Does adjusting for tumor proliferation bias prognostic gene expression estimates? Preregistered analysis code, decision rules and results

Abstract

R analysis code, preregistered decision rules, result files, run logs and figures for a study of what omitting tumor proliferation does to prognostic gene expression estimates. For every expressed gene in a cohort, two Cox models are fitted for overall survival, one adjusted for age, sex and stage and one that adds a standardized eight-gene proliferation score. The ratio of the two hazard ratios is the shift, and the shift is regressed on the gene's Spearman correlation with the score. Discovery used TCGA-LIHC, with replication in GSE14520. A confirmatory analysis across further cancer types was registered on the Open Science Framework (doi:10.17605/OSF.IO/X5DCF) on 20 September 2026 and run on 21 September 2026. Ten cancer types met the registered inclusion rule, and both registered hypotheses were refuted. The file results/CJ05_decision.csv records both verdicts as REFUTED. The structure found within that refutation is labeled post hoc wherever it appears, and it is registered separately (doi:10.17605/OSF.IO/N8DW6) for testing in cohorts that have not been examined; no analysis covered by that second registration is in this release. A survey of 50 recent prognostic gene expression papers in hepatocellular carcinoma, drawn under a fixed seed from a frame of 424 records with its design fixed before any paper was read, records how often such studies adjust for proliferation at all. Every analysis is labeled preregistered, post hoc or exploratory in the code and in its output files, and README.md maps every result file prefix to its stage and to that label. Every script writes its own decision rules to a result file on each run, so an edit made after the fact is visible in the deposited output rather than only in the code history. All data are public: TCGA from UCSC Xena, GSE14520 from GEO, gene sets from MSigDB. No new human or animal data were generated and no identifiable data are included.