(P-213) Surveillance of Postmarketing Risk-mitigation Behavior Using Tests of Model Misspecification in a Statistical Process Control Framework

Background: Surveillance of high-risk drug prescriptions, and their corresponding risk mitigating behaviors, may provide early warning of unforeseen issues with risk management. However, analysis approaches to conduct this surveillance using real-world data are underdeveloped.

Objectives: We propose a method for conducting surveillance on one or multiple outcomes of prescribing behavior using administrative claims data.

Methods: We modified Pignatiello and Runger’s MC1 statistical process control method to include elements of White’s Information Matrix test. We use the misspecification of a previously fit model on newly collected data to determine if the new data shows a change in prescribing behavior. We performed simulation studies to compare the performance of our method against other options. We applied this approach in two settings: 1) behaviors around new prescriptions and 2) behaviors around drugs that are continuously prescribed (i.e., chronic conditions). Our examples use administrative claims data (1/2012-6/2019) from the Colorado All Payers Claims Database. We fit a joint mixed data model on the first prescriptions of tapentadol extended-release (an opiate with an FDA REMS) for 2,702 patients. Mixed data includes both continuous and categorical outcomes. Our joint outcomes were the prescribing rate and the proportions of prescriptions with appropriate dosing, a risk mitigating behavior. In a second example, we modeled repeated prescriptions from 2,878 patients receiving clozapine, an antipsychotic with a REMS. Our outcome was the proportion of patients receiving a neutrophil count prior to filling each prescription. This example addresses how to analyze and conduct surveillance on data repeatedly collected on individual patients.

Results: Simulation studies showed that our method detects misspecification fastest at all levels tested against other potential methods. Our method found two periods in March 2015 (p < 0.005) and April 2017 (p < 0.005) of increased prescribing rates and increased proportions of appropriate dosing of tapentadol extended-release. Our method also detected a period in February 2019 in which the proportion of people receiving neutrophil counts prior to filling a clozapine prescription decreased 12% (p < 0.005).

Conclusions: Novel application of statistical process control to prescription surveillance can detect changes in prescribing behavior in a timely fashion with readily available administrative data. This tool may be useful for regulatory agencies to independently monitor for emerging risks and prescribing trends to complement assessments from pharmaceutical companies.