a serum-based spectroscopic triage tool for brain tumours, that could be developed to reduce diagnostic delays in the current clinical pathway.

Design
A model-based health pre-trial economic assessment. Decision tree models were constructed based on simplified diagnostic pathways. Models were populated with parameters identified from rapid reviews of the literature and clinical expert opinion.

Setting
Explored as a test in both primary and secondary care (neuroimaging) in the UK health service, as well as application to the US.

Participants
Calculations based upon an initial cohort of 10,000 patients. In primary care, it is estimated that the volume of tests would approach 75,000 per annum. The volume of tests in secondary care is estimated at 53,000 per annum.

Main Outcome Measures
The primary outcome measure was quality adjusted life years (QALYs), which were employed to derive incremental cost-effectiveness ratios (ICERs) in a cost-effectiveness analysis.

Results
Results indicate that using a blood-based spectroscopic test in both scenarios has the potential to be highly cost-effective in a health technology assessment (HTA) agency decision making process, as ICERs were well below standard threshold values of £20,000 to £30,000 per QALY. This test may be cost-effective in both scenarios with test sensitivities and specificities as low as 80%; however, the price of the test would need to be lower (less than approximately £40).

Conclusion
Use of this test as triage tool in primary care has the potential to be both more effective and cost saving for the health service. In secondary care, this test would also be deemed more effective than the current diagnostic pathway.

Strengths and Limitations of this Study
• Simplified models of clinical pathways were mapped with input from, and consensus among, a wide range of clinical experts including neurosurgeons, neurooncologists, neuropathologists, neuroradiologists, and primary care experts.
• The spectroscopic blood test was highly sensitive and specific in retrospective data, with performances of 91.5 and 83.0% respectively. There is potential for this to contribute towards improved prognosis for patients, as well as healthcare savings.
• This study is based upon proof-of-concept studies, in advance of a pending prospective clinical trial. As these samples are retrospective, there is the possibility the diagnostic performance will not be as high in prospective studies.
• A lack of clinical trial evidence necessitates the estimate of long-term benefits of improved diagnostic protocols based on disease natural history models. This creates additional uncertainties.
• The precise patient population for whom the test may be suitable in the primary care setting is difficult to establish at this stage in development. This study considers a limited definition of eligibility that may need revision in light of future evidence.  1 This, at least in part, may relate to diagnostic delays, reflecting the non-specific early symptoms, such as headache and dizziness, from which general practitioners (GPs) must identify patients at risk for further investigation. The lack of a low-cost diagnostic and/or screening tools available within the health service contributes to this delay. We have recently demonstrated that a spectroscopic test using blood-serum is able to effectively identify brain tumours in patients with sensitivities and specificities as high as 91·5% and 83% respectively, in a tissue bank casecontrol series. 2,3 This approach is based upon Fourier-transform infrared (FTIR) spectroscopy and can detect disease-specific signatures, which are extracted mathematically using pattern recognition and machine learning algorithms.

Current Diagnostic Pathway
Currently, patients who are symptomatic with a brain tumour visit their GP on average five times before being referred to secondary care. 4 Partly as a result of this diagnostic delay up to 61% of brain tumour diagnoses occur in an emergency setting, often following a seizure. 5 Patients diagnosed by the emergency route have a poorer prognosis. 6,7 For some patients this may be because the disease is at a more advanced stage at diagnosis. The complications precipitating the emergency admission make an additional contribution to mortality.
Screening programmes for breast, prostate, cervical, and colorectal cancer have proved effective for diagnosing patients at an earlier stage, which can result in a better prognosis. [8][9][10][11] These screening programmes have had a significant impact in reducing the number of patients presenting as an emergency. To date there has been no accessible and economically viable diagnostic tool for early detection of asymptomatic and symptomatic brain tumours. 12 The addition of a rapid and accurate blood test for patients with suspected primary brain tumours (symptomatic patients) therefore has the potential to improve outcomes by allowing prioritisation of patients most at risk of a brain tumour for further investigation. Under the current patient pathway it is not feasible to provide fast-track diagnostic imaging because the number of patients with non-specific headache symptoms is very large and the positive predictive value (PPV) on the basis of symptoms alone is less than 3% for all symptoms other than a new-onset seizure. 13,14 Magnetic resonance imaging (MRI) and computed tomography (CT) are the current gold standard for identifying structural brain lesions including tumours. Treatment decisions made at the neuro-oncology multidisciplinary team (MDT) meeting are often based on the imaging alone. Following surgical resection or biopsy with histopathology and molecular analysis, definitive treatment can be planned. 15 Surgery to secure the tissue diagnosis has a small risk to the patient of neurological deterioration and death. 16 The diagnostic pathway also represents a significant cost burden to the health service, with a single MRI and CT scan in the UK costing around £164 and £85, respectively (National timeframe of the diagnostic pathway for brain tumours, specifically primary gliomas, is illustrated in Figure 1, and effectively highlights the significant wait that a symptomatic patient may have before receiving brain imaging. Even from this stage, regardless of the time to GP referral or emergency presentation, full diagnosis may take a further 5 weeks.

