Using the Observational Medical Outcomes Partnership Common Data Model for a multi-registry intensive care unit benchmarking federated analysis: lessons learned

Abstract

Objective: Federated analysis is a method that allows data analysis to be performed on similar datasets without exchanging any data, thus facilitating international research collaboration while adhering to strict privacy laws. This study aimed to evaluate the feasibility of using federated analysis to benchmark mortality in 2 critical care quality registry databases converted to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM), describing challenges to and recommendations for performing federated analysis on data transformed to OMOP CDM. Materials and Methods: To identify as many challenges as possible and to be able to complete the benchmarking phase, a 2-step approach was taken during implementation. The first step was a naive implementation to allow challenges to surface naturally; the second step was developing solutions for the encountered challenges. Expected patient mortality risk was calculated by applying the Acute Physiology and Chronic Health Evaluation II (APACHE II) model to data from OMOP CDM databases containing adult ICU encounters between July 1, 2019 and December 31, 2022. An analysis script was developed to calculate comparable, registry level standardized mortality ratios. Challenges were recorded and categorized into predefined categories: "data preparation," "data analysis plan," and "data interpretation." Challenges specific to the OMOP CDM were further categorized using published steps from an existing generic harmonization process. Results: A total of 7 challenges were identified, 4 of which were related to data preparation, 1 to data analysis, and 1 to data interpretation. Out of all 7 challenges, 4 stemmed from decisions made during the implementation of OMOP CDM. Several recommended solutions were distilled from the naive approach. Discussion: Federated analysis facilitated by a CDM is a feasible option for critical care quality registries. However, future analysis is influenced by decisions made during the CDM implementation process. Thus, prior publication of data dictionaries and the use of metadata to communicate data handling and data source classification during CDM implementation will improve the efficiency and accuracy of subsequent analysis.