Presentation (CTI)

OBJECTIVES: For health technology assessment (HTA) and policy decisions, comparative estimates versus standard of care are required, but individual patient data for comparator studies are often unavailable. Population-adjusted indirect comparisons such as multilevel network meta-regression (ML-NMR) can address cross-trial differences when networks are connected. Unanchored MAIC/STCs are limited because they primarily target the aggregate-data comparator population, which may not match the decision-relevant target population. We extended the ML-NMR to time-to-event (TTE) outcomes by incorporating single-arm studies and compared the results against a connected “gold-standard” ML-NMR analysis.
METHODS: Based on the newly diagnosed multiple myeloma TTE dataset from the multinma package, which comprises five trials comparing lenalidomide and thalidomide versus placebo, we constructed an artificial disconnected network by removing placebo arm from one of the studies, leaving thalidomide as a single-arm trial. Standardized prognostic factors were used to quantify between-study population similarity using Euclidean distance matrix. The most similar connected study was selected to predict the baseline response for the disconnected study, enabling network reconnection. Standard parametric survival models were applied. Relative treatment effects were estimated as hazard ratios (HRs) with 95% Credible Intervals (CrIs) and compared with estimates from the gold-standard analysis. Model fit was assessed using leave-one-out information criterion (LOOIC) with lower values indicating better fit.
RESULTS: The Weibull model demonstrated the best fit for both the reconnected network (LOOIC = −5575.3) and gold-standard ML-NMR (LOOIC = −6134.7). Reconnected, unanchored analyses produced comparative estimates closely matching to the connected reference for thalidomide versus placebo (HR, 95%CrI: 0.82, [0.55, 1.18] vs 0.77 [0.53, 1.12]) and lenalidomide versus placebo (0.48 [0.39, 0.63] vs 0.49 [0.40-0.60]).
CONCLUSIONS: Extending ML-NMR to disconnected survival networks offers a practical route from single-arm evidence to HTA-relevant comparative estimates. Careful selection and justification of contributing studies for connection to network is essential to manage bias and uncertainty in the estimates.