Serum Spectroscopic Diagnostics
This novel blood test for early brain tumour detection is based upon the interaction of infrared (IR) light with biological components of blood serum (Figure 2). Specifically, using attenuated total reflectance (ATR)-FTIR spectroscopy, specific bond vibrations of given molecules can be elucidated from serum samples, thus providing a unique insight into the composition via an absorbance spectrum. 17 Benefits of an ATR-FTIR based approach include a robust, user-friendly methodology without extensive sample preparation, that would readily fit into a clinical setting. 18 In short, serum is obtained according to standard protocols and can be snap frozen and stored at -80° until the point of analysis. A small volume of serum is required for analysis (1-5 µL), which is pipetted onto a crystal, known as an internal reflection element (IRE), where IR non-destructively interacts with the sample and produces an IR spectrum, with peaks representative of known bond vibrations and hence biomolecular constituents.
Blood serum is a complex medium that contains a variety of biomolecules, including around 20,000 proteins, that may be employed as diagnostic biomarkers. 19 In the case of brain tumours, such blood based technologies are limited, due to a lack of an established brain tumour specific diagnostic biomarker. 20 With our spectroscopic approach, rather than derive single biomarker specific information, a global signature is obtained, which encompasses the entire biomolecular makeup. This is epitomised as an equally complex biological absorbance spectrum, which contains a wealth of diagnostic information (an example may be seen in Figure 3).
Pattern recognition and machine learning algorithms using spectral features from FTIR data have been demonstrated as rapid and accurate for separating primary brain cancer and noncancer cases. 2,21 When these algorithms are applied to this information rich dataset, the relationship between all biological components of the sample is ascertained, providing a multi-dimensional analysis of the sample.
In the case-control setting, this approach has been able to not only detect between cancer and non-cancer, but also stratify based upon cancer pathology. 2 For further information and indepth description of the methodology, we direct the readers to the following fundamental review and recent research papers. 2,3,17,21 The ability to triage patients likely to have a brain tumour based on serum sample alone raises the possibility of systematic triaging prior to investigation with more expensive (MRI/CT imaging) and invasive (biopsy) tests. One major impact of having a serum test available would be a possible reduction in the number of unnecessary brain scans; however, as this test is also able to differentiate between primary and secondary tumours, there could also be a knock-on reduction of chest and abdomen scans which are conducted to rule out primary disease elsewhere. There is also the possibility that this approach will reduce the incidence of incidental abnormalities which in themselves can cause considerable distress. Ultimately, it is expected that this could allow earlier and potentially more effective treatment of brain tumours.
It is important to note that this study is based upon proof-of-concept studies, in advance of a pending clinical trial. As these samples are retrospective, there is the possibility the diagnostic performance will not be as high in prospective studies. Additional to determining the true diagnostic accuracy of the technique, the planned clinical trial held in primary care will also reveal the suitable patient population for the test, as well as the long-term benefits of an improved diagnostic pathway.

