MT4 Mixed Treatment Comparisons Using Aggregate- and individual-Participant Level Data: An Efficient Use of Evidence for Cost-Effectiveness Modelling

Abstract

OBJECTIVES: Cost-effectiveness analysis must use all relevant sources of evidence to inform reimbursement decisions. Mixed treatment comparisons (MTC) extends the traditional pair-wise meta-analytic framework to facilitate the synthesis of information on more than two interventions. While most MTCs use aggregate data (AD), a proportion of the evidence base might be available at the individual level (IPD). This paper develops novel statistical models aimed to fully exploit the existing data, regardless of the format (i.e. AD or IPD). METHODS: We develop a series of novel Bayesian statistical MTC models to allow for the simultaneous synthesis of IPD and AD, while considering study and individual level covariates, and use these to inform a decision model. RESULTS: The effectiveness of home safety education and the provision of functioning smoke alarms (binary outcome - Yes/No) for the prevention of childhood injuries in the household was used as a motivating example. Case study included 20 trials (11 AD, 9 IPD), summing up to 11,500 participants. Seven strategies were defined and a network of evidence was constructed. Irrespective of the evidence format used, all models which did not consider information on covariate(s) showed equivalent results, i.e. more intensive interventions (providing education, equipment (with fitting) and home inspection) were more effective (OR vs usual care of 4.5 (95% credible interval: 1.4 to 14.8). Results of synthesizing IPD using information on a covariate account for possible ecological bias and show a clear improvement in accuracy over estimated treatment-covariate interactions, when compared to results obtained from synthesizing AD. CONCLUSIONS: Including evidence at IPD level in the MTC is advisable when exploring participant level covariates; even when IPD are only available for a fraction of the studies forming the evidence base. Our findings suggest that adjusting for covariates impact produces intervention effect estimates of higher accuracy, which is valuable for estimating subgroup effects or adjusting for inconsistency.