Purpose: Treatment effects are often heterogeneous; different patients respond differently to the same medication. Thus, current evidence reported as an average treatment effect across a large population may not be appropriate. Providing evidence across smaller more homogeneous subgroups may be more beneficial. However, there are multiple methods that can be used to stratify patients; each producing different subsets of patients. As a result of these differing subsets, the interpretation of their results may vary. The purpose of this study is to compare the results of the newly developed prognostic propensity score (PPS) method to the traditional propensity score (PS) method; in order to understand how they can be best utilized to identify the optimal treatment.
Methods: To compare the PS and PPS we used data derived from a 100% sample of the California Medicaid program fee-for-service paid claims data from the period 1994-2002 for patients who received at least one prescription for risperidone or olanzapine. We calculated the probability of receiving olanzapine (PS) and the probability of success on olanzapine (PPS).
Results: The final study sample was composed of 38,804 recipients of olanzapine and 37,560 of risperidone. The olanzapine PS model and PPS model accounted for 48 and 55 covariates respectively. Although the PS and PPS methods both identified HTE the predictors of receiving olanzapine (PS) and the predictors of success on olanzapine (PPS) varied significantly. For example, patients diagnosed with dementia were less likely to receive olanzapine but were more likely to benefit from it. In general, the PS results indicated that patients with the highest probability of receiving olanzapine would actually have benefited more from risperidone. While the results of the PPS confirmed that those with the highest predicted probability of success on olanzapine did indeed benefit more from olanzapine than risperidone.
Conclusions: Results from this study suggest that providing physicians with evidence presented as a simple average treatment effect may not be appropriate. Although the PS and PPS methods both have the ability to identify HTE they answer notably different questions. The PS method is beneficial in identifying if physicians have the appropriate evidence they need to individualize therapy. While, the PPS method is beneficial in identify the subgroup of patients most likely to benefit from a treatment.
Candidate for the Lee B. Lusted Student Prize Competition