Aims & Objectives
The aim of the economic evaluation is to assess the potential cost-effectiveness of this spectroscopic technology, in advance of any prospective study results being available. There are three main objectives for the evaluation. The first objective is to create a map of where the test could be used in the clinical pathway. The second is to assess the potential costeffectiveness of the technology, if the performance shown in the case-control study is replicated prospectively. This will give an indication of whether the technology would meet the criteria for acceptance for use in the National Health Service that are applied by Health Technology Assessment agencies (HTA). Related to this, the third objective is to define the level of performance in prospective trials, and any additional evidence that would be needed to meet the cost-effectiveness criteria of an HTA decision-making process. This can include diagnostic performance and also effects on long-term outcomes such as survival and resource use. To achieve these objectives a simplified economic model of two important clinical scenarios is used to explore cost effectiveness.

Cost Effectiveness Analysis
A cost-effectiveness analysis (CEA) was conducted to calculate the effects on health outcomes and health service costs of introducing spectroscopic testing in each of two scenarios. The health outcomes considered were life-years and quality-adjusted life-years (QALYs).
A decision tree was used to model the pathway for patients presenting with symptoms warranting a referral for MRI/CT imaging for suspected brain tumour (Figure 4). Separate models were considered for primary and secondary care. The time horizon of the model is 2 years and the perspective is that of the health care service. A two-year horizon was selected because of the short duration of survival in this patient group: median survival is approximately 1 year for high grade gliomas, which are the commonest malignant primary brain tumour. In all scenarios the comparator is the current diagnostic pathway (i.e. imaging alone). Further details regarding the node probabilities, simplifications and assumptions of the model can be found in Appendix 2.

Diagnostic Performance
The sensitivity and specificity of the test for detecting brain tumours has been demonstrated in a series of case-control studies using historical samples from the Brain Tumour North West and Walton Centre NHS bio-banks. 2,3 In the key case-control study nine FTIR spectra were collected from each of 433 patients. 3 Of these 134 were from patients with primary brain tumours (64 high grade glioma), 177 were from patients with cerebral metastases and 122 were from non-cancer controls. FTIR spectra were analysed using the random forest method to fit a classification model. Classification performance was estimated by applying the fitted model to a test set containing 20% of the patients from the original data set that were not used in the model fitting step (hold-out test set). Classification statistics are computed as averages of this process, iterated 96 times using random training and test sets. Under the best available model, sensitivity estimated by this method was 92·8% and specificity was 91·5% for the analysis of cancerous vs. non-cancerous serum. 3 These classification statistics were established using 96 independent iterations of a Random Forest model, and resulted in standard deviations of 1.1% and 1.9% respectively.
Establishing whether the performance demonstrated case-control data from historical samples translates to equivalent performance when applied prospectively in clinical practice is the subject of a planned clinical trial. This is critical to demonstrating the clinical and cost-

Prevalence of disease
Prevalence data were sourced from the literature based on clinical expert guidance. Brain tumour prevalence in scenario 2 (secondary care) was assumed to be 3% based on observed rates of primary brain tumour diagnosis among patients referred for brain imaging for suspected cancer in secondary care. [22][23][24] In scenario 1 (primary care) an estimated prevalence of 0·5% is used based on case-control evidence and expert opinion of the prevalence among patients who would be considered for direct access imaging, using MRI or CT, where this is available and referral to neurology where it is not. 25,26 An alternative prevalence of 1% was explored for scenario 1 based on unpublished data from a direct access imaging service in the UK (P. Brennan, personal communication).
The effect of serum spectroscopy testing on the time-to-diagnosis and time-to-treatment is discussed in Appendix 3, additional to the effect of testing on the use of imaging studies, and also on the patient outcomes. The primary assumptions in this model are that first, the expected time-to-diagnosis would match the current median time-to-diagnosis for patients presenting with brain tumour in emergency care as observed in Aggarwal et al. 5 Furthermore, based upon expert opinion, it is assumed that in secondary care all patients would continue to imaging, while in primary care 50% would continue to imaging following a negative spectroscopy result. This is a conservative estimate associated with the possibility that an imaging test will still be required in some cases based on interpretation of a patient's symptoms and the other non-tumour diagnoses being considered by the clinician. Finally, the effects of early diagnosis on the outcome of brain tumours is estimated using literature describing fitting natural history models to observational datasets of high grade glioma patients. 27,28 Utility Weights Health state utility weights are applied to life-years to generate quality adjusted life years (QALYs). A systematic review of health state utility weights for high grade glioma, the most common and aggressive primary brain tumour, was conducted to identify suitable utility weights. Due to heterogeneity it was not considered suitable to pool the estimates. The most appropriate health state utility weight was taken from a previous UK economic evaluation of glioblastoma treatment. 29 A value of 0·89 was used in the base case.

