(306) Development of matching procedures to identify comparators for real-world comparative evaluations of antipsychotics

Background: Identifying appropriate comparators for a new antipsychotic (AP) entering the market to evaluate comparative effects is challenging, as most of the new AP users exposed to comparator drugs historically and are not eligible with traditional incident new-user cohort design. The prevalent new-user cohort design provided an alternative method using all patients rather than only the treatment-naïve ones, and identified comparators with matched baseline characteristics and treatment patterns. However, further extensions of matching are needed in schizophrenia research as cumulative exposures to APs and time-related bias should be taken into account. Moreover, how to efficiently identify comparators through large number of potential comparators within big databases worths exploration.

Objectives: To develop matching procedures for identifying an appropriate comparator group in schizophrenia research and enable comparative evaluations of APs with large-scale databases, using risk assessment of cardiovascular events between initiating paliperidone palmitate once-monthly injection (PP1M) and initiating a new oral second-generation AP (SGA) as an example.

Methods: Data from Taiwan’s National Health Insurance Research Database were used to identify a cohort of adult patients with schizophrenia who received SGAs from 2013 to 2018. Patients who initiated PP1M were enrolled and they were firstly matched with other patients who initiated a new oral SGA based on calendar time of cohort entry (the date of initiating PP1M or the new oral SGA), recent exposure to SGAs, visit type, and history of cardiovascular hospitalizations. Next, conditional logistic regression was performed to obtain the propensity score (PS) of initiating either treatment, based on baseline characteristics including demographics, treatment and medical history. Finally, each patient initiating PP1M was matched with a comparator, using nearest neighbor matching with a caliper. Matching performance was assessed using standardized mean differences (SMD) between groups.

Results: Of the 182,592 users of SGAs, 11,023 patients initiating PP1M were identified. In the first step for matching key characteristics, a set of potential comparators for each PP1M user was generated efficiently for the following PS matching, and finally most patients (92%) were identified SGA comparators. The comparability between groups was well achieved, with SMDs of all baseline covariates < 0.1. The well balanced time of cohort entry between groups indicated minimized time-related bias.

Conclusions: The matching procedures developed in this study can be used to efficiently identify appropriate comparator groups in schizophrenia research and reduce bias.