Observational data can reveal treatment use, outcomes, safety, effectiveness, and patient risk in routine care. But differences in populations, measurement, treatment selection, and study design can also produce misleading results. OHDSI researchers therefore develop methods—and empirically evaluate how those methods perform—rather than relying on methodological convention alone.
OHDSI methods emphasize prespecified study designs and transparent diagnostics. Depending on the question, researchers may use active comparators, large-scale propensity scores, negative controls, empirical calibration, phenotype evaluation, covariate-balance checks, and replication across databases. These diagnostics expose assumptions and potential sources of error. If a result fails prespecified checks, it should not be treated as reliable evidence.
The goal is not to declare observational data trustworthy by default. It is to evaluate whether a particular study is fit for interpretation by making its assumptions, diagnostics, and uncertainty visible. Confidence increases when findings remain consistent across prespecified analyses, data sources, and populations—and decreases when those checks reveal important limitations.
