OBJECTIVES: Distributed data networks are an expanding focus of drug safety research.  However, interpreting results of distributed analyses is challenging when treatment effects are heterogeneous.  We assessed heterogeneity of treatment effects and explored factors influencing heterogeneity using experimental results from the Observational Medical Outcomes Partnership (OMOP).  METHODS: OMOP evaluated risk of eight health outcomes across nine drug groups, replicating analyses across eight data sources using a common data model.  We focused on the OMOP propensity score analyses and assessed heterogeneity using meta-analysis and I-squared statistics.  Plots of the relationship between influence on overall results and contribution to overall heterogeneity identified influential data sources.  Summary-level data source characteristics were examined to identify potential factors with high variability that could be influencing results.  Exploratory meta-regression further assessed the relationship of summary-level characteristics with heterogeneity using risk of bleeding with warfarin and un-related negative controls (antibiotics, benzodiazepines, and tricyclic antidepressants).  RESULTS: Heterogeneity, as measured by the I-squared statistic, ranged from 0-99% across the drug-outcome pairs studied.  In general, heterogeneity generally was lower for analyses of angioedema (0-71%) and aplastic anemia (0-84%) and higher for acute liver injury (74-91%) and bleeding (78-99%).  Plots of the relationship between influence on overall results and contribution to heterogeneity illustrated limited data source clustering, which is an indicator that heterogeneity was not driven by a single dataset but rather was unique to the drug and outcome being assessed.  Exploratory meta-regression identified many variables influencing heterogeneity, but was unable to accurately identify specific factors that were most influential.  CONCLUSIONS: Distributed data network drug safety analyses are challenging to interpret in the face of heterogeneity.  A formal assessment of heterogeneity should be conducted, and approaches illustrated here can be useful to explore heterogeneity.  Further evaluation of the value of patient-level as opposed to summary-level data in heterogeneity assessment is needed.