Article Text
PDF
Abstract
Objectives This study aims to investigate the cost-effectiveness of individually tailored self-management support, delivered via the artificial intelligence-based selfBACK app, as an add-on to usual care for people with low back pain (LBP).
Design Secondary health-economic analysis of the selfBACK randomised controlled trial (RCT) with a 9-month follow-up conducted from a Danish national healthcare perspective (primary scenario) and a societal perspective limited to long-term productivity in the form of long-term absenteeism (secondary scenario).
Setting Primary care and an outpatient spine clinic in Denmark.
Participants A subset of Danish participants in the selfBACK RCT, including 297 adults with LBP randomised to the intervention (n=148) or the control group (n=149).
Interventions App-delivered evidence-based, individually tailored self-management support as an add-on to usual care compared with usual care alone among people with LBP.
Outcome measures Costs of healthcare usage and productivity loss, quality-adjusted life-years (QALYs) based on the EuroQol-5L Dimension Questionnaire, meaningful changes in LBP-related disability measured by the Roland-Morris Disability Questionnaire (RMDQ) and the Pain Self-Efficacy Questionnaire (PSEQ), costs (healthcare and productivity loss measured in Euro) and incremental cost-effectiveness ratios (ICERs).
Results The incremental costs were higher for the selfBACK intervention (mean difference €230 (95% CI −136 to 595)), where ICERs showed an increase in costs of €7336 per QALY gained in the intervention group, and €1302 and €1634 for an additional person with minimal important change on the PSEQ and RMDQ score, respectively. At a cost-effectiveness threshold value of €23250, the selfBACK intervention has a 98% probability of being cost-effective. Analysis of productivity loss was very sensitive, which creates uncertainty about the results from a societal perspective limited to long-term productivity.
Conclusions From a healthcare perspective, the selfBACK intervention is likely to represent a cost-effective treatment for people with LBP. However, including productivity loss introduces uncertainty to the results.
Trial registration number NCT03798288.
- Self-Management
- Artificial Intelligence
- Randomized Controlled Trial
- Primary Health Care
- HEALTH ECONOMICS
- Back pain
Data availability statement
Data are available on reasonable request. Requests to access RCT data supporting this study's findings can be made to MJS (m.jensen@kiroviden.sdu.dk) at the Department of Sports Science and Clinical Biomechanics, University of Southern Denmark, Odense, Denmark and Chiropractic Knowledge Hub, University of Southern Denmark, Odense, Denmark. The cost data are available pending permission from Statistics Denmark’s Research Service.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
- Self-Management
- Artificial Intelligence
- Randomized Controlled Trial
- Primary Health Care
- HEALTH ECONOMICS
- Back pain
STRENGTHS AND LIMITATIONS OF THIS STUDY
A key strength is the ability to link trial data with national registries, which had no missing data on costs and no need for imputation, increasing the study’s internal validity.
A limitation is that this is a secondary analysis from a trial focusing on reducing low back pain-related disability through individually tailored self-management support, not specifically designed for a health-economic evaluation.
The analysis concerning productivity loss is limited to all-cause, long-term sickness absence and a few persons. Thus, these results should be interpreted cautiously.
However, adding a societal perspective to the analysis broadens the evaluation and may strengthen the evidence base for decision-makers to consider both a healthcare and societal perspective.
