Who gets referred for knee or hip replacement? A theoretical model of the potential impact of evidence-based referral thresholds using data from a retrospective review of clinic records from an English musculoskeletal referral hub

Objectives To estimate the relationship between patient characteristics and referral decisions made by musculoskeletal hubs, and to assess the possible impact of an evidence-based referral tool. Design Retrospective analysis of medical records and decision tree model evaluating policy changes using local and national data. Setting One musculoskeletal interface clinic (hub) in England. Participants 922 adults aged ≥50 years referred by general practitioners with symptoms of knee or hip osteoarthritis. Interventions We assessed the current frequency and determinants of referrals from one hub and the change in referrals that would occur at this centre and nationally if evidence-based thresholds for referral (Oxford Knee and Hip Scores, OKS/OHS) were introduced. Main outcome measure OKS/OHS, referrals for surgical assessment, referrals for arthroplasty, costs and quality-adjusted life years. Results Of 110 patients with knee symptoms attending face-to-face hub consultations, 49 (45%) were referred for surgical assessment; the mean OKS for these 49 patients was 18 (range: 1–41). Of 101 hip patients, 36 (36%) were referred for surgical assessment (mean OHS: 21, range: 5–44). No patients referred for surgical assessment were above previously reported economic thresholds for OKS (43) or OHS (45). Setting thresholds of OKS ≤31 and OHS ≤35 might have resulted in an additional 22 knee referrals and 26 hip referrals in our cohort. Extrapolating hub results across England suggests a possible increase in referrals nationally, of around 13 000 additional knee replacements and 4500 additional hip replacements each year. Conclusions Musculoskeletal hubs currently consider OKS/OHS and other factors when making decisions about referral to secondary care for joint replacement. Those referred typically have low OHS/OKS, and introducing evidence-based OKS/OHS thresholds would prevent few inappropriate (high-functioning, low-pain) referrals. However, our findings suggest that some patients not currently referred could benefit from arthroplasty based on OKS/OHS. More research is required to explore other important patient characteristics currently influencing hub decisions.


Strengths and limitations of this study
 This is the first study reporting data on the characteristics of patients being referred to a UK hub with symptoms of hip or knee osteoarthritis; such data are needed to assess the impact of changing referral criteria on the number of knee/hip replacements and on costs and quality of life.  We retrospectively reviewed clinical records for 922 men and women referred to one musculoskeletal hub.  We evaluated how the probability of referral varies with preoperative characteristics in current clinical practice.  We extrapolated our data nationally using a decision-tree and Markov model and estimated what impact that different policies might have based on several assumptions.  A pilot study would be required to quantify the real-world impact of any policy change, while data from multiple hubs and information on other patient characteristics (such as BMI) are required for a comprehensive understanding of current clinical practice. Recently, many clinical commissioning groups (CCGs) in the UK have set referral criteria for joint replacement that exclude patients with Oxford knee and hip scores (OKS and OHS) above certain thresholds, which are often as low as 19 [1] or 24 [1][2][3] out of 48. Recent papers have shown that these low thresholds are inappropriate: arthroplasty has a ≥80% chance of producing a meaningful improvement in patients with OKS/OHS of 19 or 24 [4] and would cost <£10,000 per quality-adjusted life-year (QALY). [5] Analysis of >400,000 medical records in the recent ACHE study suggested that thresholds between 31 and 41 may be justified on clinical and cost-effectiveness grounds. [4 5] Commissioners' decisions concerning thresholds may be influenced by the likely number of operations and their cost. In the UK, >94% of patients currently undergoing arthroplasty have OKS/OHS ≤31, demonstrating that nearly all operations currently conducted were costeffective and had ≥70% chance of meaningful improvement. [4 5] However, there are very little data on how patient characteristics influence the assessment process in practice, or on patients who do not currently receive surgery, making it difficult to assess the impact that different referral criteria may have on patient numbers or costs. As in many other clinical areas, hubs (also known as triage clinics, musculoskeletal interface services or intermediate musculoskeletal assessment centres) are increasingly used as gatekeepers determining access to consultations with orthopaedic surgeons, assessing all patients being considered for surgery. [6][7][8] This study aimed to describe the characteristics of patients attending hub consultations, identify how patients within a hub are currently managed and estimate the relationship between preoperative characteristics and referral decisions. To illustrate how such data could be used to inform policy, we estimated the potential impact of a policy change by which patients are referred from a hub based on evidence-based OKS/OHS thresholds.

Outline
We collected data from one UK referral hub and used it to estimate a logistic regression model predicting the probability that patients will be referred from the hub to surgical assessment based on their OKS/OHS, age and sex. We then applied this regression model to nationally-collected preoperative outcomes data [10][11][12][13][14][15] to estimate how many patients are referred for surgical assessment nationally. We predicted how many of these patients would be referred through to secondary care if different OKS/OHS thresholds were introduced and potentially how many extra operations would be performed.

Hub data collection
Anonymised data were extracted retrospectively from medical records of patients with knee or hip symptoms who had been referred by GPs between July 2015 and July 2016 to the musculoskeletal hub at the Nuffield Orthopaedic Centre (NOC) in Oxford. This audit of practice was conducted under guidance from the Hub Lead and Clinical Director of Oxford University Hospitals (OUH R&D Ref 11603). The primary statistical analysis used a prognostic model to estimate how the probability of patients attending face-to-face consultations at the hub being referred to surgical assessments in secondary care varied with OKS/OHS, age and sex. We therefore chose the sample size to provide data on ≥30 knee referrals and ≥30 hip referrals from the hub to secondary care, thereby providing ≥10 events per explanatory variable. [9] Two medically-qualified surgical research fellows extracted the following: age; sex; OKS/OHS at hub attendance; attendance date; whether the patient was referred to secondary care; date of any subsequent surgical assessment visit; and date of any subsequent arthroplasty surgery. Additional information on imaging, referrals to other clinics, and other surgeries or diagnoses was recorded in free text fields. Data on OKS at the surgical assessment visit were also extracted, when available, from the secondary care records of patients with knee pain. Body mass index (BMI) was not extracted as the analysis focused on OKS/OHS and available evidence on how capacity to benefit and cost-effectiveness vary with OKS/OHS does not consider BMI. [4 5] Patients who were on the waiting list for arthroplasty surgery at the time of data extraction (August 2016) and those for whom surgery was delayed due to comorbidities or high BMI after their attendance at the surgical assessment consultation were counted as having been referred for surgery.
Exclusion criteria comprised:  Aged <50 years (for whom knee/hip pain is unlikely to be caused by osteoarthritis).  Evidence from medical records that symptoms were due to a condition other than osteoarthritis.  Previous arthroplasty on the same joint.  Medical records inaccessible for research.  Any attendance at the hub or surgical assessment unit before July 2015.
However, patients who were referred for x-rays, magnetic resonance imaging (MRI) or physiotherapy but did not attend face-to-face consultations at the hub or surgical assessment unit were included in the descriptive analysis.

Statistical modelling
Logistic regression was used to estimate a prognostic model of the local hub data predicting how OKS/OHS, age and sex affected the odds of being referred for surgical assessment following a face-to-face attendance at the hub. Age and OKS/OHS at hub attendance were analysed as continuous variables. Explanatory variables were selected based on the Akaike information criterion using forward stepwise regression, manually testing variables in a prespecified sequence (online supplementary appendix). Analyses were conducted in Stata version 14 (StataCorp, College Station TX) on a complete case basis, excluding patients with missing data. We analysed binary variables using two-sample tests of proportions and continuous variables using unpaired t-tests preceded by F-tests for equal variance. OKS/OHS in the hub sample was compared against population means [10][11][12][13][14][15] using one-sample t-tests.

Estimating the number of referrals and effect of introducing thresholds
A decision-tree model of the treatment pathway was developed based on hub data and consultations with hub staff. Using the model, hub results were extrapolated to estimate the number of patients in England who are referred for hip or knee replacement (online supplementary appendix). The overall proportion of patients referred directly to surgical assessment and the proportion attending face-to-face consultations at the hub were calculated from the hub sample. We also estimated the proportion of patients who underwent arthroplasty after being referred from the hub to surgical assessment. Logistic regression results were used to predict the probability of being referred from the hub to surgical assessment for men and women aged 50, 60, 70, 80 and 90 with different OKS/OHS. For each patient group defined by age, sex and OKS/OHS, we then estimated the number of patients referred to hubs nationally by dividing national data on the number of joint replacements currently conducted for osteoarthritis in England [10][11][12][13][14][15] by the probability that patients in each group would undergo surgery.
We then estimated how the number of referrals to surgical assessment and the number of arthroplasty procedures might change if different OKS/OHS thresholds between 18 and 45 were used to determine referral decisions during face-to-face hub consultations. This analysis made the following assumptions: 1) Introducing OKS/OHS thresholds at the hub was assumed to have no effect on GP referrals, the hub triage or the probability of a patient with particular preoperative characteristics being referred for arthroplasty following a surgical assessment visit. 2) Since the hub audit provided no data on OKS/OHS for patients who did not attend the hub, we assumed that the probability of direct referral to surgical assessment, the probability of attending a face-to-face visit at the hub, and the probability of patients referred to surgical assessment subsequently being referred for surgery were independent of OKS/OHS, age and sex. Although symptom severity is one of the main factors considered in the hub triage, this assumption is unlikely to affect estimates of the impact of changing referral criteria applied to those patients attending the hub. 3) We assumed that the referral probabilities estimated from the NOC hub sample are broadly representative of clinical practice across the UK. We assumed that 50% of patients who are currently not referred for surgical assessment after a face-to-face hub visit would still not be considered candidates for arthroplasty (and therefore not referred) regardless of whether thresholds were introduced. This group of patients includes patients who choose not to be referred, those who have not had a complete trial of non-operative treatment and others who are be unsuitable for surgery due to comorbidities or other factors. The remaining 50% of patients were assumed to be referred only if their OKS/OHS was below the threshold. 5) Each 40-minute hub attendance was estimated to cost £58, and surgical assessment £132 (online supplementary appendix). 6) The analyses took the perspective of the NHS, focusing on costs related to hip/knee arthroplasty. Costs of GP consultations, hub triage, x-rays, imaging, physiotherapy, injections, weight loss programs, missed appointments and referrals to other clinics were excluded as there is no reason to expect the proportion of patients requiring these services to change following the introduction of OKS/OHS thresholds. 7) The reference year for costs was 2014.
The patient numbers calculated within the decision-tree were used to estimate the cost of the referral pathway. The total 10-year costs and total QALYs estimated in a related study [5] for different patient characteristics with and without arthroplasty were applied to the number of patients expected to undergo arthroplasty or have no arthroplasty in each scenario. These figures were used to estimate the net health benefit for each scenario [16 17] (assuming a £20,000/QALY ceiling ratio) and the incremental cost-effectiveness of different pairs of scenarios (online supplementary appendix).
We also estimated the number of additional referrals that might have been observed within the population of individuals attending the NOC hub if a fixed OKS/OHS threshold were introduced. Based on assumption (4), this was equal to half the number of patients who had OKS/OHS at or below the proposed threshold but were not referred, minus the number who were referred with OKS/OHS above the proposed threshold.

