Methodological considerations in the analysis of time-to-event outcomes with MAIHDA

We welcome the recent extension of multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) to time-to-event outcomes.1 Before researchers deploy the method, we highlight three considerations that could be subsumed under what the authors call ‘unobserved heterogeneities’ and that do not necessarily constitute evidence of intersectionality.

Scale dependence

MAIHDA is often applied in an intersectional framework to evaluate departures from additivity in how social identities shape health outcomes.2 A key challenge is that interactions are scale dependent: they may be absent on the multiplicative (log-hazard) scale but appear on the additive (survival probability) scale due to the non-linear transformation between cumulative hazard and survival. Moreover, the induced interaction is generally time-varying, even under proportional hazards.3

We illustrate this with two binary variables, and , simulating hazard-scale interactions from the Weibull model:

and the corresponding survival function:

where is the Weibull shape parameter. We…

Read the full article ›