Introduction
Digital health interventions (DHIs), such as receiving health information via smartphone apps (mHealth), have great potential to enhance health and healthcare delivery. The value of mHealth resides largely in the accessibility and scalability of smartphone apps, and within these lies the potential for enhanced cost-effectiveness, higher-quality care processes1 and the facilitation of health-promoting self-management behaviours.2 Under these assumptions, several countries, including Denmark, have included DHIs in their National digital health strategies, intending to bring patients and healthcare systems closer together.3
Supported self-management is a hallmark of high-quality care for most conditions,4 including patients with low back pain (LBP). Supported self-management is recommended in clinical guidelines for LBP,5 6 and healthcare professionals are encouraged to assist patients in achieving successful self-management strategies.7 LBP is a highly prevalent and costly condition, and its prevalence is expected to rise in the future.8 With limited resources in healthcare systems, mHealth may be important in meeting future resource challenges relating to the societal and healthcare costs of LBP. Meta-analyses have documented the effectiveness of DHI, including self-management programmes for LBP, in reducing pain and improving function in musculoskeletal (MSK) conditions.9 10 Some evidence indicates that DHIs across various MSK conditions are cost-effective,11 but the cost-effectiveness of mHealth solutions for LBP is currently not well established.12
One example of a mHealth intervention is the selfBACK system.13 14 selfBACK offers individually tailored self-management advice concordant with national and international clinical guidelines for LBP, delivered through an artificial intelligence (AI)-based app.6 15 In a pragmatic randomised controlled trial (RCT) conducted in Denmark and Norway, we tested the clinical effectiveness of the selfBACK system as an add-on to usual care and showed a small but statistically significant reduction in LBP-related disability with uncertain clinical significance compared with usual care after 3 months.16 Similarly, secondary outcomes favoured the intervention and results were sustained at 9 months. No adverse events were reported, and secondary analyses imply no difference in effect among subgroups.17–20 The cost-effectiveness of interventions is, alongside clinical efficacy, important information informing resource allocation decisions also in clinically comparable interventions, as the results may have implications in terms of cost savings.21
In this secondary analysis of the selfBACK RCT, we aimed to evaluate the 9 months cost–utility and cost-effectiveness of the AI-based selfBACK app in addition to usual care vs usual care alone. The analysis was restricted to Danish participants, and the specific research objectives were to compare the selfBACK app to usual care in terms of (1) the mean per-person additional costs, (2) the incremental effectiveness in terms of quality of life, LBP-related disability and self-efficacy and (3) the incremental cost–utility and cost-effectiveness.
Methods
Study design
A cost-effectiveness analysis was conducted from a national healthcare perspective (primary scenario) and a societal perspective limited to long-term productivity (secondary scenario). The national healthcare perspective includes the monetary costs of healthcare utilisation. This perspective was chosen because treatment for LBP in primary care in Denmark is primarily financed by the government-funded national healthcare sector. Further, to evaluate the potential impact of including productivity loss, we performed an analysis from a societal perspective limited to long-term productivity in the form of long-term absenteeism (see Cost measures under Outcomes). The societal perspective considers everyone who may be affected by the intervention, including patients, employers, payers and society, but it does not matter who pays the cost and who receives the benefit.
The analysis was conducted as a within-trial analysis using the 9-month time frame of the trial. Data from the selfBACK trial were combined with individual-level data on healthcare resource utilisation and sickness absence from national registers. All costs and effects were calculated for the selfBACK intervention group and the usual care group. The study followed the Consolidated Health Economic Evaluation Reporting Standards reporting guidelines for economic evaluations.22 The trial was registered with ClinicalTrials.gov (Identifier: NCT03798288) before participant recruitment, and a statistical analysis plan was also registered there for this economic evaluation. Full details of the protocol13 and the clinical outcomes of the trial16 have been published elsewhere. Here, we provide a brief overview of the RCT.
Trial design and participants
Eligible participants were 18 years or older, had non-specific LBP within the last 8 weeks, scored 6 points or higher on the Roland-Morris Disability Questionnaire (RMDQ) at the time of screening, had consulted a primary care clinician (general practitioner, physiotherapist or chiropractor) or an outpatient spine clinic (Spine Centre of Southern Denmark) from 8 March 2019 to 14 December 2019. A total of 461 participants were included in the selfBACK trial, whereby 317 participants were from Denmark. Participants were randomised in a 1:1 ratio using permuted blocks with random sizes from 4 to 20 and stratified by country and referring clinician type. Only Danish participants were asked to consent to cross-referencing their information with national register data, and hence, the Norwegian participants were excluded from the current analysis. The current study was based on the data from the 297 Danish participants who provided both their unique personal identification numbers and consented to the linkage of their data to national health registries for research purposes.