Public and Patient Involvement
Patient representatives were involved in the grant application and design of the wider ACHE study. [4] Results of this part of the study were discussed with a user group including clinicians and patient representatives; this discussion helped inform the interpretation and presentation of results. At the time of this audit, the musculoskeletal hub service was the only route for patients in Oxfordshire to access NHS elective orthopaedic surgery. When GPs refer patients to secondary care with knee/hip symptoms, the referral letter and x-rays are first reviewed by senior hub staff, who decide whether to (a) refer the patient directly to surgical assessment, (b) invite them to attend the hub clinic for an individual triage appointment or (c) manage them in primary care ( Figure 3). Triage is based on patient symptoms as described in the GP's referral letter (which will include BMI and may include OKS/OHS) and conservative treatments already given (such as anti-inflammatory drugs, physiotherapy and weight loss strategies).
Of the osteoarthritis patients included in the analysis, 23% (71/315) of knee patients and 39% (236/607) of hip patients were referred directly to surgical assessment based on the hub triage. Anecdotally, those patients referred directly to surgical assessment tend to be those with severe symptoms, those who had already exhausted all conservative measures and those who had previously been referred for surgical assessments, but had chosen not to have surgery at that time. Among 27 patients referred directly to surgical assessment for whom OKS data were available, the mean OKS was 15 (range: ; no such data were available for hip patients. Those patients who are referred for surgical assessment discuss the risks and benefits of surgery with an orthopaedic surgeon and make an informed decision about whether or not to undergo arthroplasty or other surgery, taking into account their comorbidities, symptom severity and other factors. Among the patients who were referred directly for surgical assessment and had outcomes data available, 56% (38/68) of knee patients and 69% (161/235) hip patients underwent or were awaiting knee/hip replacements at the time of data extraction.
Following triage, 36% (114/315) of knee patients and 44% (265/607) hip patients did not attend face-to-face consultations at either the hub or the surgical assessment. Typically such patients comprise those who can be managed in primary care: for example those with mild symptoms and those who have not yet exhausted conservative treatment options, such as advice and information, activity and exercise and weight loss. [18][19][20] Across the 379 hip and knee patients without face-to-face consultations, 270 (71%) had x-rays, MRI or ultrasound to identify the most appropriate care pathway. Eleven (3%) were referred for physiotherapy, while seven patients (2%), all with hip osteoarthritis, had injections. Anecdotally, other patients may not be seen face-to-face as they are unfit for surgery or have recent injuries likely to heal without further intervention. Eighteen patients opted for non-NHS care, missed/failed to book hub appointments, died before the hub attendance or were still awaiting a hub appointment at the time of data extraction (Figures 1 and 2).
The remaining 44% (130/315) of knee patients and 17% (106/607) of hip patients attended face-to-face assessments at the musculoskeletal hub, where extended-scope specialist physiotherapists or orthopaedic fellows assess patients to confirm diagnosis. Patients routinely complete OKS/OHS questionnaires to assess whether the level of symptoms warrants surgery and to guide discussions about the symptom profile. Among the patients attending the hub who had complete data, the mean OKS was 21 (range: 1-48), while mean OHS was 24 (range: ; both values were significantly higher than the average for patients undergoing arthroplasty nationally (mean: 18; p≤0.015), although only OHS was significantly different from the average for patients undergoing arthroplasty in Oxfordshire (mean OKS: 20, p=0.134; mean OHS: 19 p<0.0001). [10][11][12][13][14][15] Diagnostic imaging, landmark injections and injections for trochanteric bursitis may be done during the hub consultation. Patients with BMI ≥40 are referred for monitored weight loss programs that must be followed for 12 months before considering surgery. Staff and patients also discuss the risks and benefits of joint replacement (including recovery times, the need for support at home after hospital discharge and the potential need for revision surgery) and how these may be affected by patients' living arrangements and comorbidities. Following the hub visit, 45% (49/130) of knee patients and 36% (36/101) of hip patients were referred for surgical assessment. This includes patients who are considered candidates for arthroplasty or other procedures (e.g. arthroscopy, anterior cruciate ligament repair or interventional radiology).
Patients with higher OKS/OHS were significantly less likely to be referred for surgical assessment: each one-point increase in OKS reduced the odds of referral by 4.7% (p=0.019), while a one-point increase in OHS reduced the odds by 3.9% (p=0.062). Those patients who are not referred to surgical assessment may be referred for physiotherapy, other nonsurgical management or other outpatient clinics (e.g. rheumatology or sports injury clinics), or may choose not to be referred as they prefer not to undergo surgery at the current time.
Across the hub attendees who were referred for surgical assessment and had data on clinical outcomes, 30% (12/40) of knee patients and 37% (10/27) of hip patients underwent or were awaiting arthroplasty surgery.
Applying estimates of the probability of referral and subsequent arthroplasty from the hub audit to national data on the distribution of patients undergoing knee/hip arthroplasty suggests that GPs in England refer around 417,000 patients with knee osteoarthritis and around 235,000 patients with hip osteoarthritis to secondary care each year ( Figure 3). Of these, around 172,000 knee patients and 41,000 hip patients might attend a hub if the care pathway in Oxfordshire were followed nationally; such hub attendances would cost a total of £12 million.

Effect of introducing referral thresholds
We used the hub data to estimate the impact that different policies for using OKS/OHS to guide referral decisions might have on patient numbers, costs and health outcomes. The ACHE study reported that people with OKS ≤31 have a ≥70% chance of achieving a sevenpoint increase in OKS following knee arthroplasty. [4] Of the 110 knee patients attending the NOC hub, 94 (85%) had OKS ≤31, of whom 47 (50%) were referred for surgical assessment. Two patients were referred with OKS of 32 or 41. Assuming that 50% of the NOC patients with OKS ≤31 who are not currently referred might still choose not to be referred or might be considered clinically inappropriate for other reasons, a fixed referral threshold OKS of 31 would have resulted in an additional 22 referrals to surgical assessment: a 49% increase.
Applying the results of logistic regression to the whole population undergoing surgery nationally suggested that an OKS threshold of 31 would result in 43,000 additional surgical assessment visits and 13,000 additional knee replacement procedures in England each year (Table 1). Based on the costs and outcomes for patients with and without knee arthroplasty that were estimated in a recent economic evaluation, [5] introducing this policy would cost the NHS an additional £36 million for each annual cohort of patients (of which £5.8 million would be due to additional surgical assessment visits), but would gain 30,000 QALYs. This policy would be highly cost-effective compared with current practice, costing just £1,195 per QALY gained, and the net health benefit would be 28,000 QALYs even after subtracting the health benefits that would be foregone by spending £36 million on knee arthroplasty candidates, rather than other conditions. Introducing higher OKS thresholds would produce still greater health benefits and would continue to be cost-effective at a £20,000/QALY ceiling ratio right up to the economic threshold of 43. [5] Similarly, ACHE predicts that people with preoperative OHS ≤35 have a ≥70% chance of achieving an eight-point increase in OHS following hip arthroplasty. [4] Within the NOC hub, 87% (88/101) of patients had OHS ≤35, of whom 35 were referred. One patient was referred with OHS of 44. Introducing a threshold of 35 might therefore result in 26 additional referrals: a 30% increase. Extrapolating across England, we might expect 14,000 additional surgical assessments and 5,000 additional hip replacements each year. The impact of this policy is lower than for knee osteoarthritis as 90% of hip replacements are done in patients who were referred directly, and did not attend the hub. The policy would cost an additional £25 million (of which additional surgical assessments account for £2 million) and gain 16,000 QALYs, costing £1,560/QALY gained and generating a net health benefit of 15,000 QALYs. Introducing higher thresholds (up to the economic threshold of 45) would produce greater health benefits and continue to cost <£20,000/QALY gained.

Discussion
Although referral hubs are being used increasingly as gatekeepers determining access to consultations in different clinical areas of the NHS, there has been little research on them. This study has characterised the patients being referred to and from a musculoskeletal hub, and showed that OKS/OHS and other factors were considered when deciding which patients to refer for surgical assessment. Since most patients attending hubs have relatively severe osteoarthritis symptoms and those with high OKS/OHS are less likely to be referred for surgery, introducing evidence-based OKS/OHS thresholds for arthroplasty is unlikely to prevent large numbers of inappropriate referrals. However, since recent evidence demonstrates that arthroplasty is both cost-effective and highly likely to benefit people with OKS/OHS well above the thresholds currently used by some CCGs, [4 5] such policies are also likely to identify patients who are not currently referred, but for whom arthroplasty could be expected to be beneficial and cost-effective. These numbers are uncertain, but extrapolations from the hub examined here suggest that, across England, setting thresholds of OKS ≤31 and OHS ≤35 would result in many more patients being referred for surgical assessment, and could result in an additional 13,000 knee replacements and 4,500 hip replacements each year. More evidence is therefore urgently needed on the other factors currently influencing both the decision to refer from the hub to surgical assessment and the decision on whether arthroplasty is the appropriate treatment option.
To our knowledge our study is the first to report the characteristics of patients attending a musculoskeletal hub. However, the analysis was based on a small sample and included only eight hub attendees with OKS/OHS ≥40. Data were analysed one month after the end of the period studied; some patients may therefore have gone on to have arthroplasty after physiotherapy or weight loss treatment. Furthermore, the retrospective review of medical records may not have identified all patients meeting exclusion criteria -particularly for patients who did not attend either the hub or surgical assessment. Furthermore, no data were collected on BMI.
Estimates of the impact of different policies rely on additional assumptions and represent an approximate indication of potential patient numbers; pilot studies would be required to assess the true impact in practice. In particular, we focused on the impact of changing referral guidelines at hubs, as no data were available on patients visiting GPs with osteoarthritis symptoms. In practice, any decision aid or referral guideline is also likely to affect GPs' referral decisions, the hub triage process and (potentially) consultations with orthopaedic surgeons. The impact of such policies could be much greater in these settingsparticularly for hip replacement, where only 10% of patients undergoing arthroplasty attend the hub. We also assumed, arbitrarily, that 50% of people with OKS/OHS below the threshold who are not currently referred would choose not to be referred for surgery even if this were offered.
Although access to joint replacement may be restricted if funding for interventions is reduced, results suggest that increased numbers of knee/hip replacements could be justified on cost-effectiveness grounds. Our results suggest that the NHS is currently willing to pay no more £2,000 per QALY gained from arthroplasty, while NICE routinely approves treatments for patients with other conditions (e.g. rheumatoid arthritis) that cost £20,000 [23] or even £40,000 [24]/QALY. This suggests that it may be more efficient and more equitable to spend limited NHS funds on conducting more joint replacements, increasing the operation rate in the UK to a level similar to that of Austria or Germany,[25] rather than directing these resources to other conditions. However, the number of procedures is also limited by availability of surgeons, operating theatres and hospital beds, which may mean that it is not currently feasible to conduct all of the operations that could be justified on cost-effectiveness grounds. And, as noted earlier, OKS/OHS is clearly not the only factor taken into account in decision-making by hubs and surgeons. In particular, patient choice plays a substantial role and comorbidities and other factors are also taken into account.
Referral hubs are an effective mechanism for ensuring that patients receive cost-effective treatment in the most appropriate healthcare setting and patients are generally satisfied with this model of care. [8] One aim of introducing hubs was to improve efficiency in the referral pathway by increasing the proportion of referrals to secondary care that lead to surgery. The hub identifies patients who do not want surgery or who have mild symptoms and directs patients to other effective interventions when surgery is not appropriate. If the service works well, only patients who are suitable for surgery are referred. However, the process and criteria for arthroplasty referrals still vary greatly between CCGs. [1] Explicit guidelines and decision aids could help reduce geographical inequity and ensure that all patients likely to benefit have fair access to cost-effective treatments. Although our study is based on retrospective data from only one hub, it provides initial estimates of what potential patient numbers and costs might be if a more explicit threshold-based approach was adopted nationally. While arthroplasty rates and patient characteristics in Oxfordshire are similar to the national average,[26 27] referral guidelines and pathways vary substantially between CCGs, suggesting that a broader audit covering multiple hubs is necessary to provide a more comprehensive picture of current clinical practice. Ideally this would also collect more detailed patient information (e.g. BMI).
In conclusion, our paper has estimated the effect of introducing evidence-based OKS/OHS thresholds in England. While our results can be used to estimate the impact of other policies, including those where thresholds vary by age or other patient characteristics, a

Declarations
Competing interest statement All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: a grant from NIHR for the submitted work; HD reports personal fees from Halyard Health, outside the submitted work; DB reports grants from Zimmer Biomet, outside the submitted work; AP reports personal fees from Zimmer Biomet and DePuy outside the submitted work; all declare that there are no other relationships or activities that could appear to have influenced the submitted work. The other authors declare that there are no relevant conflicts of interests besides the NIHR grant that supports the current work.