Resource Use and Costs
Resource use includes the application of a spectroscopic serum test to all patients prior to imaging, the imaging studies used in the diagnostic process, outpatient neurology clinic visits and general practitioner visits. In the UK analysis, unit costs for imaging studies are taken from UK NHS reference costs (2014/15), clinic and GP visits from the PSSRU costs schedule (Table 1). In the USA analysis, unit costs are taken from Medicare reimbursement schedules.

Incremental cost effectiveness ratios (ICER)
The comparative cost-effectiveness of spectroscopic testing compared to no testing is summarized by the ICER defined as: and ‫ܥ‬ are the total costs with spectroscopic testing and no testing respectively. Equivalently ‫ܪ‬ ௦ and ‫ܪ‬ are the total QALYs with and without spectroscopic testing. The ICER can be interpreted as the additional cost per QALY gained.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Sensitivity analyses included one-way sensitivity analysis (OWSA), systematically varying a single parameter in the model, and scenario analysis in which specific model assumptions were altered. OWSA were conducted for test sensitivity, specificity and test cost. Scenario analyses included assuming an additional consultation cost for discussion of test results, assuming a higher proportion of patients continue to imaging following a negative spectroscopy result in primary care, and using mean survival rather than median survival.
A probabilistic sensitivity analysis (PSA) was conducted to explore the effects of joint uncertainty in the parameter estimates on the model results. 30

Patient and Public Involvement
Patients and the public were not actively involved in the formation of this study. The impact of the test on clinical decision making was the priority in this instance; however, the involvement of patients going forward will be fundamental to understanding the tests uptake into the health services.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o n l y

Mapping the clinical pathway
For the first study objective, initial discussion with clinical experts indicated that there are potential uses for the test in both primary and secondary care. The main advantage of employing a cost-effective spectroscopic blood test in the diagnostic pathway is to use it as a triage test. This prioritises more urgent cases for access to services, and acts as a gate-keeper (requiring a positive result in some conditions to give access to services) for imaging studies.
Two clinical scenarios are mapped out below and subsequently explored in this early economic evaluation of the serum spectroscopy test for aiding diagnosis of brain tumours:

Triage tool in primary care
The primary care scenario explores a population of patients with a clinical presentation that warrants further investigation of possible brain tumour. This would include some patients with headaches and some with focal neurological deficits. This is the group of patients who would be considered for direct access imaging, using MRI or CT, where this is available and referral to neurology where it is not. 31 The blood test is used to provide rapid information, within 24 hours, where a positive result would lead to patients receiving more timely access to imaging. It may also be the case that negative test results, in addition to establishing the low probability of a brain tumour, could also provide some reassurance for those patients that must wait for imaging. The total volume of tests would be approximately 75,000 per year in the UK (see Appendix 4 for further details).

Triage tool in secondary care
In this scenario the population is the group of patients that are currently referred for imaging studies from secondary care for suspected brain tumour, typically via neurology clinics. This is the patient group for whom the clinical presentation has the highest positive predictive value (PPV). However, even in this high risk group, the odds of a brain tumour being present are approximately 1:33. [22][23][24] Again, the spectroscopy test is used to provide rapid information to allow a subset of these patients to access immediate imaging and provide reassurance to other patients who may have to wait longer for definitive imaging studies and diagnosis. The extent of the benefits of triage in this scenario is likely to vary by locality depending on the capacity constraints on imaging and pathology services. This evaluation uses estimates of the delays in diagnosis, and potential improvements in the speed of diagnosis, from a consecutive patient case series in London, UK. 5 The total volume of tests if this scenario occurred would be approximately 53,000 per year in the UK (see Appendix 4 for further details).  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n l y

Cost Effectiveness Assessment
The standard threshold value per QALY gained in the UK, is considered to be between £20,000 to £30,000. Below this value, a healthcare intervention may be considered cost effective, whereas a negative ICER value would be deemed cost saving. Base case results for primary care (scenario 1) and secondary care (scenario 2) are presented in Table 2. Note results are reported for cohorts of 10,000 patients.