Intervention
The selfBACK intervention comprises an app-delivered, evidence-based decision support system that, guided by case-based reasoning (a branch of knowledge-based AI), provides weekly, individually tailored self-management recommendations for three main components that are endorsed by clinical LBP guidelines: (1) physical activity (number of daily steps), (2) strength and flexibility exercises and (3) daily educational messages. In addition, the app provides general information about LBP and access to several tools (goal setting, mindfulness audios, pain-relieving exercises and sleep reminders) that participants could use at their convenience.23 The intervention group had access to the selfBACK app during the entire trial period as an add-on to usual care. The control group managed their LBP according to the advice or treatment offered by their healthcare professional.
Patient and public involvement
The content of the selfBACK system was developed using an intervention mapping process that included stakeholders and representatives from the target population. In addition, a process evaluation of the trial explored participants’ experience with the selfBACK system. No patient or public involvement was involved in the study design or interpretation of the current study’s results.
Outcomes
Cost measures
Data from the selfBACK trial were linked with individual-level data on healthcare resource utilisation from relevant registers hosted by Statistics Denmark and The Danish Health Data Authority. These included (1) all primary sector services provided by general practitioners, chiropractors, physiotherapists and medical specialists from the National Health Service Registry (Sygesikringsregisteret),24 except fully self-financed or private health insurance-paid consultations (estimated less than 10%) and (2) all secondary sector care utilisation related to assessments, tests and procedures at spine centres and rheumatology, neurology, orthopaedic and neurosurgery departments from the National Patient Registry (Landspatientregisteret)25 and (3) relevant redeemed prescriptive medication from the National Prescription Registry (Lægemiddeldatabasen).26
All directly and indirectly related costs covered by the publicly funded healthcare budget were included. Primary healthcare costs included all services (not restricted to LBP) and were valued according to the prevailing fee schedules agreed on in the collective agreement negotiated between the organisations of private healthcare practitioners and the Danish Regions (public payer). Hospital healthcare costs were valued using official hospital diagnosis-related group tariffs provided by the National Health Data Authority. Costs of prescriptive medication were calculated based on prices charged by the pharmacies (excluding VAT) (see table 1 in the Results section for details regarding included types of medication).
Baseline characteristics
In the scenario applying a limited societal perspective, the cost of productivity loss due to any long-term absence (≥4 weeks) from paid work was also included. Productivity loss was estimated using the human capital method based on weeks of all-cause long-term absence from work for participants who were not retired at baseline. The value of forgone earnings was assessed by the unadjusted national average gross wage. Data on formal production loss, measured in terms of long-term sickness absence, were retrieved from the Danish Register for Evaluation of Marginalisation,27 which contains weekly information on social transfer payments, including sickness absence benefits beyond 4 weeks for all residents in Denmark. The societal perspective was limited as no information was available regarding costs related to short-term sickness absence (<4 weeks), presenteeism or other types of time loss from work or other usual activities.
The intervention costs, defined as the valuation of the selfBACK intervention, were based on the list price of the selfBACK app reported by the company SelfBack Aps. A 3-month licence fee costs €67 (DKK500) and €3.6 (DKK27) per subsequent month. Participants had access to the app for 9 months, corresponding to a total cost of €89. All monetary values were inflated to 2022 euros using the Net Price Index maintained by Statistics Denmark. Since all costs and outcomes occur within 1 year, discounting was not applied.