Details of contributors
HD is the guarantor for this study. AP, DB, HD, AG and KB conceived, designed and conducted the ACHE study. HD, AP and JS conceived and designed the audit, which was conducted by HD and JS with assistance from Reza Mafi, and Sarah Dorman. HD, AG and PE conceived and designed the economic evaluation. HD conducted the statistical and economic analysis under the supervision of AG. AP and KB helped interpret the results. HD drafted the manuscript. All authors edited the manuscript for important intellectual content and approved the final version.
The following members of the ACHE study group all contributed to the developing the research question, planning the protocol of work, gaining funding and/or delivery of this research project: Ray Fitzpatrick, David Murray, Nigel Arden, Andy Carr, Karen Barker, James Smith, Stephanie Smith, Kristina Harris, Rob Middleton, Elizabeth Gibbons, Elena Benedetto, Jill Dawson, Adrian Sayers, Laura Miller, Elsa Marques, Rachael Gooberman-Hill, Ashley Blom, Andrew Judge, and Sion Glyn-Jones.

Copyright/licence
The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, an exclusive licence (or non-exclusive for government employees) on a worldwide basis to the BMJ Publishing Group Ltd and its Licensees to permit this article (if accepted) to be published in BMJ Open and any other BMJPGL products to exploit all subsidiary rights, as set out in our licence http://authors.bmj.com/submitting-yourpaper/copyright-and-authors-rights/ and the Corresponding Author accepts and understands that any supply made under these terms is made by BMJPGL to the Corresponding Author. Where the Corresponding Author wishes to make the article available on an open access basis (and intends to pay the relevant open access fee), the terms of such open access shall be governed by a Creative Commons licence -details of these licences and which Creative Commons licence will apply to this article are set out in our licence referred to above.

Ethics statement
The audit of practice was conducted under guidance of the Hub Lead and Clinical Director of OUH (OUH R&D Ref 11603) and did not require ethical approval. The audit was conducted retrospectively on patients who had received normal clinical management. All data were anonymised by clinical staff.

Funding and role of study sponsors
The ACHE study and economic evaluation was funded by the NIHR Health Technology Assessment Programme (project number 11/63/01) and will be published in full in Health Technology Assessment. Further information available at: https://www.journalslibrary.nihr.ac.uk/programmes/hta/116301. This report presents independent research commissioned by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, MRC, CCF, NETSCC, the Health Technology Assessment Programme, NHS, NHS Digital or the Department of Health. The funder had no involvement in: the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication. The authors had full access to the data (including statistical reports and tables), can take responsibility for the integrity of the data and the accuracy of the data analysis and are responsible for submitting this paper.

Transparency declaration
The manuscript's guarantor (HD) affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

Data sharing
The decision tree and Markov models used in this paper are available from the corresponding author on request. PROMs/HES data are available from NHS Digital (http://content.digital.nhs.uk/dars). with data cleaning. Data on NHS PROMs linked to HES APC data were reused with the permission of NHS Digital, copyright 2015, with all rights reserved.

Patient involvement
The research question and study design (prior to funding) was informed by patient interest groups. The study was based on pre-existing data from medical records so no patients were directly recruited in the course of the work. Throughout the study, a User Group, including patients and other stakeholders, informed progress and development of the work. Results of the audit were discussed with user group including clinicians and patient representatives, to inform the presentation of results. The results will not be directly disseminated to patients whose records were included in the audit, but the publication will be available for the target population.    Table A2 in online supplementary appendix for a list of the conditions other than osteoarthritis for which patients aged ≥50 years were excluded from the analysis.   Table A2 in online supplementary appendix for a list of the conditions other than osteoarthritis for which patients aged ≥50 years were excluded from the analysis.

Methods for stepwise regression models
We selected variables for inclusion in the logistic regression model in a pre-specified sequence. We first estimated a model only including the constant term and evaluated whether adding OKS/OHS to the model reduced Akaike information criterion (AIC); if AIC was reduced, the model used in the next step controlled for OKS/OHS and otherwise remained with just a constant term. Similarly, the next step assessed whether adding age into the model reduced AIC and in the final step we assessed whether adding sex into the model reduced AIC. We specified the order in which variables were considered for the model (OKS/OHS, then age, then sex) in advance of data analysis: it was hypothesised that OKS/OHS would be the most important factor determining referrals, followed by age (since this is a proxy for comorbidities) and finally sex (since we had no prior hypotheses about how sex would affect referrals). Age and OKS/OHS were assumed to have linear effects on the log-odds of being referred for surgical assessment as there were insufficient data to evaluate nonlinear functions.

Methods for estimating patient numbers
Hub data were extrapolated to estimate patient numbers across England by applying the probability of referral estimated on the hub data to national data on the characteristics of patients undergoing arthroplasty and the total number of operations.
We considered 490 groups of patients, defined by gender, age (using age bands <55, 55-64, 65-74, 75-84 and ≥85) and individual preoperative OKS/OHS between 0 and 48. An extract of individual patient data from the national patient reported outcome measures (PROMs) dataset linked to Hospital Episode Statistics (HES) covering operations conducted between 2009 and 2015 was used to estimate the proportion of arthroplasty patients in different groups defined by age and gender. The same dataset was used to obtain the proportion of patients with each preoperative OKS/OHS. We then calculated the number of patients in each group who are likely to have been referred by GPs with knee/hip symptoms by dividing the number of operations in each group by the probability that each group of patients will undergo surgery.

Unit costs
The unit costs used in the current analysis are shown in Table A1. The costs and QALYs accrued in the 10-year period after arthroplasty (or a decision not to perform arthroplasty) have been reported previously. [5]

Additional regression results
For patients attending the hub with knee symptoms, adding age and sex into the model predicting the odds of hub attendees being referred for surgical assessment increased AIC, so the final model included only OKS (Table A4). Similarly, in the model of hip patients, adding age into the model worsened AIC, so the final model included only OHS and gender.
Omitting gender from the model had negligible effect on the coefficient for OHS, which was 0.0397 (SE: 0.0211) when OHS was the only explanatory variable. In addition to the logistic regression model predicting the odds of patients being referred from the hub to surgical assessment, we also conducted a secondary analysis predicting the odds of undergoing arthroplasty in patients who had already been referred to surgical assessment by the hub, using the same methods. This analysis suggested that OKS/OHS, age and sex had no significant effect on the odds of referral from surgical assessment to surgery (p>0.05) and that including any of these variables in the regression increased AIC compared with a model with only a constant term. This validates the assumption that the probability of patients referred directly to surgical assessment subsequently being referred for surgery is independent of OKS/OHS, age and sex and the assumption that introducing OKS/OHS thresholds at the hub would have no effect on the probability of a patient with particular preoperative characteristics being referred for arthroplasty following a surgical assessment visit.  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

Strengths and limitations of this study
 This is the first analysis reporting data on the characteristics of patients being referred to a UK hub with symptoms of hip or knee osteoarthritis; such data are needed to assess the impact of changing referral criteria on the number of knee/hip replacements and on costs and quality of life.  We retrospectively reviewed clinical records for 922 men and women referred to a single musculoskeletal hub, although only 221 patients underwent arthroplasty.  We used several assumptions to extrapolate our estimates of how the probability of referral varies with preoperative characteristics using a decision-tree and Markov model, which enabled us to estimate the potential impact of different referral policies nationally.  A prospective pilot study would be required to quantify the real-world impact of any policy change, while data from multiple hubs and information on other patient characteristics (such as BMI) are required for a comprehensive understanding of current clinical practice. Recently, many clinical commissioning groups (CCGs) in the UK have set referral criteria for joint replacement that exclude patients with Oxford knee and hip scores (OKS and OHS) above certain thresholds, which are often as low as 19 [1] or 24 [1][2][3] out of 48. Recent papers have shown that these low thresholds are inappropriate: arthroplasty has a ≥80% chance of producing a meaningful improvement in patients with OKS/OHS of 19 or 24 [4] and would cost <£10,000 per quality adjusted life-year (QALY). [5] The recent ACHE study analysed >400,000 medical records and concluded that thresholds between 31 and 41 may be justified on clinical and cost-effectiveness grounds. [4 5] Commissioners' decisions concerning thresholds may be influenced by the likely number of operations and their cost. In the UK, >94% of patients currently undergoing arthroplasty have OKS/OHS ≤31, demonstrating that nearly all operations currently conducted were costeffective and had ≥70% chance of meaningful improvement. [4 5] However, there are very little data on how patient characteristics influence the assessment process in practice, or on patients who do not currently receive surgery, making it difficult to assess the impact that different referral criteria may have on patient numbers or costs. As in many other clinical areas, hubs (also known as triage clinics, musculoskeletal interface services or intermediate musculoskeletal assessment centres) are increasingly used as gatekeepers determining access to consultations with orthopaedic surgeons, assessing all patients being considered for surgery. [6][7][8] We aimed to review the characteristics of patients attending hub consultations, identify how patients within a hub are currently managed and explore how referral decisions vary with preoperative characteristics. To illustrate how such data could be used to inform policy, we estimated the potential impact of a policy change by which patients are referred from a hub based on evidence-based OKS/OHS thresholds.

Outline
We collected data from one UK referral hub and used it to estimate a logistic regression model predicting the probability that patients will be referred from the hub to surgical assessment based on their OKS/OHS, age and sex. We then applied this regression model to nationally-collected preoperative outcomes data [9][10][11][12][13][14] to estimate how many patients are referred for surgical assessment nationally. We predicted how many of these patients would be referred through to secondary care if different OKS/OHS thresholds were introduced and potentially how many extra operations would be performed.

