I'm proud to report a new publication. In lieu of a long discussion about the hurdles (and immense potential) of integrating real-world data to drive medical evidentiary advances, statistical techniques and causal frameworks deserving to be more widely adopted, and how our recent publication tries to sing these tunes, I will instead try to give a TLDR version:
Diagnostics exist to improve decision-making
We sought to better understand the decisions made without the diagnostic, to better infer causality of outcomes with the diagnostic
We utilized real world data from of standard of care treatment
We “borrowed” causal inference techniques from epidemiology and economics (i.e. propensity analyses)
We trained a risk score using existing clinical and laboratory features to predict differential drug class survival, and made this model compete with the AR-V7 biomarker. AR-V7 still out-performed the "poor man's version" of the biomarker.
Graf RP, Hullings M, Barnett ES, Carbone E, Dittamore R, Scher HI. Clinical Utility of the Nuclear-localized AR-V7 Biomarker in Circulating Tumor Cells in Improving Physician Treatment Choice in Castration-resistant Prostate Cancer. European Urology. 2020. PubMed
I will say that this was definitely one of the more stimulating and challenging projects I’ve had in recent years (not to mention the 6 peer-reviewers and staff statistical reviewer. European Urology doesn’t mess around!). It was a great experience and an honor to work closely with the team at MSKCC, and I have to thank Ryan Dittamore and Howard Scher for supporting and seeing the value in this project. We hope that these methods of causal inference and counterfactual testing of real-world evidence will become more common in the years to come, becoming an integral part of biomarker evidentiary builds.