Effectiveness measures
The primary effectiveness measure for this secondary analysis was quality-adjusted life-years (QALYs). QALYs were calculated based on the generic utility-based health status instrument, EuroQol-5L Dimension questionnaire (EQ-5D) collected at baseline, week 6 and months 3, 6 and 9, and weighted according to the Danish value set status.28 The 9-month QALYs were calculated by multiplying the utilities by the time a person spent in a particular health state. Transitions between health states were linearly interpolated. Two secondary clinical effect measures were assessed using the proportion of participants with a minimal important change (MIC) from baseline to 9 months on the (1) RMDQ (improvement ≥4 points)29 and (2) the Pain Self-Efficacy Questionnaire (PSEQ; improvement ≥5.5).30 31
Statistical analysis
The statistical analyses were conducted based on the intention-to-treat principle and with the current guidelines on conducting economic evaluations alongside clinical studies.21
To account for missing outcome data at follow-up, multiple imputation based on fully conditional specifications was used. As described in the statistical analysis plan for the main trial, 20 replications of imputed data sets were created and analysed separately and the results pooled using Rubin’s rules.32 There were no missing cost data.
Multivariable regression analyses were used to estimate incremental costs and QALYs. As costs were right-skewed and QALYs left-skewed, general linear models with a gamma family and a log function were applied for both the costs and QALY regressions. The choice of distributional family and link function was informed using the Hosmer-Lemeshow test and the Akaike information criteria. The secondary clinical effectiveness measures were analysed using general linear models with a Poisson family and a log function.
Two models were estimated for all outcomes; a base-case model and a model adjusting for baseline characteristics following evidence-based recommendations as in the main trial: age (years), sex (male vs female), educational achievement (<10, 10–12 or >12 years of education), duration of current pain episode (<1, 1–4, 5–12 or >12 weeks), and average pain intensity level in the past week at baseline (0–10 scale). In estimating costs and QALYs, the base-case model included healthcare costs during 12 months of prebaseline and baseline QALY values.33
Several sensitivity analyses were performed. The models were estimated for each outcome as a complete case analysis restricted to participants who completed respectively all EQ-5D questions (n=202), the RMDQ (n=201) and the PSEQ (n=205). Furthermore, a tornado diagram was conducted for the primary scenario to show the influence of changing selected parameter values on the estimated incremental cost-effectiveness ratio (ICER) when other factors remain at their base values. The following parameters were examined: QALYs in the intervention group changed by ±20%, primary care costs in the intervention group changed by ±20%, medicine costs in the intervention group changed by ±20% and hospital costs in the intervention group changed by ±20% and the app price halved or doubled.
Cost-effectiveness
The primary outcome was the ICER calculated as the difference in mean costs between the selfBACK and control groups divided by the difference between the two groups’ mean QALYs. For the secondary outcomes, ICERs were calculated as the cost per additional participant with an MIC on the RMDQ and PSEQ scores, respectively. The statistical uncertainty in the estimated ICERs was assessed using non-parametric bootstrapping with 1000 iterations, and the ICER for each iteration was plotted in a cost-effectiveness plane (ICER plane). The cost-effectiveness planes comprise four quadrants determined by the combination of the costs (incremental cost) and health effects (incremental effectiveness) associated with the intervention compared with the control. ICERs with a positive value fall into either the north-east (more effective, but more costly) or south-west quadrants (less effective but saves money). Lower ICERs suggest the intervention is more cost-effective (ie, benefits come at a relatively lower cost). In addition, the probability that the selfBACK intervention was cost-effective was evaluated using a range of hypothetical threshold values for the willingness to pay (WTP) for a QALY. The results were presented graphically as cost-effectiveness acceptability curves (CEACs), which show the probability of cost-effectiveness on the y-axis and WTP thresholds (cost per QALY gained) on the x-axis.
P values <0.05 were considered statistically significant. Analyses were performed by using STATA V.18.
Results
Descriptives
Of the 297 Danish participants, 148 were randomised to the selfBACK intervention while 149 were randomised to the control group. The participant characteristics at baseline are provided in table 1. No statistically significant differences were found between the two groups. The typical participant was a well-educated woman around 50 years old, recruited from a chiropractor, who had experienced moderate pain for more than 12 weeks (see online supplemental table 1 for EQ-5D index scores for the intervention and control groups).