Hub data collection
Anonymised data were extracted retrospectively from medical records of patients with knee or hip symptoms who had been referred by GPs between July 2015 and July 2016 to the musculoskeletal hub at the Nuffield Orthopaedic Centre (NOC) in Oxford. This study discussed and agreed through the Clinical Governance Group for hip and knee replacement as part of a wider review of outcomes in hip and knee arthroplasty (Oxford University Hospitals Research and Development Ref 11603). The primary statistical analysis used a prognostic model to estimate how the probability of patients attending face-to-face consultations at the hub being referred to surgical assessments in secondary care varied with OKS/OHS, age and sex. We therefore chose the sample size to provide data on ≥30 knee referrals and ≥30 hip referrals from the hub to secondary care, thereby providing ≥10 events per explanatory variable. [15] Two medically-qualified surgical research fellows extracted the following: age; sex; OKS/OHS at hub attendance; attendance date; whether the patient was referred to secondary care; date of any subsequent surgical assessment visit; and date of any subsequent arthroplasty surgery. Additional information on imaging, referrals to other clinics, and other surgeries or diagnoses was recorded in free text fields. Data on OKS at the surgical assessment visit were also extracted, when available, from the secondary care records of patients with knee pain. Body mass index (BMI) was not extracted as the analysis focused on OKS/OHS and the available evidence on how capacity to benefit and cost-effectiveness vary with OKS/OHS does not consider BMI. [4 5] Patients who were on the waiting list for arthroplasty surgery at the time of data extraction (August 2016) and those for whom surgery was delayed due to comorbidities or high BMI after their attendance at the surgical assessment consultation were counted as having been referred for surgery.
Exclusion criteria comprised:  Aged <50 years (for whom knee/hip pain is unlikely to be caused by osteoarthritis).  Evidence from medical records that symptoms were due to a condition other than osteoarthritis.  Previous arthroplasty on the same joint.  Medical records inaccessible for research.  Any attendance at the hub or surgical assessment unit before July 2015.
However, patients who were referred for x-rays, magnetic resonance imaging (MRI) or physiotherapy but did not attend face-to-face consultations at the hub or surgical assessment unit were included in the descriptive analysis.

Statistical modelling
Logistic regression was used to estimate a prognostic model of the local hub data predicting how OKS/OHS, age and sex affected the odds of being referred for surgical assessment following a face-to-face attendance at the hub. Age and OKS/OHS at hub attendance were analysed as continuous variables. Explanatory variables were selected based on the Akaike information criterion using forward stepwise regression, manually testing variables in a prespecified sequence (online supplementary appendix). Analyses were conducted in Stata version 14 (StataCorp, College Station TX) on a complete case basis, excluding patients with missing data. We analysed binary variables using two-sample tests of proportions and continuous variables using unpaired t-tests preceded by F-tests for equal variance. OKS/OHS in the hub sample was compared against population means [9][10][11][12][13][14] using one-sample t-tests.

Estimating the number of referrals and effect of introducing thresholds
A decision-tree model of the treatment pathway was developed based on hub data and the authors' clinical experience of running hubs and/or surgical clinics. Using the model, hub data were extrapolated to estimate the number of patients in England who are referred for hip or knee replacement (online supplementary appendix). The overall proportion of patients referred directly to surgical assessment and the proportion attending face-to-face consultations at the hub were calculated from the hub sample. We also estimated the proportion of patients who underwent arthroplasty after being referred from the hub to surgical assessment. Logistic regression results were used to predict the probability of being referred from the hub to surgical assessment for men and women aged 50, 60, 70, 80 and 90 with different OKS/OHS. For each patient group defined by age, sex and OKS/OHS, we then estimated the number of patients referred to hubs nationally by dividing national data on the number of joint replacements currently conducted for osteoarthritis in England [9][10][11][12][13][14] by the probability that patients in each group would undergo surgery.
We then estimated how the number of referrals to surgical assessment and the number of arthroplasty procedures might change if different OKS/OHS thresholds between 18 and 45 were used to determine referral decisions during face-to-face hub consultations. This analysis made the following assumptions: 1) Introducing OKS/OHS thresholds at the hub was assumed to have no effect on GP referrals, the hub triage or the probability of a patient with particular preoperative characteristics being referred for arthroplasty following a surgical assessment visit. 2) Since patient records provided no data on OKS/OHS for patients who did not attend the hub, we assumed that the probability of direct referral to surgical assessment or the probability of attending a face-to-face visit at the hub were independent of OKS/OHS, age and sex. Although symptom severity is one of the main factors considered in the hub triage, this assumption is unlikely to affect estimates of the impact of changing referral criteria applied to those patients attending the hub. 3) We also assumed that the probability of patients referred to surgical assessment subsequently being referred for surgery was independent of OKS/OHS, age and sex, which was supported by a secondary regression analysis (online supplementary material). 4) We assumed that the referral probabilities estimated from the NOC hub sample are broadly representative of clinical practice across the UK. 5) We assumed, based on the experience of clinical co-authors, that 50% of patients who are currently not referred for surgical assessment after a face-to-face hub visit would still not be considered candidates for arthroplasty (and therefore not referred) regardless of whether thresholds were introduced. This group of patients includes patients who choose not to be referred, those who have not had a complete trial of nonoperative treatment and others who are unsuitable for surgery due to comorbidities or other factors. The remaining 50% of patients were assumed to be referred only if their OKS/OHS was below the threshold. 6) Since there are no data on how OKS/OHS changes over time in the absence of arthroplasty, we assumed for simplicity patients who do not undergo arthroplasty following their hub attendance would not be referred back to clinic for 10 years. 7) Each 40-minute hub attendance was estimated to cost £58, and surgical assessment £132 (online supplementary appendix, Table A1). 8) The analyses took the perspective of the NHS, focusing on costs related to hip/knee arthroplasty. Costs of GP consultations, hub triage, x-rays, imaging, physiotherapy, injections, weight loss programs, missed appointments and referrals to other clinics were excluded as there is no reason to expect the proportion of patients requiring these services to change following the introduction of OKS/OHS thresholds. 9) The reference year for costs was 2014.
The patient numbers calculated within the decision-tree were used to estimate the cost of the referral pathway. The total 10-year costs and total QALYs estimated in a related study [5] for different patient characteristics with and without arthroplasty were applied to the number of patients expected to undergo arthroplasty or have no arthroplasty in each scenario. These figures were used to estimate the net health benefit for each scenario [16 17] (assuming a £20,000/QALY ceiling ratio) and the incremental cost-effectiveness of different pairs of scenarios (online supplementary appendix).
We also estimated the number of additional referrals that might have been observed within the population of individuals attending the NOC hub if a fixed OKS/OHS threshold were introduced. Based on assumption (4), this was equal to half the number of patients who had OKS/OHS at or below the proposed threshold but were not referred, minus the number who were referred with OKS/OHS above the proposed threshold.

Current referral pathway
From the 1,638 records reviewed, we identified 315 patients with knee osteoarthritis and 607 with hip osteoarthritis: 922 in total (Figures 1, 2 and A1, Tables A2-A3).
During the time period for which data were collected, the musculoskeletal hub service was the only route for patients in Oxfordshire to access NHS elective orthopaedic surgery. When GPs refer patients to secondary care with knee/hip symptoms, the referral letter and x-rays are first reviewed by senior hub staff, who decide whether the patient should (a) be referred directly to surgical assessment, (b) attend the hub clinic (c) be managed in primary care (Figure 3). Triage is based on patient symptoms as described in the GP's referral letter (which will include BMI and may include OKS/OHS) and conservative treatments already given (such as anti-inflammatory drugs, physiotherapy and weight loss strategies).
Of the osteoarthritis patients included in the analysis, 23% (71/315) of knee patients and 39% (236/607) of hip patients were referred directly to surgical assessment based on the hub triage. In the authors' experience, these patients tend to be those with severe symptoms, those who had already exhausted all conservative measures and those who had previously been referred for surgical assessments, but had chosen not to have surgery at that time. Among 27 patients referred directly to surgical assessment for whom OKS data were available, the mean OKS was 15 (range: 1-30); no such data were available for hip patients.
At the surgical assessment visit, patients discuss the risks and benefits of surgery with an orthopaedic surgeon and make an informed decision about whether or not to undergo arthroplasty or other surgery, taking into account their comorbidities, symptom severity and other factors. Among the patients who were referred directly for surgical assessment and had outcomes data available, 56% (38/68) of knee patients and 69% (161/235) hip patients underwent or were awaiting knee/hip replacements at the time of data extraction.
Following triage, 36% (114/315) of knee patients and 44% (265/607) hip patients did not attend face-to-face consultations at either the hub or surgical assessment. Typically such patients comprise those who can be managed in primary care: for example those with mild symptoms and those who have not yet exhausted conservative treatment options, such as advice and information, activity and exercise and weight loss. [18][19][20] Across the 379 hip and  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   10 knee patients without face-to-face consultations, 270 (71%) had x-rays, MRI or ultrasound to identify the most appropriate care pathway. Eleven (3%) were referred for physiotherapy, while seven patients (2%), all with hip osteoarthritis, had injections. Other patients may not be seen face-to-face as they are unfit for surgery or have recent injuries likely to heal without further intervention. Eighteen patients opted for non-NHS care, missed/failed to book hub appointments, died before the hub attendance or were still awaiting a hub appointment at the time of data extraction (Figures 1 and 2).
The remaining 44% (130/315) of knee patients and 17% (106/607) of hip patients attended face-to-face assessments at the musculoskeletal hub, where extended-scope specialist physiotherapists or orthopaedic fellows assess patients to confirm diagnosis. Patients routinely complete OKS/OHS questionnaires to assess whether the level of symptoms warrants surgery and to guide discussions about the symptom profile. Among the patients attending the hub, the mean OKS was 21 (range: 1-48, Table A3, Figure A1), while mean OHS was 24 (range: 5-46); both values were significantly higher than the average for patients undergoing arthroplasty nationally (mean: 18; p≤0.015), although only OHS was significantly different from the average for patients undergoing arthroplasty in Oxfordshire (mean OKS: 20, p=0.134; mean OHS: 19 p<0.0001). [9][10][11][12][13][14] Diagnostic imaging, landmark injections and injections for trochanteric bursitis may be done during the hub consultation. Patients with BMI ≥40 are referred for monitored weight loss programs that must be followed for 12 months before considering surgery. Staff and patients also discuss the risks and benefits of joint replacement (including recovery times, the need for support at home after hospital discharge and the potential need for revision surgery) and how these may be affected by patients' living arrangements and comorbidities. Following the hub visit, 45% (49/130) of knee patients and 36% (36/101) of hip patients were referred for surgical assessment. This includes patients who are considered candidates for arthroplasty or other procedures (e.g. arthroscopy, anterior cruciate ligament repair or interventional radiology).
Patients with higher OKS/OHS were significantly less likely to be referred for surgical assessment: each one-point increase in OKS reduced the odds of referral by 4.7% (p=0.019), while a one-point increase in OHS reduced the odds by 3.9% (p=0.062; Table A4). Those patients who are not referred to surgical assessment may be referred for physiotherapy, other non-surgical management or other outpatient clinics (e.g. rheumatology or sports injury clinics), or may choose not to be referred as they prefer not to undergo surgery at the current time.
Across the hub attendees who were referred for surgical assessment and had data on clinical outcomes, 30% (12/40) of knee patients and 37% (10/27) of hip patients underwent or were awaiting arthroplasty surgery. Applying estimates of the probability of referral and subsequent arthroplasty from the hub dataset to national data on the distribution of patients undergoing knee/hip arthroplasty suggests that GPs in England refer around 417,000 patients with knee osteoarthritis and around 235,000 patients with hip osteoarthritis to secondary care each year (Figure 3). Of these, around 172,000 knee patients and 41,000 hip patients might attend a hub if the care pathway in Oxfordshire were followed nationally; such hub attendances would cost a total of £12 million.

