A reproducibility benchmark of transcriptomic prognostic signatures in clear-cell renal cell carcinoma: protocol, feasibility assessment, and a survival-availability obstacle

Protocol and feasibility assessment for a systematic review and independent multicohort benchmark of published transcriptomic prognostic signatures for overall survival in clear-cell renal cell carcinoma (ccRCC). IMPORTANT: no comparative outcome analysis has been performed. This record documents what was planned, what was found to be feasible, and one obstacle that constrains the study design. It contains no C-index, no hazard ratio and no pooled estimate, and none exists. Results. A frozen PubMed query retrieved 489 records, 481 after deduplication. Screening against prespecified eligibility criteria excluded 177 with recorded reasons and retained 304 candidates. Six models with literally printed coefficients were staged, carrying 41 coefficients; all resolve against the GENCODE v36 annotation used by GDC matrices. Two cohorts were verified from repository metadata: TCGA-KIRC (336 usable survival records and 175 deaths, from 523 clear-cell cases of 537) and GSE29609 (39 and 17), totalling 375 patients and 192 deaths. Central finding. Public ccRCC cohorts rarely publish overall survival in their repository metadata. Of 40 GEO ccRCC expression series surveyed, none carried both a survival time and an event field in their sample characteristics. E-MTAB-1980, the external validation set used throughout this literature, records no survival column in its SDRF. GSE3538 carries only image and annotation fields. GSE22541 reports disease-free survival only and is ineligible against an overall-survival endpoint. For those cohorts the survival data exists solely in primary-report supplements and must be extracted and linked to sample identifiers by hand. The contrast with the companion PDAC study is sharp: there, four of five GEO cohorts carried usable survival directly. Feasibility verdict. The patient (375 of 300 required) and death (192 of 100) thresholds are met. The three-cohort threshold is not met, and neither is the five-model threshold: six formulas are staged but none has received the two independent human confirmations the protocol requires. The two failures differ in kind. The model criterion is a governance state; the cohort criterion is a data-availability problem that no automated work resolves. If a third cohort cannot be assembled, the protocol's own fallback applies and the output becomes a systematic review of reconstructability, external validation, reporting and risk of bias. Status. The protocol is marked 0.1.0-draft: it was adapted from the companion PDAC protocol and is not approved, and its search date must be frozen on approval rather than on the date the search ran. The analysis lock was created locked, with empty hashes and no mentor name, and never modified. The production model registry is empty. The 40-series survey is not exhaustive; ArrayExpress and ICGC were not surveyed systematically, and only PubMed was searched. Companion record: 10.5281/zenodo.22843630 (the PDAC benchmark protocol and feasibility assessment), which shares the screening rules and the formula-recovery implementation. Preprint. Not peer reviewed. This work was produced with substantial AI assistance (Claude, Anthropic). The author takes full responsibility for the content. The two independent human confirmations the protocol requires have not been performed and remain outstanding.