Supplemental material
Costs
Table 2 shows the costs by type of healthcare service, pharmaceuticals and productivity loss between baseline and 9 months of follow-up. Pharmaceuticals constitute the largest part of the healthcare costs. Average costs related to productivity losses were higher than total healthcare costs. However, as the median and IQR indicate, only very few patients experienced long-term sickness absence. Apart from the intervention costs, no statistically significant differences in costs were found between the groups.
Unadjusted costs within the 9 months of follow-up, 2022 euros
Table 3 shows the estimated incremental costs for the selfBACK group compared with the control group. The average costs for the selfBACK group were similar to the costs in the control group (mean difference €230 (95% CI −€136 to €595)) (see online supplemental table 2 for unadjusted results). Compared with the results from the healthcare perspective (primary scenario), the analysis in the scenario including productivity costs showed higher average costs (mean difference €589 (95% CI −€2120 to €3298)); but again, the difference in costs between the two groups was insignificant (see table 3B).
Incremental costs, effect and cost-effectiveness ratio based on imputed observations—model adjusted for baseline characteristics
Effectiveness
Table 3 also shows the estimated difference for the three outcome measures. The selfBACK group showed a statistically significant improvement in QALYs (mean difference 0.03 (95% CI 0.01 to 0.05)) compared with the control group. The percentage of participants with an MIC on the RMDQ and PSEQ was 14% (95% CI 0.02% to 0.26%) and 18% (95% CI 0.06% to 0.29%) higher in the selfBACK group, respectively. The unadjusted estimates were similar.
Cost-effectiveness
The point estimates of costs and effects indicated higher costs and higher effects for the intervention, resulting in positive ICERs for all outcomes. The results showed an ICER of €7336 per QALY gained, and ICERs of €1302 and €1634 for an additional participant with an MIC on the RMDQ and PSEQ scores, respectively. ICERs based on unadjusted models showed similar results (online supplemental table 2). Including productivity loss in the estimation of the ICERs increased the ICERs (ie, less cost-effective) to €18 821 per QALY gained and €3341 and €4192 for an additional participant with an MIC on PSEQ and RMDQ, respectively.
ICER plane
Figure 1 illustrates the results of the bootstrapped iterations of differences in costs and effects in ICER planes, with the results from the analysis in the primary scenario and the scenario including productivity. In the primary scenario, most of the iterations (91.4%) in the plane (figure 1A) were in the north-east quadrant, meaning that the selfBACK intervention was more effective but also more costly. However, 8.6% of the iterations indicated that the selfBACK intervention was more effective and less costly, thereby dominating usual care. The probability of the selfBACK intervention dominating usual care increased when including productivity costs with 23.0% of the iterations in the south-east quadrant (figure 1B). However, most (77.0%) iterations were still in the north-east quadrant. For PSEQ, the distribution of iterations across the four quadrants in the ICER plane resembled that of the primary scenario. RMDQ had iterations in all four quadrants, however, most (90.3%) of the iterations were found in the north-east quadrant similar to QALY and PSEQ (figure 1C,D). Adding productivity loss to the PSEQ and RMDQ ICER planes had the same effect as in the primary scenario.
Cost-effectiveness planes. The cost-effectiveness planes of costs per QALY for the model adjusted for baseline characteristics based on 1000 bootstrap replications showing (A) the ICER plane of costs per QALY in the primary scenario (healthcare perspective including healthcare costs and costs of intervention); (B) the ICER plane of costs per QALY in the secondary scenario (societal perspective limited to healthcare costs, costs of intervention and long-term productivity loss in the form of long-term absenteeism) ; (C) the ICER plane of costs (healthcare costs and costs of intervention) per minimal important change (MIC) on the PSEQ score for the primary scenario and (D) the ICER plane of costs (healthcare costs and costs of intervention) per MIC on the RMDQ score for the primary scenario. ICER, incremental cost-effectiveness ratio; PSEQ, Pain Self-Efficacy Questionnaire Score; QALY, quality-adjusted life-year; RMDQ, Roland-Morris Disability Questionnaire.