Effect of introducing referral thresholds
We used the hub data to estimate the impact that different policies for using OKS/OHS to guide referral decisions might have on patient numbers, costs and health outcomes. The ACHE study reported that people with OKS ≤31 have a ≥70% chance of achieving a sevenpoint increase in OKS following knee arthroplasty. [4] Of the 110 knee patients attending the NOC hub, 94 (85%) had OKS ≤31, of whom 47 (50%) were referred for surgical assessment. Two patients were referred with OKS of 32 or 41. Assuming that 50% of the NOC patients with OKS ≤31 who are not currently referred might still choose not to be referred or might be considered clinically inappropriate for other reasons, a fixed referral threshold OKS of 31 would have resulted in an additional 22 referrals to surgical assessment: a 49% increase.
Applying the results of logistic regression to the whole population undergoing surgery nationally suggested that an OKS threshold of 31 would result in 43,000 additional surgical assessment visits and 13,000 additional knee replacement procedures in England each year (Table 1). Based on the costs and outcomes for patients with and without knee arthroplasty that were estimated in a recent economic evaluation, [5] introducing this policy would cost the NHS an additional £36 million for each annual cohort of patients (of which £5.8 million would be due to additional surgical assessment visits), but would gain 30,000 QALYs. This policy would be highly cost-effective compared with current practice, costing just £1,195 per QALY gained, and the net health benefit would be 28,000 QALYs even after subtracting the health benefits that would be foregone by spending £36 million on knee arthroplasty candidates, rather than other conditions. Introducing higher OKS thresholds would produce still greater health benefits and would continue to be cost-effective at a £20,000/QALY ceiling ratio right up to the economic threshold of 43. [5] Similarly, ACHE predicts that people with preoperative OHS ≤35 have a ≥70% chance of achieving an eight-point increase in OHS following hip arthroplasty. [4] Within the NOC hub, 87% (88/101) of patients had OHS ≤35, of whom 35 were referred. One patient was referred with OHS of 44. Introducing a threshold of 35 might therefore result in 26 additional referrals: a 30% increase. Extrapolating across England, we might expect 14,000 additional surgical assessments and 5,000 additional hip replacements each year. The impact of this policy is lower than for knee osteoarthritis as 90% of hip replacements are done in patients  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   12 who were referred directly, and did not attend the hub. The policy would cost an additional £25 million (of which additional surgical assessments account for £2 million) and gain 16,000 QALYs, costing £1,560/QALY gained and generating a net health benefit of 15,000 QALYs. Introducing higher thresholds (up to the economic threshold of 45) would produce greater health benefits and continue to cost <£20,000/QALY gained.

Discussion
Although referral hubs are being used increasingly as gatekeepers determining access to consultations in different clinical areas of the NHS, there has been little research on them. This retrospective analysis characterised the patients being referred to and from a musculoskeletal hub, and showed that OKS/OHS and other factors were considered when deciding which patients to refer for surgical assessment. Since most patients attending hubs have relatively severe osteoarthritis symptoms and those with high OKS/OHS are less likely to be referred for surgery, introducing evidence-based OKS/OHS thresholds for arthroplasty is unlikely to prevent large numbers of inappropriate referrals. However, since recent evidence demonstrates that arthroplasty is both cost-effective and highly likely to benefit people with OKS/OHS well above the thresholds currently used by some CCGs, [4 5] such policies are also likely to identify patients who are not currently referred, but for whom arthroplasty could be expected to be beneficial and cost-effective. These numbers are uncertain, but extrapolations from the hub examined here suggest that, across England, setting thresholds of OKS ≤31 and OHS ≤35 would result in many more patients being referred for surgical assessment, and could result in an additional 13,000 knee replacements and 4,500 hip replacements each year. More evidence is therefore urgently needed on the other factors currently influencing both the decision to refer from the hub to surgical assessment and the decision on whether arthroplasty is the appropriate treatment option.
To our knowledge our study is the first to report the characteristics of patients attending a musculoskeletal hub. However, the analysis was based on a small sample and included only eight hub attendees with OKS/OHS ≥40. Data were analysed one month after the end of the period studied; some patients may therefore have gone on to have arthroplasty after physiotherapy or weight loss treatment. Furthermore, the retrospective review of medical records may not have identified all patients meeting exclusion criteria -particularly for patients who did not attend either the hub or surgical assessment. Furthermore, no data were collected on BMI.
Estimates of the impact of different policies rely on additional assumptions and represent an approximate indication of potential patient numbers; prospective pilot studies would be required to assess the true impact in practice. In particular, we focused on the impact of changing referral guidelines at hubs, as no data were available on patients visiting GPs with osteoarthritis symptoms. In practice, any decision aid or referral guideline is also likely to affect GPs' referral decisions and the hub triage process, which could increase the number of additional operations resulting from any policy change. We assumed that decisions made by the patient and surgeon about whether to proceed to surgery after surgical assessment would be unaffected by the use of thresholds by the hub as OKS/OHS would have already been taken into account at the hub; the budget impact for patients referred via the hub could be lower than shown in Tables 1 and 2 if patients referred with high OKS/OHS were less likely to undergo surgery. We also assumed, arbitrarily, that 50% of people with OKS/OHS below the threshold who are not currently referred would choose not to be referred for surgery even if this were offered. The model also assumed that patients who do not undergo arthroplasty following their hub attendance would not have surgery for 10 years, since there are currently no data on how OKS/OHS change over time without arthroplasty. [5] In practice, many patients who would be eligible for surgery if the OKS/OHS threshold was raised would otherwise have had surgery later, after their condition had deteriorated. If the OKS/OHS thresholds were raised, the number of operations and budget impact might decrease over time as these patients would have been treated earlier, before their disease progresses, and would not need primary arthroplasty in the future.
Although access to joint replacement may be restricted if funding for interventions is reduced, results suggest that increased numbers of knee/hip replacements could be justified on cost-effectiveness grounds. Our results suggest that the NHS is currently willing to pay no more £2,000 per QALY gained from arthroplasty, while NICE routinely approves treatments for patients with other conditions (e.g. rheumatoid arthritis) that cost £20,000 [21] or even £40,000 [22]/QALY. This suggests that it may be more efficient and more equitable to spend limited NHS funds on conducting more joint replacements, increasing the operation rate in the UK to a level similar to that of Austria or Germany, [23] rather than directing these resources to other conditions. However, the number of procedures is also limited by availability of surgeons, operating theatres and hospital beds, which may mean that it is not currently feasible to conduct all of the operations that could be justified on cost-effectiveness grounds. And, as noted earlier, OKS/OHS is clearly not the only factor taken into account in decision-making by hubs and surgeons. In particular, patient choice plays a substantial role and comorbidities and other factors are also taken into account.
Referral hubs are an effective mechanism for referring patients to the most appropriate healthcare interventions/setting and patients are generally satisfied with this model of care. [8] One aim of introducing hubs was to improve efficiency in the referral pathway by increasing the proportion of referrals to secondary care that lead to surgery. The hub identifies patients who do not want surgery or who have mild symptoms and directs patients to other effective interventions when surgery is not appropriate. If the service works well, only patients who are suitable for surgery are referred. However, the process and criteria for arthroplasty referrals still vary greatly between CCGs. [ 14 and decision aids could help reduce geographical inequity and ensure that all patients likely to benefit have fair access to cost-effective treatments. Although our analysis is based on retrospective data from only one hub, it provides initial estimates of what potential patient numbers and costs might be if a more explicit threshold-based approach was adopted nationally. While arthroplasty rates and patient characteristics in Oxfordshire are similar to the national average, [24 25] referral guidelines and pathways vary substantially between CCGs, suggesting that a broader study covering multiple hubs is necessary to provide a more comprehensive picture of current clinical practice. Ideally this would also collect more detailed patient information (e.g. BMI or comorbidities).

1] Explicit guidelines
In conclusion, our paper has estimated the effect of introducing evidence-based OKS/OHS thresholds in England. While our results can be used to estimate the impact of other policies, including those where thresholds vary by age or other patient characteristics, a pilot study would be required to evaluate the actual impact of a policy change on routine clinical practice.

Declarations
Competing interest statement All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: a grant from NIHR for the submitted work; HD reports personal fees from Halyard Health, outside the submitted work; DB reports grants from Zimmer Biomet, outside the submitted work; AP reports personal fees from Zimmer Biomet and DePuy outside the submitted work; all declare that there are no other relationships or activities that could appear to have influenced the submitted work. The other authors declare that there are no relevant conflicts of interests besides the NIHR grant that supports the current work.

Details of contributors
AP is the guarantor for this study. AP, DB, HD, AG and KB conceived, designed and conducted the ACHE study. HD, AP and JS conceived and designed the study, which was conducted by HD and JS with assistance from Reza Mafi, and Sarah Dorman. HD, AG and PE conceived and designed the economic evaluation. HD conducted the statistical and economic analysis under the supervision of AG. AP and KB helped interpret the results. HD drafted the manuscript. All authors edited the manuscript for important intellectual content and approved the final version.

Copyright/licence
The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, an exclusive licence (or non-exclusive for government employees) on a worldwide basis to the BMJ Publishing Group Ltd and its Licensees to permit this article (if accepted) to be published in BMJ Open and any other BMJPGL products to exploit all subsidiary rights, as set out in our licence http://authors.bmj.com/submitting-yourpaper/copyright-and-authors-rights/ and the Corresponding Author accepts and understands that any supply made under these terms is made by BMJPGL to the Corresponding Author. Where the Corresponding Author wishes to make the article available on an open access basis (and intends to pay the relevant open access fee), the terms of such open access shall be governed by a Creative Commons licence -details of these licences and which Creative Commons licence will apply to this article are set out in our licence referred to above.

Ethics statement
Trust management approval from the Oxford University Hospitals NHS Trust was obtained for retrospective analysis of anonymised data from medical records concerning outcomes in hip and knee arthroplasty (Oxford University Hospitals Research and Development Reference 11603). The plan to study referral patterns through the musculoskeletal hub, under the terms of the research and development agreement noted above, was discussed and agreed through the Clinical Governance Group for hip and knee replacement. The analysis was conducted to improve service understanding and used retrospective data on patients who had received normal clinical management. All data were anonymised by clinical staff.

Funding and role of study sponsors
The ACHE study and economic evaluation was funded by the NIHR Health Technology Assessment Programme (project number 11/63/01) and will be published in full in Health Technology Assessment.
Further information available at: https://www.journalslibrary.nihr.ac.uk/programmes/hta/116301. This report presents independent research commissioned by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, MRC, CCF, NETSCC, the Health Technology Assessment Programme, NHS, NHS Digital or the Department of Health. The funder had no involvement in: the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication. The authors had full access to the data (including statistical reports and tables), can take responsibility for the integrity of the data and the accuracy of the data analysis and are responsible for submitting this paper.

Transparency declaration
The manuscript's guarantor (AP) affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

Data sharing
The decision tree and Markov models used in this paper are available from the corresponding author on request. PROMs/HES data are available from NHS Digital (http://content.digital.nhs.uk/dars). Woollacott. We would like to extend thanks to Reza Mafi, and Sarah Dorman who helped with data cleaning. We would also like to thank the reviewers for their insightful comments on the manuscript. Data on NHS PROMs linked to HES APC data were reused with the permission of NHS Digital, copyright 2015, with all rights reserved.

