Meta-Analyzing Diagnostic and Predictive Modeling Studies

While meta-analysis for randomized controlled trials (RCTs) is a well-established procedure, meta-analyzing scores of studies that build observational, diagnostic predictive models is considerably harder. These studies usually report confusion matrices and AUC/ROC metrics rather than more conventional mean differences or odds ratios, making the standard techniques for meta-analysis fairly unusable.

Fortunately, Debray and team recently published an insightful paper that provides guidelines, estimators and approaches for conducting meta-analysis of predictive modeling studies. It is available at the link below, and we recommend that anyone working in the predictive analytics area checks it out!

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6728752/