Cost-effectiveness acceptability curves
Figure 2 illustrates the CEACs for cost per QALY from a healthcare perspective (primary scenario) and a societal perspective limited to including productivity loss (secondary scenario) and the impact of including productivity in the analysis. The threshold values are plotted on the horizontal axis against the probability of selfBACK being cost-effective on the vertical axis. The shape of the CEACs resembles a logistic function; the probability increased steeply at low values and slowly at higher values, reaching 100% at a value of approximately €35 000. According to NICE (The National Institute for Health and Care Excellence) guidelines in UK, the cost-effectiveness threshold is approximately €23 250 (£20 000). At this threshold, the probability of the selfBACK intervention being cost-effective was 98%. Including productivity loss in the analysis reduced the probability to approximately 50%. This probability never reached 100% within a range WTP of up to €50 000.
Cost-effectiveness acceptability curve for costs per QALY. Cost-effectiveness acceptability curve for cost per QALY from a healthcare perspective (primary scenario) and a societal perspective limited to including productivity loss (scenario including productivity). The curves indicate the probability (y-axis) of the selfBACK app (as an add-on to usual care) being cost-effective compared with usual care, given a specific threshold value (x-axis) for an additional QALY. QALY, quality-adjusted life-year; WTP, willingness to pay.
Sensitivity analyses
Restricting the sample to complete cases did not impact the cost difference from a healthcare perspective nor the probability of an MIC on RMDQ (online supplemental table 3). The point estimate of the MIC on PSEQ increased slightly from 0.18 to 0.26 in the complete case analysis. Including productivity loss increased the cost difference by approximately €40–€100 and the CIs widened substantially. The higher cost estimates resulted in a small increase in the ICERs of two out of three outcomes (health benefits at a higher cost) and a small decrease for MIC on PSEQ due to the increased effect for that outcome. Furthermore, around 27% of the iterations fell in the south-east quadrant of the ICER plane (more effective and saves money), indicating a considerable chance that the intervention dominated usual care (not shown). Contrarily, including productivity loss in the unadjusted model (online supplemental table 3) decreased cost estimates substantially resulting in much lower ICERs. The tornado diagram (figure 3) reveals the ICER varied from €6114/QALY to €9170/QALY, and that changes in the medicine costs have the largest effect on the estimated ICER. A reduction of 20% in medication costs resulted in a negative ICER (more effective and saves money). For all parameters, increasing the parameters by 20% or doubling the app price, still resulted in an ICER below a threshold value of €23 250.
Tornado diagram. Tornado diagram comparing the relative impact of selected parameters on the estimated ICER. The vertical line in the tornado diagram represents the ICER (€7336/QALY) from the primary scenario (model adjusted for baseline characteristics) to provide a reference to the changes in the ICER. ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year.
Discussion
The results show that the selfBACK intervention is associated with an increase in health-related quality of life and clinical outcomes, but also higher costs compared with usual care. From a national healthcare perspective, the intervention has a 98% probability of being cost-effective at the NICE threshold value. The probability of being cost-effective reduces to 50% when including productivity loss. Even though the selfBACK intervention is found to have an additional clinical effect compared with usual care, it does not seem to affect the healthcare resource utilisation pattern within the 9-month follow-up period. Including productivity loss in the analyses adds noise to the results. Costs increased for the selfBACK group in both the primary model and the complete-case analysis, with the width of the CIs increasing. However, when not adjusting for baseline characteristics in the complete-case analysis incremental costs decreased substantially. This suggests that the higher productivity cost in the intervention group is driven by specific characteristics of participants with missing outcome data. However, adjusting for baseline characteristics seems to remediate this issue. This indicates that results are very sensitive to the characteristics of the analysed sample and that few participants with large productivity losses impact the results considerably.