Patient involvement
The ACHE research question and study design (prior to funding) was informed by patient interest groups. The current analysis was based on pre-existing data from medical records so no patients were directly recruited in the course of the work. Throughout the ACHE study, a User Group, including patients and other stakeholders, informed progress and development of the work. Results of the analysis were discussed with user group including clinicians and patient representatives, to inform the presentation of results. The results will not be directly disseminated to patients whose records were included in the dataset, but the publication will be available for the target population.     Table A2 in online supplementary appendix for a list of the conditions other than osteoarthritis for which patients aged ≥50 years were excluded from the analysis.   Table A2 in online supplementary appendix for a list of the conditions other than osteoarthritis for which patients aged ≥50 years were excluded from the analysis.

Online supplementary file Methods for stepwise regression models
We selected variables for inclusion in the logistic regression model in a pre-specified sequence. We first estimated a model only including the constant term and evaluated whether adding OKS/OHS to the model reduced Akaike information criterion (AIC); if AIC was reduced, the model used in the next step controlled for OKS/OHS and otherwise remained with just a constant term. Similarly, the next step assessed whether adding age into the model reduced AIC and in the final step we assessed whether adding sex into the model reduced AIC. We specified the order in which variables were considered for the model (OKS/OHS, then age, then sex) in advance of data analysis: it was hypothesised that OKS/OHS would be the most important factor determining referrals, followed by age (since this is a proxy for comorbidities) and finally sex (since we had no prior hypotheses about how sex would affect referrals). Age and OKS/OHS were assumed to have linear effects on the log-odds of being referred for surgical assessment as there were insufficient data to evaluate nonlinear functions.

Methods for estimating patient numbers
Hub data were extrapolated to estimate patient numbers across England by applying the probability of referral estimated on the hub data to national data on the characteristics of patients undergoing arthroplasty and the total number of operations. We then calculated the number of patients in each group who are likely to have been referred by GPs with knee/hip symptoms by dividing the number of operations in each group by the probability that each group of patients will undergo surgery.

Unit costs
The unit costs used in the current analysis are shown in Table A1. The costs and QALYs accrued in the 10-year period after arthroplasty (or a decision not to perform arthroplasty) have been reported previously. [4]   Additional regression results For patients attending the hub with knee symptoms, adding age and sex into the model predicting the odds of hub attendees being referred for surgical assessment increased AIC, so the final model included only OKS (Table A4). Similarly, in the model of hip patients, adding age into the model worsened AIC, so the final model included only OHS and gender.
Omitting gender from the model had negligible effect on the coefficient for OHS, which was 0.0397 (SE: 0.0211) when OHS was the only explanatory variable. In addition to the logistic regression model predicting the odds of patients being referred from the hub to surgical assessment, we also conducted a secondary analysis predicting the odds of undergoing arthroplasty in patients who had already been referred to surgical assessment by the hub, using the same methods. This analysis suggested that OKS/OHS, age and sex had no significant effect on the odds of referral from surgical assessment to surgery (p>0.05) and that including any of these variables in the regression increased AIC compared with a model with only a constant term. This validates the assumption that the probability of patients referred directly to surgical assessment subsequently being referred for surgery is independent of OKS/OHS, age and sex and the assumption that introducing OKS/OHS thresholds at the hub would have no effect on the probability of a patient with particular preoperative characteristics being referred for arthroplasty following a surgical assessment visit. However, this analysis is likely to be underpowered, since only 12 knee patients and 10 hip patients underwent surgery after face-to-face consultations at the hub and surgical assessment.    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

Strengths and limitations of this study
 This is the first analysis reporting data on the characteristics of patients being referred to a UK hub with symptoms of hip or knee osteoarthritis; such data are needed to assess the impact of changing referral criteria on the number of knee/hip replacements and on costs and quality of life.  We retrospectively reviewed clinical records for 922 men and women referred to a single musculoskeletal hub, although only 221 patients underwent arthroplasty.  We used several assumptions to extrapolate our estimates of how the probability of referral varies with preoperative characteristics using a decision-tree and Markov model, which enabled us to estimate the potential impact of different referral policies locally and nationally.  A prospective pilot study would be required to quantify the real-world impact of any policy change, while data from multiple hubs and information on other patient characteristics (such as body mass index) are required for a comprehensive understanding of current clinical practice.  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  Introduction Many clinical commissioning groups (CCGs) in the UK have set referral criteria for joint replacement that exclude patients with Oxford knee and hip scores (OKS and OHS) above certain thresholds, which are often as low as 19 [1] or 24 [1][2][3] out of 48. Recent papers have shown that these low thresholds are inappropriate: arthroplasty has a ≥80% chance of producing a meaningful improvement in patients with OKS/OHS of 19 or 24 [4] and would cost <£10,000 per quality adjusted life-year (QALY). [5] The recent ACHE study analysed >400,000 medical records and concluded that referral thresholds between 31 and 41 may be justified on clinical and cost-effectiveness grounds. [4 5] Commissioners' decisions concerning thresholds may be influenced by the likely number of operations and their cost. In the UK, >94% of patients currently undergoing arthroplasty have OKS/OHS ≤31, demonstrating that nearly all operations currently conducted are costeffective and have ≥70% chance of meaningful improvement. [4 5] However, there are very little data on how patient characteristics influence the assessment process in practice, or on patients who do not currently receive surgery, making it difficult to assess the impact that different referral criteria may have on patient numbers or costs. As in many other clinical areas, hubs (also known as triage clinics, musculoskeletal interface services or intermediate musculoskeletal assessment centres) are increasingly used as gatekeepers determining access to consultations with orthopaedic surgeons, assessing all patients being considered for surgery. [6][7][8] We aimed to review the characteristics of patients attending hub consultations, estimate the proportion of patients referred for surgical assessment and surgery and explore how referral decisions vary with preoperative characteristics. To illustrate how such data could be used to inform policy, we estimated the potential impact of a change in referral criteria: namely basing referrals from the hub to surgical assessment on evidence-based OKS/OHS thresholds, rather than the hub's current referral criteria.

Outline
We collected data from one UK referral hub and used it to estimate a logistic regression model predicting the probability that patients will be referred from the hub to surgical assessment based on their OKS/OHS, age and sex. We then subsequently applied this regression model to nationally-collected preoperative outcomes data [9][10][11][12][13][14] to estimate how many patients are referred for surgical assessment nationally. We predicted how many of these patients would be referred through to secondary care if different OKS/OHS thresholds were introduced at the hub and potentially how many extra operations would be performed.

Hub data collection
Anonymised data were extracted retrospectively from medical records of patients with knee or hip symptoms who had been referred by GPs between July 2015 and July 2016 to the musculoskeletal hub at the Nuffield Orthopaedic Centre (NOC) in Oxford. This study was discussed and agreed through the Clinical Governance Group for hip and knee replacement as part of a wider review of outcomes in hip and knee arthroplasty (Oxford University Hospitals Research and Development Ref 11603). The primary statistical analysis used a prognostic model to estimate how the probability of patients attending face-to-face consultations at the hub being referred to surgical assessments in secondary care varied with OKS/OHS, age and sex. We therefore chose the sample size to provide data on ≥30 knee referrals and ≥30 hip referrals from the hub to secondary care, thereby providing ≥10 events per explanatory variable. [15] Two medically-qualified surgical research fellows extracted the following: age; sex; OKS/OHS at hub attendance; attendance date; whether the patient was referred to secondary care; date of any subsequent surgical assessment visit; and date of any subsequent arthroplasty surgery. Additional information on imaging, referrals to other clinics, and other surgeries or diagnoses was recorded in free text fields. Data on OKS at the surgical assessment visit were also extracted, when available, from the secondary care records of patients with knee pain. Body mass index (BMI) was not extracted as the analysis focused on OKS/OHS and the available evidence on how capacity to benefit and cost-effectiveness vary with OKS/OHS does not consider BMI. [4 5] Patients who were on the waiting list for arthroplasty surgery at the time of data extraction (August 2016) and those for whom surgery was delayed due to comorbidities or high BMI after their attendance at the surgical assessment consultation were counted as having been referred for surgery.
Exclusion criteria comprised:  Aged <50 years (for whom knee/hip pain is unlikely to be caused by osteoarthritis).  Evidence from medical records that symptoms were due to a condition other than osteoarthritis.  Previous arthroplasty on the same joint.  Medical records inaccessible for research.  Any attendance at the hub or surgical assessment unit before July 2015.
However, patients who were referred for x-rays, magnetic resonance imaging (MRI) or physiotherapy but did not attend face-to-face consultations at the hub or surgical assessment unit were included in the descriptive analysis.

Statistical modelling
We used logistic regression to estimate a prognostic model of the local hub data that predicted how OKS/OHS, age and sex affected the odds of being referred for surgical assessment following a face-to-face attendance at the hub. Age and OKS/OHS at hub attendance were analysed as continuous variables. Explanatory variables were selected based on the Akaike information criterion using forward stepwise regression, manually testing variables in a pre-specified sequence (online supplementary appendix). Analyses were conducted in Stata version 14 (StataCorp, College Station TX) on a complete case basis, excluding patients with missing data. We analysed binary variables using two-sample tests of proportions and analysed continuous variables using unpaired t-tests preceded by F-tests for equal variance. OKS/OHS in the hub sample was compared against population means [9][10][11][12][13][14] using one-sample t-tests.