Sensitivity Analysis Results
The performance of the test with regards to levels of sensitivity and specificity are addressed using sensitivity analysis. OWSA results for a range of test specificities are displayed in Figure 5 and 6, for primary and secondary care respectively, displaying the ICER with  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y varying test specificity. Note that the estimated QALYs do not change with specificity in the model therefore changes in the ICER are due solely to changes in incremental costs. Varying sensitivity changes both estimated QALYs and estimated costs therefore results of the OWSA for test sensitivity are presented on the cost-effectiveness plane (Appendix 5). In primary care, using the upper cost limit of £100 it is evident that the test is deemed cost effective at specificities of around 0·9 and above, where the ICER is below standard thresholds. In contrast, at the lower cost limit, the test is cost effective at specificities above 0·8. Although the serum spectroscopy test is not cost saving at low cost (or near perfect specificities), the test is still considered cost effective at specificity levels around 0·7 and 0·8 for £50 and £100 pricing respectively. Additional scenario analyses are also reported in Appendix 5. These demonstrate that results are robust to using mean survival estimates rather than median survival estimates and including additional consultation costs for positive test results. If the prevalence of brain tumours in Scenario 1 is 1% rather than 0·5% the incremental QALYs increase substantially and the ICERs are reduced.

Discussion
This economic evaluation establishes the potential for serum spectroscopy to have a role in the diagnosis of both benign and malignant brain tumours in both primary and secondary care. The potential costs and health benefits of testing using a spectroscopic method prior to CT/MRI tests (or in some scenarios to avoid imaging) have been estimated based on a mathematical model with parameter values taken from published studies and expert opinion. This diagnostic tool is sensitive to all brain tumours (benign or malignant), however, this assessment is closely aligned with the diagnosis of primary gliomas, where there is a maximum potential benefit to the health service.
The major limitations of this analysis relate to the use of proof-of-concept studies and a disease natural history model rather than direct clinical trial evidence. This creates additional uncertainties. Results should be interpreted as indicative and used primarily to guide future evidence generation. Furthermore, the scenarios explored were limited in scope; future studies should continue to refine understanding of the role of the test in real-world clinical decision making.
When used as a triage tool in primary care, this novel test has the potential to deliver improvements in health outcomes and also to reduce costs. At the lower end of test costs, the technology would be cost-saving for the health service. At higher test costs the technology is still likely to be considered cost-effective in HTA agency decision processes.
In Scenario 2, in which serum spectroscopy is used as a triage tool in secondary care, the technology will create additional costs but also produce sizable health benefits. At test costs of under £100 ($200) the technology would be likely to be considered as a cost-effective use of resources in HTA agency decision processes in the UK (and USA). It is assumed that in both scenarios, the uptake of the test in the USA would mirror that of the UK; however, this would need to be explored further, alongside clinical experts of the USA care pathways.
Sensitivity analyses have demonstrated the importance of diagnostic performance on the costeffectiveness of the test. In particular, test specificity is important in the primary care setting. If test specificity is 87·5% or worse the technology may not be considered cost-effective at higher values of assumed test cost. This is due to the increased number of false positive results in this low prevalence population, generating a greatly increased proportion of fasttrack imaging studies which increases costs.
To strengthen the case that this approach represents a cost-effective use of healthcare resources it is necessary to establish the diagnostic performance of the test prospectively. This can be accomplished by a suitably large cohort study in which serum spectroscopy is used alongside current clinical practice in one of the patient groups included in this evaluation. It would be appropriate to initially target the secondary care patient population, because the higher prevalence of disease in this group will reduce the sample size needed to accurately estimate diagnostic performance.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y Decision makers are often most interested in patient outcomes, such as survival, rather than intermediate outcomes, such as accuracy or speed of diagnosis (although this latter point is vital for treatment of high grade gliomas). From this perspective, a randomized trial, or a prospective cohort study with extended follow-up, may be required to fully establish the size of survival and quality-of-life benefits of including a serum spectroscopy test in the diagnostic pathway. A trial with primary outcomes relating to survival and quality-of-life would be specific to either the primary or secondary care setting (rather than generalisable to both), would need a large sample size, and would also require a follow-up period to capture survival benefits. In the case of malignant glioma this would require a period of at least 24 months. Such a trial would clearly be expensive, time consuming and may be unfeasible. Decision makers may be willing to make a decision on implementation of the blood test based on the modeled effects of improvements in intermediate outcomes on later patient outcomes. In this situation, the model proposed in this evaluation, populated with diagnostic performance and other data from a prospective trial, could be used to inform decisions about the wider adoption of the technology.
Future developments beyond trials such as emerging epidemiological evidence and new technologies should also be included in any future evaluations. It was not possible to foresee and include all such possible scenarios in this early evaluation but that should not preclude assessment in the light of new evidence. Updated analysis should inform any decisions about system wide implementation.
Several results in this analysis suggest cost savings through reduced use of imaging for patients with a negative test result. To make the case that a serum spectroscopic test can improve the efficiency of the diagnostic pathway prospective studies will also need to explore the impact of these test results on clinician and patient imaging study decisions. The possibility remains that the test may triage patients, but may not reduce the number of scans being conducted, and could potentially increase the demand on imaging. For example, if the test is applied to a wider population than intended in primary care due to the availability of such a non-invasive test effectively lowering the threshold for investigation. This highlights the need to stud decision making in this area prior to any implementation in primary care. Nevertheless, this triaging of patients would still benefit each patient that is provided with an early diagnosis.

