Drawing Reproducible Conclusions from Observational Clinical Data with OHDSI

Area of Science:

• Health Informatics
• Observational Research
• Open Science

Background:

• Observational research literature is plagued by publication bias and contradictory findings.
• The Observational Health Data Sciences and Informatics (OHDSI) initiative aims to enhance research reproducibility.
• Open science principles are central to OHDSI's approach.

Purpose of the Study:

• To improve the reproducibility of observational health research.
• To leverage a global federated data network for large-scale studies.
• To implement rigorous methods for addressing confounding and systematic error.

Main Methods:

• Utilized an international federated network of electronic health records and claims data covering nearly 10% of the global population.
• Employed a common data model with standardized schema and vocabulary mappings for data harmonization.
• Applied large-scale propensity score adjustment, negative/positive control hypotheses, and multi-database assessments for robust analysis.
• Ensured complete openness of protocols, software, models, parameters, and results for external verification.

Main Results:

• Generated findings on hypertension treatment now being integrated into clinical practice.
• Produced rigorous COVID-19 studies informing treatment decisions and disease characterization.
• Estimated comparative treatment effects and predicted serious complication likelihood for COVID-19 patients.

Conclusions:

• OHDSI effectively implements open science practices and reproducible research methodologies.
• The initiative has yielded significant results in critical health areas, including hypertension and COVID-19.
• OHDSI's approach demonstrates the potential for large-scale, reproducible observational research.