Estimating the number of referrals and effect of introducing thresholds
A decision-tree model of the treatment pathway was developed based on hub data and the authors' clinical experience of running hubs and/or surgical clinics. The overall proportion of patients referred directly to surgical assessment and the proportion attending face-to-face hub consultations were calculated from the hub sample. We also estimated the proportion of patients who underwent arthroplasty after being referred from the hub to surgical assessment.
The model then extrapolated hub data to estimate the number of patients in England who are referred for hip or knee replacement (online supplementary appendix). Logistic regression results were used to predict the probability of being referred from the hub to surgical assessment for men and women aged 50, 60, 70, 80 and 90 with different OKS/OHS. For each patient group defined by age, sex and OKS/OHS, we then estimated the number of patients referred to hubs nationally by dividing national data on the number of joint replacements currently conducted for osteoarthritis in England [9][10][11][12][13][14] by the probability that patients in each group would undergo surgery.
We then conducted an exploratory analysis estimating how the number of referrals to surgical assessment and the number of arthroplasty procedures might change if different OKS/OHS thresholds between 18 and 45 were used to determine referral decisions during face-to-face hub consultations. Although in practice thresholds could be used at various stages in the referral pathway, our analysis focused on the impact of changing referral criteria during face-to-face hub consultations since OKS/OHS data were only routinely available for this patient group.
This analysis made the following assumptions: Introducing OKS/OHS thresholds at the hub was assumed to have no effect on GP referrals, the hub triage or the probability of a patient with particular preoperative characteristics being referred for arthroplasty following a surgical assessment visit. 2) Since patient records provided no data on OKS/OHS for patients who did not attend the hub, we assumed that the probability of direct referral to surgical assessment and the probability of attending a face-to-face visit at the hub were independent of OKS/OHS, age and sex. Although this assumption may not hold in practice (as symptom severity is one of the main factors considered in the hub triage), it is unlikely to affect estimates of the impact of changing referral criteria solely at the hub. 3) We also assumed that the probability of patients referred to surgical assessment subsequently being referred for surgery was independent of OKS/OHS, age and sex, which was supported by a secondary regression analysis (online supplementary material). 4) We assumed that the referral probabilities estimated from the NOC hub sample are broadly representative of clinical practice across the UK. 5) We assumed, based on the experience of clinical co-authors, that 50% of patients who are currently not referred for surgical assessment after a face-to-face hub visit would still not be considered candidates for arthroplasty (and therefore not referred) regardless of whether thresholds were introduced. This group of patients includes patients who choose not to be referred, those who have not had a complete trial of nonoperative treatment and others who are unsuitable for surgery due to comorbidities or other factors. The remaining 50% of patients were assumed to be referred only if their OKS/OHS was below the threshold. 6) Since there are no published data on how OKS/OHS changes over time in the absence of arthroplasty, we assumed for simplicity patients who do not undergo arthroplasty following their hub attendance would not be referred back to clinic for 10 years. 7) Each 40-minute hub attendance was estimated to cost £58 and surgical assessment £132 (online supplementary appendix, Table A1). 8) The analyses took the perspective of the NHS, focusing on costs related to knee/hip arthroplasty. Costs of GP consultations, hub triage, x-rays, imaging, physiotherapy, injections, weight loss programs, missed appointments and referrals to other clinics were excluded as there is no reason to expect the proportion of patients requiring these services to change following the introduction of OKS/OHS thresholds. 9) The reference year for costs was 2014.
The patient numbers calculated within the decision-tree were used to estimate the cost of the referral pathway. The total 10-year costs and total QALYs estimated in a related study [5] for different patient characteristics with and without arthroplasty were applied to the number of patients expected to undergo arthroplasty or have no arthroplasty in each scenario. These figures were used to estimate the net health benefit for each scenario [16 17] (assuming that the NHS is willing/able to pay £20,000 per QALY gained [18]) and the We also estimated the number of additional referrals for surgical assessment that might have been observed within the population of individuals attending the NOC hub if a fixed OKS/OHS threshold were introduced. Based on assumption (5), this was equal to half the number of patients who had OKS/OHS at or below the proposed threshold but were not referred, plus the number who were referred with OKS/OHS below the proposed threshold: for example, if (hypothetically) 60 out of 100 patients were referred and 55 of those referred and 24 of those not referred had OKS<31, we would estimate that 67 (55+24/2) would be referred if a threshold of 31 were introduced.

Public and patient involvement
Patient representatives were involved in the grant application and design of the wider ACHE study. [4] The ACHE research question and study design (prior to funding) was informed by patient interest groups. The current analysis was based on pre-existing data from medical records so no patients were directly recruited in the course of the work. Throughout the ACHE study, a User Group, including patients and other stakeholders, informed progress and development of the work. Results of the analysis were discussed with user group including clinicians and patient representatives, to inform the presentation of results.

Current referral pathway
From the 1,638 records reviewed, we identified 315 patients with knee osteoarthritis and 607 with hip osteoarthritis: 922 in total (Figures 1, 2 and A1, Tables A2-A3). Data on these patients were used to construct the treatment pathway shown in Figure 3.
During the time period for which data were collected, the musculoskeletal hub service was the only route for patients in Oxfordshire to access NHS elective orthopaedic surgery and also directed patients to physiotherapy in the absence of a direct referral physiotherapy service. When GPs refer patients to secondary care with knee/hip symptoms, the referral letter and x-rays are first reviewed by senior hub staff, who decide whether the patient should (a) be referred directly to surgical assessment, (b) attend the hub clinic, or (c) be managed in primary care (Figure 3). Triage is based on the symptoms described in the GP's referral letter (which will include BMI and may include OKS/OHS) and the conservative treatments already given (such as anti-inflammatory drugs, physiotherapy and weight loss strategies). Of the osteoarthritis patients included in the analysis, 23% (71/315) of knee patients and 39% (236/607) of hip patients were referred directly to surgical assessment based on the hub triage. In the authors' experience, these patients tend to be those with severe symptoms, those who had already exhausted all conservative measures and those who had previously been referred for surgical assessments, but had chosen not to have surgery at that time. Among 27 patients referred directly to surgical assessment for whom OKS data were available, the mean OKS was 15 (range: 1-30); no such data were available for hip patients.
At the surgical assessment visit, patients discuss the risks and benefits of surgery with an orthopaedic surgeon and make an informed decision about whether or not to undergo arthroplasty or other surgery, taking into account their comorbidities, symptom severity and other factors. Those not undergoing arthroplasty may have interventional radiology, or other operations. Among the patients who were referred directly for surgical assessment and had outcomes data available, 56% (38/68) of knee patients and 69% (161/235) hip patients underwent or were awaiting knee/hip replacements at the time of data extraction.
Following triage, 36% (114/315) of knee patients and 44% (265/607) hip patients did not attend face-to-face consultations at either the hub or surgical assessment. Typically such patients comprise those who can be managed in primary care: for example those with mild symptoms and those who have not yet exhausted conservative treatment options, such as advice and information, activity and exercise and weight loss. Across the 379 hip and knee patients without face-to-face consultations, 270 (71%) had x-rays, MRI or ultrasound to identify the most appropriate care pathway. Eleven (3%) were referred for physiotherapy, while seven patients (2%), all with hip osteoarthritis, had injections. Other patients may not be seen face-to-face as they are unfit for surgery or have recent injuries likely to heal without further intervention. Eighteen patients opted for non-NHS care, missed/failed to book hub appointments, died before the hub attendance or were still awaiting a hub appointment at the time of data extraction (Figures 1 and 2).
The remaining 41% (130/315) of knee patients and 17% (106/607) of hip patients attended face-to-face assessments at the musculoskeletal hub, where extended-scope specialist physiotherapists or orthopaedic fellows assess patients to confirm diagnosis. Patients routinely complete OKS/OHS questionnaires to assess whether the level of symptoms warrants surgery and to guide discussions about the symptom profile. Among the patients attending the hub, the mean OKS was 21 (range: 1-48, Table A3, Figure A1), while mean OHS was 24 (range: 5-46); both values were significantly higher than the average for patients undergoing arthroplasty nationally (mean: 18; p≤0.015), although only OHS was significantly different from the average for patients undergoing arthroplasty in Oxfordshire (mean OKS: 20, p=0.134; mean OHS: 19 p<0.0001). [9][10][11][12][13][14]  Diagnostic imaging, landmark injections and injections for trochanteric bursitis may be done during the hub consultation. Patients with BMI ≥40 are referred for monitored weight loss programs that must be followed for 12 months before considering surgery. Staff and patients also discuss the risks and benefits of joint replacement (including recovery times, the need for support at home after hospital discharge and the potential need for revision surgery) and how these may be affected by patients' living arrangements and comorbidities. Following the hub visit, 45% (49/130) of knee patients and 36% (36/101) of hip patients were referred for surgical assessment. This includes patients who are considered candidates for arthroplasty or other procedures (e.g. arthroscopy, anterior cruciate ligament repair or interventional radiology).
Patients with higher OKS/OHS were significantly less likely to be referred for surgical assessment: each one-point increase in OKS reduced the odds of referral by 4.7% (p=0.019), while a one-point increase in OHS reduced the odds by 3.9% (p=0.062; Table A4). Those patients who are not referred to surgical assessment may be referred for physiotherapy, other non-surgical management or other outpatient clinics (e.g. rheumatology or sports injury clinics), or may choose not to be referred as they prefer not to undergo surgery at the current time.
Across the hub attendees who were referred for surgical assessment and had data on clinical outcomes, 30% (12/40) of knee patients and 37% (10/27) of hip patients underwent or were awaiting arthroplasty surgery.
Applying estimates of the probability of referral and subsequent arthroplasty from the hub dataset to national data on the distribution of patients undergoing knee/hip arthroplasty suggests that GPs in England refer around 417,000 patients with knee osteoarthritis and around 235,000 patients with hip osteoarthritis to secondary care each year ( Figure 3). Of these, around 172,000 knee patients and 41,000 hip patients might attend a hub if the care pathway in Oxfordshire were followed nationally; such hub attendances would cost a total of £12 million.

Knees
We used the hub data to estimate the impact that using OKS to guide referral decisions during face-to-face hub consultations might have on patient numbers, costs and health outcomes. Table 1 shows results for thresholds between 18 and 43, although we focus here on the impact of an OKS threshold of 31, since the ACHE study reported that people with pre-operative OKS ≤31 have a ≥70% chance of achieving a seven-point increase in OKS following knee arthroplasty. [4]  Of the 110 knee patients attending the NOC hub, 94 (85%) had OKS ≤31, of whom 47 (50%) were referred for surgical assessment. Two patients were referred with OKS of 32 or 41. Assuming that 50% of the NOC patients with OKS ≤31 who are not currently referred might still choose not to be referred or might be considered clinically inappropriate for other reasons, a fixed referral threshold OKS of 31 would have resulted in an additional 22 referrals to surgical assessment: a 49% increase.
Applying the results of logistic regression to the whole population undergoing surgery nationally suggested that an OKS threshold of 31 might result in 43,000 additional surgical assessment visits and 13,000 additional knee replacement procedures in England each year ( Table 1). Introducing this policy could cost the NHS an additional £36 million for each annual cohort of patients (of which £5.8 million would be due to additional surgical assessments), but would gain 30,000 QALYs. This policy would be highly cost-effective compared with current practice, costing just £1,195 per QALY gained. Introducing OKS thresholds between 32 and 43 would produce still greater health benefits and cost less than £20,000/QALY gained.

Hips
Similarly, the ACHE study found that people with preoperative OHS ≤35 have a ≥70% chance of achieving an eight-point increase in OHS following hip arthroplasty. [4] Within the NOC hub, 87% (88/101) of patients had OHS ≤35, of whom 35 were referred. One patient was referred with OHS of 44. Introducing a threshold of 35 might therefore result in 26 additional referrals: a 30% increase.
Extrapolating across England, we might expect 14,000 additional surgical assessments and 5,000 additional hip replacements each year if a threshold of 35 were introduced at the hub ( Table 2). The impact of this policy is lower than for knee osteoarthritis as 90% of hip replacements are done in patients who were referred directly, and did not attend the hub. The policy could cost an additional £25 million (of which additional surgical assessments account for £2 million) and gain 16,000 QALYs, costing £1,560/QALY gained. Introducing thresholds between 36 and 45 would produce greater health benefits and cost <£20,000/QALY gained.