Appendix 2 -Decision tree model
The decision node probabilities that populate the decision tree model are reported in Table S1. The model is simplified by assuming that the reference test (MRI/CT) has perfect accuracy (100% sensitivity and 100% specificity). This is desirable if we wish to evaluate the test against an MRI/CT only diagnostic pathway in which these imaging studies are considered as the 'gold standard'. It is also difficult to obtain valid estimates of MRI/CT sensitivity and specificity for this indication as these are not reported in a comparable manner in the literature. This simplification will not substantially alter the economic results compared to using more realistic values for the scenarios under consideration. This is because it is assumed confirmation of the diagnosis always requires imaging and therefore outcomes for patients that would hypothetically be classified correctly using this test but incorrectly by MRI/CT will remain the same whether or not the test is used. The effect of less than perfect accuracy of the reference test is therefore the same as a (small) decrease in prevalence as a proportion of patients with the disease cannot benefit from the additional test due to incorrect subsequent testing. Outcomes are calculated by 'rolling-back' the tree for both cancer cases and non-cases and taking an average of the two groups weighted by the prevalence of brain tumours in the scenario population. Referral decision S1:0·5, S2: 1 S1:0·5, S2: 0 S1 & S2: Scenarios 1 &2  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

Effect of testing on time-to-diagnosis and time-to-treatment
To estimate the expected time-to-diagnosis under a fast-track pathway it is assumed that this would match the current median time-to-diagnosis for patients presenting with brain tumour in emergency care as observed in Aggarwal et al 5 . This study reported a cohort of high grade glioma patients from a single hospital in London, UK. No other suitable estimates of time-to-treatment or time-to-diagnosis were identified in the literature. This is the most common primary brain tumour in the indications considered in this evaluation. Median time-todiagnosis for the standard pathway is also taken from the same source.

Effect of testing on use of imaging studies
The effect of spectroscopic testing on imaging study decisions of clinicians and patients is uncertain. Based on clinical expert input it was assumed, in the base case, that in secondary care all patients would continue to imaging, while in primary care 50% would continue to imaging following a negative spectroscopy result.

