Development and Evaluation of a Global Propensity Score for Data Mining with Tree-Based Scan Statistics

The aims of this methods project are to develop and evaluate candidate global propensity scores for application with the propensity score cohort matched design and tree-based scan statistics. The examples used in this project are for methods development purposes and are not part of a regulatory evaluation of any medical product or safety issue.

The Workgroup will compare the relative performance of candidate global propensity scores which could be applied generally in cohort studies involving tree-based scan statistics by evaluating 4 case studies for drug safety signal detection. The selected case studies have well characterized safety profiles. After evaluating the relative performance of candidate global propensity scores, the Workgroup will conduct clinical and epidemiologic review of alerts observed with the pre-specified primary global propensity score to adjust for confounding.

The Workgroup will use the CDER-curated Multi-Level Clinical Classification Software (MLCCS) outcome tree. This tree contains thousands of hierarchically grouped adverse events and is based on International Classification of Diseases 9^th edition (ICD-9-CM) codes. Further curation may occur for specific drug evaluation examples based on the outcomes that are closely related to the reason for prescribing the drug.

This protocol was posted for public comment from August 10, 2018 through September 7, 2018. A revised version was approved for implementation by FDA on September 26, 2018. A second revised version was approved for implementation by FDA on March 4, 2019, a third revised version was approved for implementation by FDA on May 9, 2019. A log of changes is included in the revised protocol (v4).