Discussion
This retrospective analysis characterised the patients being referred to and from a musculoskeletal hub, and showed that OKS/OHS and other factors were considered when deciding which patients to refer for surgical assessment. Since most patients attending hubs have relatively severe osteoarthritis symptoms and few patients with high OKS/OHS undergo arthroplasty (partly due to the low OKS/OHS thresholds used by some CCGs), introducing evidence-based OKS/OHS thresholds for arthroplasty is unlikely to prevent large numbers of inappropriate referrals. However, since recent evidence demonstrates that arthroplasty is both cost-effective and highly likely to benefit people with OKS/OHS well above the thresholds currently used by some CCGs, [4 5] such policies are also likely to identify patients who are not currently referred, but for whom arthroplasty could be expected to be beneficial and cost-effective. Extrapolating from the hub data suggests that setting thresholds of OKS ≤31 and OHS ≤35 would result in many more patients being referred for surgical assessment and surgery, although the exact patient numbers are uncertain and rely on assumptions. More evidence is therefore urgently needed on the other factors currently influencing both the decision to refer from the hub to surgical assessment and the decision on whether arthroplasty is the appropriate treatment option.
To our knowledge, our study is the first to report the characteristics of patients attending a musculoskeletal hub. However, the analysis was based on a small sample from only one hub and included only eight hub attendees with OKS/OHS ≥40. The proportion of patients undergoing surgery after surgical assessment is uncertain as only 22 hub attendees underwent arthroplasty; however, varying the probability of undergoing surgery over its 95% confidence interval did not materially change our conclusions. Data were analysed one month after the end of the one-year period studied; some patients may therefore have gone on to have arthroplasty after physiotherapy or weight loss treatment after our data were extracted. Furthermore, the retrospective review of medical records may not have identified all patients meeting exclusion criteria -particularly for patients who did not attend either the hub or surgical assessment. Furthermore, no data were collected on BMI.
Estimates of the impact of different policies rely on additional assumptions and represent an approximate indication of potential patient numbers; prospective pilot studies would be required to assess the true impact in practice. In particular, we focused on the impact of changing referral guidelines at hubs, as no data were available on patients visiting GPs with osteoarthritis symptoms. In practice, any decision aid or referral guideline is also likely to affect GPs' referral decisions and the hub triage process, which could increase the number of additional operations resulting from any policy change. We assumed that decisions made by the patient and surgeon about whether to proceed to surgery after surgical assessment would be unaffected by the use of thresholds by the hub as OKS/OHS would have already been taken into account at the hub; the budget impact for patients referred via the hub could be lower than shown in Tables 1 and 2 if patients referred with high OKS/OHS were less likely to undergo surgery. We also assumed, arbitrarily, that 50% of people with OKS/OHS below the threshold who are not currently referred would choose not to be referred for surgery even if this were offered. The model also assumed that patients who do not undergo arthroplasty following their hub attendance would not have surgery for 10 years, since there are currently no data on how OKS/OHS change over time without arthroplasty. [5] In practice, many patients who would be eligible for surgery if the OKS/OHS threshold was raised would otherwise have had surgery later, after their condition had deteriorated. If the OKS/OHS thresholds were raised, the number of operations and budget impact might decrease over time as these patients would have been treated earlier, before their disease progresses, and would not need primary arthroplasty in the future.
Although access to joint replacement may be restricted if funding for interventions is reduced, results suggest that increased numbers of knee/hip replacements could be justified on cost-effectiveness grounds. Our results suggest that the NHS is currently willing to pay no more £2,000 per QALY gained from arthroplasty, while NICE routinely approves treatments for patients with other conditions (e.g. rheumatoid arthritis) that cost £20,000 [18] or even £40,000 [19]/QALY. This suggests that it may be more efficient and more equitable to spend limited NHS funds on conducting more joint replacements, increasing the operation rate in the UK to a level similar to that of Austria or Germany, [20] rather than directing these resources to other conditions. However, the number of procedures is also limited by availability of surgeons, operating theatres and hospital beds, which may mean that it is not currently feasible to conduct all of the operations that could be justified on cost-effectiveness grounds. And, as noted earlier, OKS/OHS is clearly not the only factor taken into account in decision-making by hubs and surgeons. In particular, patient choice plays a substantial role and comorbidities and other factors are also taken into account.
Referral hubs have previously been shown to be an effective mechanism for referring patients to the most appropriate healthcare interventions/setting and patients are generally satisfied with this model of care. [8] One aim of introducing hubs was to identify patients who do not want surgery or who have mild symptoms and direct them to other effective interventions if surgery is not appropriate. However, the process and criteria for arthroplasty referrals still vary greatly between CCGs. [1] Explicit guidelines and decision aids could help reduce geographical inequity and ensure that all patients likely to benefit have fair access to cost-effective treatments. Although our analysis is based on retrospective data on a small sample from only one hub, it provides initial estimates of what the potential patient numbers and costs might be if a more explicit threshold-based approach was adopted nationally. While arthroplasty rates and patient characteristics in Oxfordshire are similar to the national average, [21 22] referral guidelines and pathways vary substantially between CCGs, suggesting that a larger study covering multiple hubs is necessary to provide a more comprehensive picture of current clinical practice. Ideally this would also collect more detailed patient information (e.g. BMI or comorbidities).

Declarations
Competing interest statement All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: a grant from NIHR for the submitted work; HD reports personal fees from Halyard Health, outside the submitted work; DB reports grants from Zimmer Biomet, outside the submitted work; AP reports personal fees from Zimmer Biomet and DePuy outside the submitted work; all declare that there are no other relationships or activities that could appear to have influenced the submitted work. The other authors declare that there are no relevant conflicts of interests or funding besides the NIHR grant that supports the current work.

Details of contributors
AP is the guarantor for this study. AP, DB, HD, AG and KB conceived, designed and conducted the ACHE study. HD, AP and JS conceived and designed the study, which was conducted by HD and JS. HD, AG and PE conceived and designed the economic evaluation. HD conducted the statistical and economic analysis under the supervision of AG. AP, DB and KB helped interpret the results. HD drafted the manuscript. All authors edited the manuscript for important intellectual content and approved the final version.

Copyright/licence
The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, an exclusive licence (or non-exclusive for government employees) on a worldwide basis to the BMJ Publishing Group Ltd and its Licensees to permit this article (if accepted) to be published in BMJ Open and any other BMJPGL products to exploit all subsidiary rights, as set out in our licence http://authors.bmj.com/submitting-yourpaper/copyright-and-authors-rights/ and the Corresponding Author accepts and understands that any supply made under these terms is made by BMJPGL to the Corresponding Author. Where the Corresponding Author wishes to make the article available on an open access basis (and intends to pay the relevant open access fee), the terms of such open access shall be governed by a Creative Commons licence -details of these licences and which Creative Commons licence will apply to this article are set out in our licence referred to above.

Ethics statement
Trust management approval from the Oxford University Hospitals NHS Trust was obtained for retrospective analysis of anonymised data from medical records concerning outcomes in hip and knee arthroplasty (Oxford University Hospitals Research and Development Reference 11603). The plan to study referral patterns through the musculoskeletal hub, under the terms of the research and development agreement noted above, was discussed and agreed through the Clinical Governance Group for hip and knee replacement. The  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 analysis was conducted to improve service understanding and used retrospective data on patients who had received normal clinical management. All data were anonymised by clinical staff.

Funding and role of study sponsors
The ACHE study and economic evaluation was funded by the NIHR Health Technology Assessment Programme (project number 11/63/01) and will be published in full in Health Technology Assessment. Further information available at: https://www.journalslibrary.nihr.ac.uk/programmes/hta/116301. This report presents independent research commissioned by the National Institute for Health Research (NIHR). AG was supported by the NIHR Biomedical Research Centre, Oxford. The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, MRC, CCF, NETSCC, the Health Technology Assessment Programme, NHS, NHS Digital or the Department of Health. The funder had no involvement in: the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication. The authors had full access to the data (including statistical reports and tables), can take responsibility for the integrity of the data and the accuracy of the data analysis and are responsible for submitting this paper.

Transparency declaration
The manuscript's guarantor (AP) affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

Methods for stepwise regression models
We selected variables for inclusion in the logistic regression model in a pre-specified sequence. We first estimated a model only including the constant term and evaluated whether adding OKS/OHS to the model reduced Akaike information criterion (AIC); if AIC was reduced, the model used in the next step controlled for OKS/OHS and otherwise remained with just a constant term. Similarly, the next step assessed whether adding age into the model reduced AIC and in the final step we assessed whether adding sex into the model reduced AIC. We specified the order in which variables were considered for the model (OKS/OHS, then age, then sex) in advance of data analysis: it was hypothesised that OKS/OHS would be the most important factor determining referrals, followed by age (since this is a proxy for comorbidities) and finally sex (since we had no prior hypotheses about how sex would affect referrals). Age and OKS/OHS were assumed to have linear effects on the log-odds of being referred for surgical assessment as there were insufficient data to evaluate nonlinear functions.

Methods for estimating patient numbers
Hub data were extrapolated to estimate patient numbers across England by applying the probability of referral estimated on the hub data to national data on the characteristics of patients undergoing arthroplasty and the total number of operations.
We considered 490 groups of patients, defined by gender, age (using age bands <55, 55-64, 65-74, 75-84 and ≥85) and individual preoperative OKS/OHS between 0 and 48. An extract of individual patient data from the national patient reported outcome measures (PROMs) dataset linked to Hospital Episode Statistics (HES) covering operations conducted between 2009 and 2015 was used to estimate the proportion of arthroplasty patients in different groups defined by age and gender. The same dataset was used to obtain the proportion of patients with each preoperative OKS/OHS. We then calculated the number of patients in each group who are likely to have been referred by GPs with knee/hip symptoms by dividing the number of operations in each group by the probability that each group of patients will undergo surgery.

Unit costs
The unit costs used in the current analysis are shown in Table A1. The costs and QALYs accrued in the 10-year period after arthroplasty (or a decision not to perform arthroplasty) have been reported previously. [4]  Based on 40 minutes' of patient-related activity for a grade 8a physiotherapist. Based on activity at the NOC hub, each hub attendance lasts 30 minutes, plus an additional 10 minutes' administration per patient and will, on average, be conducted by a grade 8a physiotherapist who spends around 80% of their time on patient-related activity. The salary for a band 8a, point 36 physiotherapist was based on Agenda for Change salaries [5] and converted to an hourly rate by applying the same methodology as was applied by PSSRU for a hospital physiotherapist, [6] assuming that working hours and the percentages for salary on costs, qualifications and overheads would be the same for all hospital physiotherapists. Outpatient consultation with an orthopaedic surgeon

Additional regression results
For patients attending the hub with knee symptoms, adding age and sex into the model predicting the odds of hub attendees being referred for surgical assessment increased AIC, so the final model included only OKS (Table A4). Similarly, in the model of hip patients, adding age into the model worsened AIC, so the final model included only OHS and gender.
Omitting gender from the model had negligible effect on the coefficient for OHS, which was 0.0397 (SE: 0.0211) when OHS was the only explanatory variable. In addition to the logistic regression model predicting the odds of patients being referred from the hub to surgical assessment, we also conducted a secondary analysis predicting the odds of undergoing arthroplasty in patients who had already been referred to surgical assessment by the hub, using the same methods. This analysis suggested that OKS/OHS, age and sex had no significant effect on the odds of referral from surgical assessment to surgery  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 6 (p>0.05) and that including any of these variables in the regression increased AIC compared with a model with only a constant term. This validates the assumption that the probability of patients referred directly to surgical assessment subsequently being referred for surgery is independent of OKS/OHS, age and sex and the assumption that introducing OKS/OHS thresholds at the hub would have no effect on the probability of a patient with particular preoperative characteristics being referred for arthroplasty following a surgical assessment visit. However, this analysis is likely to be underpowered, since only 12 knee patients and 10 hip patients underwent surgery after face-to-face consultations at the hub and surgical assessment. 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 Introduction pg 5