Robust-ODAL: Learning from heterogeneous health systems without sharing patient-level data

Abstract

Electronic Health Records (EHR) contain extensive patient data on various health outcomes and risk predictors, providing an efficient and wide-reaching source for health research. Integrated EHR data can provide a larger sample size of the population to improve estimation and prediction accuracy. To overcome the obstacle of sharing patient-level data, distributed algorithms were developed to conduct statistical analyses across multiple clinical sites through sharing only aggregated information. However, the heterogeneity of data across sites is often ignored by existing distributed algorithms, which leads to substantial bias when studying the association between the outcomes and exposures. In this study, we propose a privacy-preserving and communication-efficient distributed algorithm which accounts for the heterogeneity caused by a small number of the clinical sites. We evaluated our algorithm through a systematic simulation study motivated by real-world scenarios and applied our algorithm to multiple claims datasets from the Observational Health Data Sciences and Informatics (OHDSI) network. The results showed that the proposed method performed better than the existing distributed algorithm ODAL and a meta-analysis method.

Leverage Real-world Longitudinal Data in Large Clinical Research Networks for Alzheimer’s Disease and Related Dementia (ADRD)

“…Secondly, ODAC is based on the fact that we treat the pooled analysis as the gold standard method, which essentially requires data to be homogenous distributed across sites. However, the effect sizes as well as the baseline hazard function at each site can be different in practice 31 . To efficiently synthesize evidence from heterogeneous clinical sites in large CRN, we are planning to extend the current distributed algorithms by allowing site-specific hazard function and effect sizes in the future.…”

Comparative Effectiveness of Knowledge Graphs- and EHR Data-Based Medical Concept Embedding for Phenotyping

“…In addition, EHR co-occurrence statistics are often specific to a local medical institution due to differences in local practices and the difficulties in patient data sharing. Co-occurrence statistics should be calculated across multiple institutions to learn more efficient and generalizable MCEs while minimizing bias of EHR from each institution [36][37][38] . The OMOP CDM can be leveraged to accomplish this since it provides a mechanism to convert local data models to its common data model.…”

Exploring the Feasibility of Using Real-World Data from a Large Clinical Data Research Network to Simulate Clinical Trials of Alzheimer’s Disease

“…Therefore, it requires that the data are homogeneously distributed across sites. However, when there exists heterogeneity in the distribution of data across sites, the pooled analysis may not be a gold standard and will require correcting the model to address the heterogeneity 34 . Taking this into consideration, we plan to extend our method to handle heterogeneity across clinical sites by allowing site-specific effects and covariates.…”