Cause of asymmetric robustness in the tipping-point analysis

Determine whether the robustness of the lower tail of the credible interval to the degree of historical-data borrowing, contrasted with the greater movement of the upper tail when the exchangeability proportion exceeds 0.5, is caused by separation between the historical control data and the current data before 12 months and their congruence after 12 months.

Background

The paper applies the NP-LEAP model to assess how survival-probability conclusions change as the assumed proportion of exchangeable historical individuals varies from 0 to 1. The reported analysis finds that the lower tail of the 95% credible interval is relatively stable across borrowing levels, whereas the upper tail shifts substantially for exchangeability proportions above 0.5.

The authors attribute this observed asymmetry to the relationship between the historical and current control-data survival curves: the curves appear separated before 12 months but more congruent after 12 months. They explicitly present this attribution as a conjecture rather than a demonstrated explanation. They further note that changing the estimand to 24-month survival alters the lower bound, motivating the unresolved question of whether the time-specific alignment of the datasets explains the differing sensitivity of the interval bounds.