Effect of early diagnosis on patient outcomes
A systematic review of the literature failed to identify any studies that directly estimated the effect of earlier (or later) diagnosis on outcomes for primary brain tumours. Instead, based clinical expert guidance, estimates of the effects of time-to-treatment were sourced from the literature. The estimated effects are based on fitting natural history models to observational datasets of high grade glioma patients. 27,28 It is necessary to use a natural history model to produce survival estimates rather than using the survival data stratified by time-to-treatment directly, in order to adjust for potential confounding factors influencing both time-to-treatment and survival.
A one-to-one correspondence is assumed for the effects of an additional day added to the time-to-diagnosis and an additional day added to the time-to-treatment. Median survival by weeks between diagnosis and treatment are displayed in Table 2. The hazard ratio per additional day delay calculated in Do et al of 1·015 was applied to each day between diagnosis and treatment up to 28 days to calculate median survival time. Time between diagnosis and treatment beyond 28 days were fixed at the same median survival as 28 days.

Resource use and cost
It is assumed that there is no additional cost for converting a small proportion of standard referral to additional urgent referrals. This could be justified on the basis that patients receiving negative results allows more flexibility in the scheduling for these patients. This could create more opportunity to schedule patients with positive spectroscopy results sooner. As a larger proportion of patients are referred for immediate imaging it was assumed costs would increase because achieving this would require additional capacity. It was assumed that unit costs for MRI and CT grow exponentially with increasing proportion of immediate referrals such that costs double if 50% of patients are referred for immediate imaging.
Treatment costs are assumed to be the same for both fast-track and standard referrals that are diagnosed as cases. This may be a conservative assumption as treatment costs are likely to increase as disease progresses and therefore earlier treatment may, on average, be less costly than later treatment. However, it is also possible that The balance of these effects is unknown given the lack of appropriate data to estimate the effects. Secondary treatment and end-of-life costs are similarly assumed to be equal between fast track and standard referrals. This reflects the model assumption that all patients ultimately progress.
The time intervals considered (0-4 weeks) are too short for discounting to have an important impact on cost estimates therefore the effects of moving forward treatment on present value is not considered.

Primary care
An assumption that patients currently referred in direct-access imaging service would be selected for spectroscopic testing can provide a conservative estimate of the total number of patients likely to be tested. Three sources from the literature were considered: 1. A direct-access brain MRI service in Nottingham (UK) reported 130 patients per quarter (520 per year) among a population of 342,000. 32,33 The population of the UK is approximately 61 million therefore, assuming Nottingham is representative of the UK general population demand for brain MRI, there would be approximately 93,000 per year in the UK. 2. A direct access CT service for chronic headache in Glasgow (UK) reported 4404 patients referred over 8 years (551 per year) among a population of approximately 1 million. 31 Assuming this was representative of the UK generally then the demand for this type of service would be approximately 34,000 for the UK. 3. A direct access CT neuroimaging service in Lothian (UK) reported 389 exams per year for a population of approximately 442,000. 34 Assuming this was representative of the UK generally then the demand for this type of service would be approximately 54,000 for the UK.
The Glasgow estimate is particularly low and may result from limited uptake in the early part of the 8 year period for which data were reported. As it was not possible to determine if uptake increased over the period or was stable this estimate was not considered reliable. The Nottingham and Lothian estimates report potentially more stable and established services and are therefore more likely to reliably estimate demand. The likely demand for this novel test, if the indication is considered identical to neuroimaging, is therefore between 54,000 and 93,000. This is likely to be conservative as the test may well have a slightly broader indication than current neuroimaging. A reasonable point estimate may be approximately 75,000.

Secondary Care
An estimate of the total demand for serum spectroscopic tests in the UK can be made based on the reported incidence of brain tumour in the UK and the proportion of cases currently diagnosed through secondary (nonemergency) care. Annual incidence of malignant brain tumours in the UK is approximately 4,700. 35 Approximately one third of these are likely to present through secondary, non-emergency care. Using the positive predictive value for patients currently referred to neuroimaging in this setting of 3% suggests that for each case there will be 33 non-case patients receiving neuroimaging. In total, in the UK, this would imply 53,000 patients per year in secondary care may be suitable for spectroscopic testing.

