There is growing interest in tailoring treatment decisions to individual patient characteristics, but few studies have examined the implementation and performance of individualized treatment rules (ITRs) for count data. In this work, we compared ITR methods for randomized trials with count outcomes and explored how sample size and the distribution of heterogeneity of treatment effect (HTE) influence the validity of treatment recommendations.

Using a simulation study, patients were randomized to one of two treatments and assigned to five responder strata representing different HTE scenarios. We evaluated several ITR methods in terms of value function and accuracy, and additionally illustrated their application in a case study of patients with multiple sclerosis. Overall, ITR methods performed better under favorable conditions such as larger sample sizes, greater treatment heterogeneity, or fewer neutral patients, but were outperformed by fixed treatment strategies when sample sizes were small or HTE was limited. Larger sample sizes could compensate for smaller HTEs, while stronger HTEs could compensate for more limited data.

In the case study, we identified evidence of HTE and developed a tree-based ITR that outperformed fixed treatment recommendations. Overall, the findings highlight that the performance of ITRs depends not only on sample size, but also on the magnitude and distribution of treatment effect heterogeneity. Simulation scenarios informed by clinical knowledge may help determine whether HTE estimation is feasible and, if so, which ITR approach is most appropriate.

Source code: https://github.com/phoebejiang/precmed_sim