Pharmacology Epidemiology
When observational, genetic, and experimental evidence agree, the mechanism is probably real. When they disagree, the pattern of disagreement classifies what went wrong.
Two Kinds of Drug Failure
Drug Failures Partition into Two Kinds: Cross-Design Evidence Diagnosis Across Ten Disease Domains
Zenodo 10.5281/zenodo.21227354
MR evidence predicts drug success, but when a drug fails despite genetic support — or succeeds without it — the MR screen alone cannot explain why. A deterministic cross-design classification rule comparing observational, MR, and RCT evidence across 41 mechanism families in ten disease domains correctly classifies 24/32 evaluable families (p = 0.004). The eight misclassified families split by MR effect size into two structurally distinct groups: five with null MR (effector-neutralization and mechanism-bypass) and three with causal MR (translation gaps and exposure mismatches). Each group implies a different pipeline response. Pre-registered accuracy: 18/22 (81.8%, CI 61.5–92.7%, p < 0.001). Extension accuracy: 6/10 (60.0%, not significant in isolation). The observational leg carries zero independent predictive information — a structural consequence of the fact that mechanisms reaching Phase III have non-trivial observational support by construction.
Ecological Bias in Multi-Site COVID-19 Studies
Ecological Bias in Multi-Site COVID-19 Studies: Two Failure Modes in 42 Million Patient Records
Zenodo 10.5281/zenodo.21252372
Using 42 million individual COVID-19 patient records from two countries, two failure modes of ecological bias are identified. First, the signal disappears: site-level analysis of 38 million CDC records cannot detect the relationship between age and mortality (p = 0.12), while individual-level analysis finds a 9.9-fold odds ratio (p < 10^-300). Second, the signal distorts: site-level analysis of 4 million Mexican records detects a significant relationship (p < 0.0001), but the ecological effect size bears no resemblance to the individual-level truth. Applied to the 4CE consortium’s 22,000-patient neurological COVID-19 dataset (21 hospitals, 6 countries), meta-analytic methods diverge by 350-fold on the pooled effect (I² = 99.8%), and a multivariate consistency test detects coordinated cross-site heterogeneity (z = 25, p < 0.0001) that standard tests miss entirely.