Qualifying NaS Core Through a TCGA-BRCA Survival Study

Before NaS Core supports a novel discovery study, it must demonstrate that it can carry a scientific question through governed data access, deterministic cohort construction, prespecified analysis, diagnostics, review, and release.

NAS-BRCA-001 uses a well-established relationship as a demanding systems test. The scientific direction is known. The challenge is whether the complete research record remains traceable, reproducible, and honest when the analysis encounters missing data, exclusions, warnings, or failed assumptions.

The study is a retrospective secondary analysis of adults in the Genomic Data Commons TCGA-BRCA project. It requests open clinical fields only. No controlled genomic files are used.

The primary comparison groups pathologic stages I and II as early stage, and stages III and IV as advanced stage. Overall survival is derived from death or the last documented follow-up. Age at diagnosis is the prespecified covariate.

• Primary model: age-adjusted Cox proportional hazards regression.
• Primary estimand: adjusted hazard ratio for advanced versus early stage with a 95 percent confidence interval.
• Supporting analyses: Kaplan-Meier estimates, log-rank testing, categorical stage analysis, and five prespecified sensitivity analyses.
• Missing-data strategy: complete-case analysis with no imputation and a retained exclusion record.

The analysis is required to test proportional hazards, examine the functional form of age, identify influential observations, retain convergence warnings, and report the number of participants and deaths in each exposure group.

The primary estimate cannot be interpreted if either group has fewer than ten deaths, the model fails to converge, or the stage coefficient cannot be estimated. Prespecified sensitivity analyses remain part of the record even when they weaken, contradict, or complicate the primary result.

The workflow verifies source manifests and checksums before cohort construction. It records the selected diagnosis, original stage value, survival fields, exclusions, code revision, environment, parameters, and hashes for generated artifacts.

Outcome-bearing artifacts remain outside the code repository in controlled external storage. The public website will not expose numerical results until the review and release gates are complete.

This retrospective observational analysis cannot establish that stage causes mortality. TCGA recruitment, treatment era, eligibility, and follow-up can limit generalizability. Overall survival does not distinguish breast cancer death from death due to other causes.

The early-versus-advanced contrast simplifies disease stage, AJCC staging editions can differ across participants, and excluding records with unusable or zero-day survival can introduce additional limitations. These constraints remain part of the final interpretation.