Objectives

Estimates of uncertainty can have important consequences for reimbursement decisions and associated data collection plans. Heterogeneity in the treatment effect is one such form of uncertainty and may be particularly apparent in histology-independent (HI) interventions as evidence is typically gathered across a number of different tumour types. This study demonstrates the implications of additional data on empirical estimates of heterogeneity in HI trial data.

Methods

Two clinical effectiveness data cuts (24 months apart) for the HI intervention, larotrectinib, were obtained from the literature. A Bayesian hierarchical model (BHM) was used to quantify the between-tumour heterogeneity present in the data, as measured by the standard deviation, σ. In addition, the pooled response rate for the tumour types present in the trial was estimated accounting for the heterogeneity.

Results

The results of the BHM show the early data cut, comprising a smaller patient population (n=55), to have substantial between tumour heterogeneity in response, σ = 2.86 (95% credible interval (CrI) 0.92 to 4.83). This heterogeneity reduced considerably in the updated data cut, comprising increased patient numbers (n=153), σ = 1.08 (0.15 to 2.63). Heterogeneity was also estimated with less uncertainty.

In the first data cut, the naïve mean response was 74.5%. This was reduced to 60.9% (16.0 to 91.8%) when accounting for heterogeneity and included considerable uncertainty. In the later data cut, the naive response was 81.0%, which was approximately the same when accounting for heterogeneity, 80.5% (65.5% to 91.7%). Uncertainty in the probability of response was also reduced.

Conclusions

This study illustrates how estimates of heterogeneity may evolve as further data becomes available. This may have consequences for how decision makers value HI technologies across specific tumour types; the choice of modelling approach taken; and, the value of further data collection.