ORIGINAL RESEARCH article
Front. Pharmacol.
Sec. Translational Pharmacology
Cross-Design Evidence: Mendelian Randomization Status Alone Explains the Classification of Phase III Drug Outcomes Across Ten Disease Domains
Independent Researcher, Cambridge, United States
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Abstract
Background: Mendelian randomization (MR) evidence is associated with drug success. We tested whether observational evidence adds to MR status in classifying Phase III outcomes, using a preregistered rule comparing the two evidence types, and performed an exploratory analysis of the misclassified families. Methods: We assembled 41 mechanism families (a drug target or exposure paired with a disease indication) across ten disease domains. Observational and MR effect sizes were standardized to Cohen’s d with a fixed d = 0.10 threshold, and a deterministic rule classified each family from the observational and MR evidence pattern: a non-trivial observational association with null MR evidence was classified as an expected failure, and supportive MR evidence as an expected success. The rule and threshold, the extension families, and two prospective predictions were preregistered before their outcomes were examined; selection of the original 27 families and the exploratory failure-mode taxonomy were not. Five families whose genetic leg is a variant–disease association are reported apart from MR-instrumented families. Results: Under the registered analysis, 24/32 families were classified correctly (75.0%; outcome-permutation p = 0.005 at family and instrument-cluster level): 18/22 in the original domains and 6/10 in the blind extension, the latter not significant alone (permutation p = 0.55) and not independent confirmation. MR-only and the cross-design rule produced identical classifications across all 32 scored families, because 31 of 32 have non-trivial observational support. Repeating the analysis with source-reported values changed no classification (24/32). Excluding four families whose MR value is not a published estimate, a set specified with outcomes known, gave 21/28 (75.0%; descriptive permutation p = 0.011). The eight misclassified families split by MR status: five had null MR evidence yet an approved drug, and three had supportive MR evidence yet a failed drug. Conclusions: The observational leg added no information about outcomes beyond MR status, consistent with the genetic-support findings of Nelson et al. and Minikel et al. Set against MR and trial evidence, the observational leg yields exploratory hypotheses about why an MR-based classification fails. The framework is an exploratory proof of concept requiring prospective validation on unseen readouts before use in drug-development decisions.
Summary
Keywords
cross-design concordance, Drug Development, Drug target validation, Evidence triangulation, Mendelian randomization, Observational evidence, pharmacogenomics, phase III trials
Received
09 July 2026
Accepted
16 October 2026
Copyright
© 2026 Tower. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Elliot Tower
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.