Meta-analysis of multi-jurisdictional health administrative data from distributed networks approximated individual-level multivariable regression
Dheri AK, Kuenzig ME, Mack DR, Murthy SK, Kaplan GG, Donelle J, Smith G, Benchimol EI. J Clin Epidemiol. 2022; May 19 [Epub ahead of print]. DOI: https://doi.org/10.1016/j.jclinepi.2022.05.006
Objective — Compare meta-analysis in a distributed network to individual-level analysis for assessment of time trends of health services utilization with health administrative data.
Study Design and Setting — We used administrative data from Ontario, Canada to analyze temporal trends in pediatric inflammatory bowel disease health services use. Beta coefficients were obtained using negative binomial, logistic, and Cox proportional hazards regression models. We replicated the individual-level analyses in each Ontario Local Health Integration Network (LHIN), then meta-analyzed aggregate trends using both fixed and random effects meta-analysis. We compared the pooled estimates of effect with individual-level analysis.
Results — Beta coefficients, summary effect estimates, and 95% confidence intervals from the meta analysis of data from distributed networks were not different than those from individual-level data, regardless of meta-analytic approach used. For example, the 5-year odds ratio of colectomy in ulcerative colitis using individual-level analysis was 0.978 (95%CI 0.950 to 1.007) compared to distributed network fixed effects meta-analysis: 0.982 (95%CI 0.950 to 1.015), and random effects meta-analysis: 0.982 (95%CI 0.950 to 1.015).
Conclusion — Meta-analysis of multi-jurisdictional estimates were similar to estimates obtained from individual-level analysis. This method is a valid alternative for analysis of multi-jurisdictional data when individual-level data cannot be shared.
View full text