Table 4 Performance of non-experimental methods according to the EU-ADR and OMOP reference sets with a comparison baseline a TrialProbe subset that consists of all entries that are statistically significant according to the Fisher exact test instead of an empirical Bayes odds ratio threshold
From: Using public clinical trial reports to probe non-experimental causal inference methods
Non-experimental | OMOP (n=140) | EU-ADR (n=32) | Fisher exact test TrialProbe Subset (n = 293) | |||
|---|---|---|---|---|---|---|
Method | % Concordant | % Recovered | % Concordant | % Recovered | % Concordant | % Recovered |
Unadjusted Cox | 50 % | 45 % | 35 % | 22 % | 53 % | 42 % |
PSM Cox | 47 % | 20 % | 60 % | 9 % | 59 % | 30 % |
IPSW Cox | 65 % | 65 % | 53 % | 53 % | 58 % | 27 % |