Appendix 5 -Sensitivity Analysis
One-way sensitivity analysis (OWSA) OWSA results for a range of test sensitivity are displayed in S1 and S2 as a series of points on a costeffectiveness (CE) plane with incremental QALYs (intervention-control) on the x-axis and incremental costs (intervention -control) on the y-axis. Corresponding OWSA results for a range of test specificities in primary and secondary care are shown in Figure 3 and 4 in the main article text respectively. These are displayed with the ICER on the y-axis and the test specificity on the x-axis. Note that the estimated QALYs do not change with specificity in the model therefore changes in the ICER are due solely to changes in incremental costs.Error! Reference source not found. All OWSA analyses show results assuming the spectroscopy based test costs of £50 and £100.   Provide an explicit statement of the broader context for the study. Present the study question and its relevance for health policy or practice decisions.

Target population and subgroups 4
Describe characteristics of the base case population and subgroups analysed, including why they were chosen. Setting and location 5 State relevant aspects of the system(s) in which the decision(s) need(s) to be made. Study perspective 6 Describe the perspective of the study and relate this to the costs being evaluated. Comparators 7 Describe the interventions or strategies being compared and state why they were chosen. Time horizon 8 State the time horizon(s) over which costs and consequences are being evaluated and say why appropriate. Discount rate 9 Report the choice of discount rate(s) used for costs and outcomes and say why appropriate. Choice of health outcomes 10 Describe what outcomes were used as the measure(s) of benefit in the evaluation and their relevance for the type of analysis performed. Measurement of effectiveness 11a Single study-based estimates: Describe fully the design features of the single effectiveness study and why the single study was a sufficient source of clinical effectiveness data. 11b Synthesis-based estimates: Describe fully the methods used for identification of included studies and synthesis of clinical effectiveness data. Measurement and valuation of preference based outcomes 12 If applicable, describe the population and methods used to elicit preferences for outcomes.

Estimating resources and costs
13a Single study-based economic evaluation: Describe approaches used to estimate resource use associated with the alternative interventions. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs. 13b Model-based economic evaluation: Describe approaches and data sources used to estimate resource use associated with model health states. Describe primary or secondary research methods for valuing each resource item in terms of its unit cost. Describe any adjustments made to approximate to opportunity costs. Currency, price date, and conversion 14 Report the dates of the estimated resource quantities and unit costs. Describe methods for adjusting estimated unit costs to the year of reported costs if necessary. Describe methods for converting costs into a common currency base and the exchange rate. Describe all analytical methods supporting the evaluation. This could include methods for dealing with skewed, missing, or censored data; extrapolation methods; methods for pooling data; approaches to validate or make adjustments (such as half cycle corrections) to a model; and methods for handling population heterogeneity and uncertainty.

Study parameters 18
Report the values, ranges, references, and, if used, probability distributions for all parameters. Report reasons or sources for distributions used to represent uncertainty where appropriate. Providing a table to show the input values is strongly recommended. Incremental costs and outcomes 19 For each intervention, report mean values for the main categories of estimated costs and outcomes of interest, as well as mean differences between the comparator groups. If applicable, report incremental cost-effectiveness ratios. Characterising uncertainty 20a Single study-based economic evaluation: Describe the effects of sampling uncertainty for the estimated incremental cost and incremental effectiveness parameters, together with the impact  of methodological assumptions (such as discount rate, study perspective). 20b Model-based economic evaluation: Describe the effects on the results of uncertainty for all input parameters, and uncertainty related to the structure of the model and assumptions. Characterising heterogeneity 21 If applicable, report differences in costs, outcomes, or costeffectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.

Discussion
Study findings, limitations, generalisability, and current knowledge 22 Summarise key study findings and describe how they support the conclusions reached. Discuss limitations and the generalisability of the findings and how the findings fit with current knowledge.

Other
Source of funding 23 Describe how the study was funded and the role of the funder in the identification, design, conduct, and reporting of the analysis. Describe other non-monetary sources of support.

Conflicts of interest 24
Describe any potential for conflict of interest of study contributors in accordance with journal policy. In the absence of a journal policy, we recommend authors comply with International Committee of Medical Journal Editors recommendations.