Several systematic reviews exist on economic evaluations of DHIs for a variety of health conditions,34–36 but only one considers MSK conditions in general,11 and few studies exist for self-management of LBP. Despite the diverse conditions, outcomes and perspectives examined, these studies generally underscore a potentially beneficial impact of DHIs. However, they also emphasise the necessity for future research to adhere more closely to established reporting guidelines for economic evaluations. Two economic evaluations exist on the German mHealth app Kaia. One study found it cost-effective with an ICER of €5486 per QALY based on a Markov model simulation.37 The other reported an ICER of €34 315 per QALY, with the difference between the two results explained by a minimal incremental effect on QALYs in the Monte Carlo simulation.38 The main differences from our study are that the German analyses were model-based with a time horizon of 3 years, the app was a stand-alone intervention,38 39 their study sample size was relatively small, and the quality of life estimates was uncertain due to data inadequacies.
A strength of our study is the ability to link the trial data with data on resource use and costs from various national registries. The registries included complete information for all participants so there were no missing data on costs. Furthermore, the estimation of costs was based on actual activity for the entire period and not on extrapolations of participant questionnaires with shorter recall periods. This strengthens the internal validity of the study. Including all-cause production loss in economic evaluations can be problematic as some participants may stay off work for numerous reasons unrelated to the condition and intervention under scrutiny. Conversely, adding a societal perspective to the analysis broadens the evaluation36 40 and may strengthen the evidence base as it renders the possibility for decision-makers to consider both a healthcare and societal perspective.40 The costs associated with productivity loss in our sample were high in magnitude compared with the healthcare costs. We only have access to data on all-cause, long-term sickness absence and were thus unable to separate absence related to LBP from other causes for sickness absence or to elucidate the effect of the intervention on short-term sickness absence and presenteeism. Not having access to short-term sickness absence data likely causes a significant underestimation of overall productive loss. The RCT was powered to a sample size of at least 350 participants. As the cost-effectiveness is based only on the Danish participants, our cost-effectiveness study is underpowered. Furthermore, the study was not powered to detect differences in costs and the selfBACK trial was not aimed at reducing sickness absence and included people outside the workforce. It would require a larger sample size to draw firm conclusions regarding the effect on productivity loss. Due to the nature of the register data, we have included non-LBP-related costs, and we cannot exclude that some of the incurred costs are unrelated to LBP. However, the randomisation and analytical strategy account for any systematic group differences. Prescriptive medication constituted the largest part of the healthcare costs, but unfortunately, we did not have data on over-the-counter medication utilisation (unless prescribed by a doctor (pack sizes higher than 20 pills require a prescription)). Over-the-counter analgesics are the most frequently used medication for LBP41 and this implies that a large part of the potential effect on the use of analgesics is not included in the cost estimates. Furthermore, this study did not have data on any treatment received outside the public healthcare system. Almost half of the Danish population has access to private health insurance through work, and although not all participants are in the workforce, we have no reason to believe that the groups differ in terms of access to health insurance. Nevertheless, we cannot rule out that we would have seen a difference in the utilisation of healthcare services if the study had access to data on treatment not financed by the national healthcare sector.
DHIs are complex interventions,42 and generalising results between DHIs may be particularly challenging with a lack of transparency concerning methods and context and rapid technological development.40 While the selfBACK app seems valuable for people with LBP, uncertainties exist in the data, necessitating further studies to inform the possible implementation in clinical practice. The selfBACK app aimed to reduce LBP-related disability by promoting better self-management. Self-management encompasses understanding when and how to manage your condition independently and when to seek assistance from the healthcare system. Consequently, effective self-management can lead to reduced healthcare resource utilisation and decreased sickness absence for some individuals. However, for others, it may result in more frequent, appropriate and timely healthcare-seeking behaviours. Making solutions such as selfBACK generally available to people with persistent or recurrent LBP has the potential for real public health impact,43 but the complex patient pathways through the healthcare systems highlight the need for further research to uncover the underlying mechanisms of change to inform healthcare administrators and politicians in making future healthcare decisions.
