BackgroundIn economic evaluations, average intervention effects are usually applied to a population. However, this fails to reflect the change in the distribution of HbA1c due to heterogeneous responses to weight loss. We aimed to investigate whether allowing heterogeneous treatment effects using a beta regression better represented the distribution of HbA1c after a weight-loss intervention, and how this affected cost effectiveness. MethodsThe Glucose Lowering through Weight Management (GLoW) trial evaluated the effectiveness of a diabetes education and weight-loss intervention against a standard diabetes education programme. Adults diagnosed with type 2 diabetes within 3 years were recruited from Clinical Commissioning Groups across 159 sites in England from July 20, 2018, to July 22, 2018. Ethics approval (18/ES/0048) and participant informed consent were obtained. Considering the between-treatment-arm difference in HbA1c after 12 months, we compared a mean-effect estimated from a mixed-effects regression to a heterogeneous effect estimated from a beta regression performed on 12-month HbA1c conditional on baseline HbA1c, gender, diabetes duration and intervention group. We used the School of Public Health Research (SPHR) Diabetes Treatment model to apply these treatment effects and evaluate the lifetime NHS costs and quality-adjusted life-years (QALYs), discounted at 3·5%. The microsimulation model estimated diabetes-related health outcomes using the UK Prospective Diabetes Study Outcomes Model 2 risk equations and risk factor trajectory equations, alongside estimating diabetes remission, osteoarthritis, and cancer. We calculated the incremental net benefit (INB) of the intervention using a £20 000 per QALY valuation, by deterministic analysis. The GLoW trial is registered with the ISRCTN Registry, ISRCTN18399564. FindingsThe trial recruited 577 participants (mean age 60 years; 278 [53%] female, 247 [47%] male; 474 [91%] white ethnic background). Applying heterogeneous HbA1c changes better reproduced the skewness in post-intervention HbA1c than applying a mean-effect (Kolmogorov-Smirnov test p=0·02 compared with p=0·0000007). The beta-regression method suggested the intervention was more cost-effective, estimating an INB of £736 per person, compared with £584 when applying the mean-effect. InterpretationAlternative regression specification methods should be considered when evaluating the cost-effectiveness of interventions if the key intervention outcomes are not normally distributed. However, this alternative method requires further investigation to conclude its appropriateness in evaluating cost-effectiveness in different contexts. FundingNational Institute for Health and Care Research (NIHR) Programme Grants for Applied Research Programme (Reference Number RP-PG0216-20010).