Presentation (CTI)

OBJECTIVES: Evaluating comparative effectiveness for novel targeted therapies often requires indirect treatment comparisons against mutation/biomarker-positive subgroups in comparator trials. Population-adjusted indirect comparisons (PAIC) can be used to adjust for differences in patient characteristics for comparisons involving two studies—one with available individual patient data, and the other with only aggregate data (AgD). While published AgD often includes outcomes by key subgroups, subgroup-level information on baseline characteristics is rarely available. Without complete subgroup AgD, standard PAIC approaches are infeasible at the subgroup level. We propose a method for quantifying the impact of covariate adjustment via PAIC on effect estimates when subgroup AgD is incomplete.
METHODS: AgD trial subgroup mean baseline characteristics can be expressed in terms of (i) the share of patients in the subgroup, (ii) the overall mean baseline characteristics, and (iii) the difference in subgroup versus overall means (the “imbalance term”). If (i) and (ii) are known, we can compute the subgroup mean under a proposed value for the imbalance term and use the computed subgroup mean as an input into a standard PAIC. We propose that PAIC estimates be computed under a range of values for the imbalance term to assess the sensitivity of effect estimates and determine how large the imbalance would need to be to reach a tipping point (e.g. for the estimated effect to cease to be statistically significant).
RESULTS: Results of a simulation will be presented demonstrating the application of the method under several PAIC approaches: matching-adjusted indirect treatment comparison (MAIC), simulated treatment comparison (STC), and multi-level network meta-regression (ML-NMR). Point estimates and confidence bounds will be plotted by imbalance factor and a tipping point threshold for the imbalance term will be computed.
CONCLUSIONS: The proposed method allows for quantifying potential bias due to confounding where it is infeasible to directly adjust for confounders via standard PAIC methods.