Participants in both the control and intervention groups were offered a step-detecting wristband as part of the clinical trial. The wristband is not required for the app to work, and the app can easily be linked to other devices, such as smartwatches, or be used without an activity tracker. Therefore, the cost of the wristband was not included in the analysis. However, it might be relevant to include this cost in an assessment of the implementation costs associated with using selfBACK in clinical practice. In addition, the selfBACK system includes the possibility of sharing patient data with a healthcare practitioner through a clinical dashboard. This module was not part of the trial, but future research should consider including costs related to the time spent by the practitioner to be trained and apply the clinical dashboard, and to the implementation and maintenance of the dashboard to their IT system, as well as any potential benefits of using the dashboard. Analyses of equity impact need consideration,44 and finally, there is a need to scrutinise the evidence base for determining the threshold value for willingness to pay, as implementing DHIs based on questionable thresholds may result in ineffective priorities.45
Conclusion
The results show that the selfBACK intervention is associated with an increase in health-related quality of life and clinical outcomes, but also higher costs compared with usual care alone. From a healthcare perspective, the probability of the selfBACK intervention being cost-effective in the primary scenario is 98% at a threshold value of €23 250. The probability of the selfBACK intervention to dominate usual care is 8.6%. The analysis from a limited societal perspective is highly sensitive, and the results’ uncertainty warrants cautious interpretation.
Data availability statement
Data are available on reasonable request. Requests to access RCT data supporting this study's findings can be made to MJS (m.jensen@kiroviden.sdu.dk) at the Department of Sports Science and Clinical Biomechanics, University of Southern Denmark, Odense, Denmark and Chiropractic Knowledge Hub, University of Southern Denmark, Odense, Denmark. The cost data are available pending permission from Statistics Denmark’s Research Service.
Ethics statements
Patient consent for publication
Ethics approval
This study involves human participants and was approved by Danish ethics committee with ID no. S-20182000-24. Participants gave informed consent to participate in the study before taking part.
Acknowledgments
We want to thank all participants in the study and those who agreed to cross-reference their information with national register data.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
X @@LineLpk, @CeciKOveraas, @LouiseSandal, @janhartvigsen, @KASogaard, @mork_paul, @MStochkendahl
Contributors MJS, the PI on this cost-effectiveness analysis and the one who secured funding for this project, conceived and designed the study with LPK, CKØ, CVS, LFS, JH, KS and PJM (the PI on the selfBACK project). A statistical analysis plan was registered in ClinicalTrials.gov with CKØ as the author, LPK and CVS as coauthors and contributions from the rest of the author group. LPK, with assistance from CVS and LFS, led the application for data access, the statistical analyses and the drafting of the method and results sections. CKØ drafted the abstract, introduction and discussion part of the manuscript. Interpretation of the data and critical revision of the manuscript for important intellectual content was done by all authors; LPK, CKØ, CVS, LFS, JH, KS, PJM and MJS. All authors have given their approval prior to publication of the final version of this paper. MJS is the guarantor.
Funding The selfBACK project was funded by the European Union Horizon 2020 research and innovation programme under grant agreement No. 689043. This cost-effectiveness analysis has received separate funding from the Danish Chiropractic Foundation (grant no. A4526).
Competing interests All authors of this paper, except the health economists LPK and CVS, have been involved in the EU project selfBACK. The company selfBack Aps is introducing the selfBACK system to the commercial market via a licensing agreement with the NTNU Technology Transfer Office. Authors affiliated with NTNU may in the future receive personal compensation according to the licensing agreement with the NTNU Technology Transfer Office. Furthermore, Center for Muscle and Joint Health at the University of Southern Denmark may in the future receive research funding according to the licensing agreement with the NTNU Technology Transfer Office.
Patient and public involvement Patients and/or the public were involved in the design, or conduct, or reporting, or dissemination plans of this research. Refer to the Methods section for further details.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.