We conducted meta-analysis on coefficients (e.g., log-hazard ratios [log-HRs]) by setting weights as the inverse of the variance across the four cohorts with neuroinflammation-related conditions, and across the four cohorts without neuroinflammation-related conditions. First, we estimated the HRs, and the false discovery rates (FDRs) by using the Benjamini-Hochberg procedure in patients with neuroinflammation-related conditions for all drugs.25 Second, we used z-test to compare the log-HRs between individuals with and without neuroinflammation-related conditions in a quasi-difference-in-difference setting. For a HR <1 (or >1) in individuals with neuroinflammation-related conditions, the null hypothesis was that the HR was lower (or higher) in individuals with neuroinflammation-related conditions compared to without. We used 0.05 as the threshold for one-sided P-value. We selected drugs that had: 1) FDR <0.05 for testing HR =1 in metanalysis of neuroinflammation-related cohorts; and 2) P-value <0.05 for comparing HRs between individuals with and without neuroinflammation-related conditions. All analyses were conducted in R.
