The Effects of Heterogeneity in the Comparative Effectiveness of Individual Treatments in Randomised Trials

In some randomised trials, the new treatment can be compared with
usual care which can include multiple treatments and result in a heterogeneous
control group. In this paper, we use simulation to assess the performance of
various statistical methods to infer the individual effects of the various control
treatments. These methods include inverse Probability Weighting, Doubly Robust
Inverse Probability Weighting, Propensity Score, Disease Risk Score,
Standardization and Multivariable Logistic Regression. Different scenarios were
tested including unmeasured heterogeneity with or without confounding. The
methods perform well when heterogeneity and confounding are both fully
captured; however, for the scenarios where heterogeneity is not fully captured this
leads to biased effect estimates, particularly where there is also unobserved
confounding. Thus, leading to potentially misleading comparative effectiveness of
individual treatments.