How can within-trial economic evaluations support evidence generation prospectively?
The HEOR team of Hardian Health is offering a new service for digital health companies to support them with evidence generation when their products are evaluated post-market through randomised controlled trials.
Our new package involves the analysis of primary data to estimate health outcomes, costs, and Incremental Cost-Effectiveness Ratios (ICERs) for decision making. By leveraging primary data collection, health outcomes and costs derived from randomised trials can also be extrapolated as robust inputs for subsequent, post-market cost-effectiveness models.
What is a within-trial economic evaluation and what are its benefits?
Within-trial economic evaluations are conducted alongside randomised controlled trials. By collecting patient- and investigator-reported data alongside clinical endpoints, health economists can derive key metrics such as Health-Related Quality of Life (HRQoL) and healthcare resource use over a trial’s follow-up period. These can be valued, typically using EQ-5D value sets and unit costs, to estimate average QALYs and costs for each comparator. Finally, by adopting robust statistical methods, such as multiple imputation by chained equations (MICE) to account for missing data, non-parametric bootstrapping methods are usually applied to derive incremental cost-effectiveness ratios (ICERs). Such evaluations present health economic outcomes that reflect the PICOTS framework under which the randomised trial is designed and conducted. These economic outcomes can reflect high internal validity, and -in the case of pragmatic randomised trials- real-world evidence without making additional assumptions about clinical practice.
The benefits of within-trial economic evaluations can include:
Low marginal cost of introducing economic evaluations alongside randomised trials- The marginal cost of collecting health outcome and cost data can be relatively low, as the data collection process can be feasibly integrated into a randomised trial, leveraging existing clinical site and data collection infrastructure.
Primary data can be extrapolated to inputs for model-based economic evaluations- Primary health outcome and cost data can be extrapolated and used as reliable inputs for longer-term, model-based cost-effectiveness analyses. If further external evidence is available on comparators apart from the randomised trial, network meta-analysis methods can be adopted to generate inputs capable of optimising the robustness of the model-based cost-effectiveness results, thus facilitating decision making.
Can produce inputs reflecting different health states in health economic models- Changes in health status (e.g. adverse events occurring) during the follow-up of a randomised trial can be linked to changes in resource utilisation or HRQoL because of the availability of primary data. Such changes can be extrapolated to inputs for model-based economic evaluations.
Subgroup analyses might be feasible- If the sample size of recruited trial participants is sufficient, subgroup analyses of health outcomes, costs and ICERs can be undertaken, although these are typically exploratory as randomised trials are usually powered for primary clinical endpoints rather than for economic outcomes.
Can account for correlation between QALYs and costs- Patient-level data can show the joint distribution of costs and effects, enabling proper probabilistic sensitivity analysis (e.g. bootstrapped estimates on incremental costs and incremental QALYs) instead of relying on independence assumptions that are typically made in model-based economic evaluations.
Can reflect real-world evidence under certain conditions- If the study is a pragmatic, randomised controlled trial, primary health outcome and cost data can reflect real-world resource use, HRQoL and treatment delivery patterns that can effectively inform decision-making. Such evidence can be sought by payers for digital health technologies in addition to diagnostic or clinical efficacy evidenced by retrospective studies.
Can inform evidence generation- Within-trial economic evaluations can strengthen the certainty of evidence for digital health companies, e.g. go/no-go decision making or pricing, before choosing the commercialisation pathway for their product.
Can achieve internal validity- If health outcome and cost data are collected alongside a randomised trial, the derived (cost-)effectiveness estimates can achieve high internal validity, thus reducing the risk of confounding.
How can a within-trial economic evaluation be successful?
Prior to initiating any randomised controlled trial, clinical sites, trial coordinators, manufacturers of digital health technologies and independent evaluators (such as statisticians and health economists) should establish an aligned data collection and trial management strategy to optimise the clinical and health economic evaluation's value for the health technologies compared. Key components for the within-trial economic evaluation’s specification include:
Questionnaire Design: Designing baseline and follow-up questionnaires, as well as choosing and agreeing on questionnaire items to be completed by study participants and investigators from the involved clinical sites over the study’s follow-up period.
Analysis Specification: Drafting a Health Economics Analysis Plan (HEAP) to ensure transparency, reproducibility, and bias reduction, alongside a Statistical Analysis Plan (SAP) tailored to primary and secondary clinical outcomes. The HEAP should also specify the software that will be used for analysing and generating datasets (e.g. STATA, R).
Data sharing agreement: Ensuring that primary data can be shared with the independent evaluators in a manner that complies with ethical approvals obtained for the randomised controlled trial, i.e.through anonymised participant data, as well as selecting a data repository where all datasets will be stored.
Following data collection, health economists should compile and analyse the shared datasets according to the pre-specified HEAP document:
Estimation of QALYs: Responses to HRQoL questionnaire items (e.g. EQ-5D-5L) are mapped to value sets recommended by NICE or other national reimbursement agencies. QALYs are then calculated using the under the area curve (AUC) approach.
Estimation of costs: Resource utilisation items are monetised using appropriate unit costs at pre-specified price levels. Total costs are then calculated by aggregating costs recorded at the baseline and follow-up periods.
Accounting for missing data: Patterns of data missingness are thoroughly evaluated to prevent biased cost and QALY estimates. For instance, if data are found to be missing at random, multiple imputation methods can be applied.
Cost-Effectiveness analysis: Base-case mean QALYs and costs should be estimated and presented for every comparator. Non-parametric bootstrapping methods should be used to derive incremental QALYs and incremental costs associated with the comparators involved, as they can account for baseline differences in the intervention groups and correlations between QALYs and costs. Incremental QALYs and incremental costs can then be synthesised to calculate ICERs.
Sensitivity analysis: Sensitivity analyses should be undertaken to account for uncertainty in the within-trial economic evaluation results.
Dissemination: The health economic results can be disseminated via peer-reviewed publications and conference abstracts.
Data sharing: For reproducibility purposes and for extrapolating primary data to inputs suitable for longer-term health economic models, all datasets generated and analysed during a randomised trial should be ideally stored in a repository, in line with the pre-specified data sharing agreements.
Provided that clinical sites or digital health technology manufacturers are able to share anonymised participant data with Hardian, in line with data sharing agreements established prior to commencing any prospective study, our team possesses the experience and expertise to guide your post-market HEOR evidence generation journey through this specialised package.
