Pitfalls of Administrative Censoring in Survival Models with Time-Indexed Inputs

Recent work has argued normatively, on synthetic data, that evaluating survival models by discrimination alone (concordance index) yields systematically misleading model comparisons, because the metric ignores calibration and time-dependent accuracy. Whether this matters for real, published, non-clinical models has not been tested. We reproduce three published survival-ML models across three structurally distinct domains -- hard-drive failure prediction, peer-to-peer credit default, and user disengagement on digital platforms -- validate our instrument against the anchor paper's own synthetic experiment, and test five pre-registered hypotheses under a Holm-corrected family-wise error rate. Three of five reject (though one pre-registered threshold clears by a narrow margin). A model reproducing the published literature's discrimination almost exactly (C = 0.9595 vs. 0.958 reported) fails a formal calibration test at p < 0.001; a broad feature-ablation search finds no single attribute responsible for its discrimination, so the calibration failure is not a trivial shortcut artifact. A lender's estimated default risk is biased upward by roughly two percentage points, growing to nearly four in the riskiest segment, when loan prepayment is treated as non-informative censoring rather than a competing risk. A platform's churn model shows probability estimates that degrade with the horizon even as global discrimination stays within the pre-registered C-index band. A direct test of whether metric choice inverts model preference does not reject, though with limited power given two to three models per domain; the failure mode we document is better characterized as misplaced confidence in a chosen model than as choosing the wrong one. We release a pre-registered evaluation harness with full code and an annotated notebook, so these results can be verified independently and the audit extended.

Objective: Survival analysis is central to medical prediction, yet large language models (LLMs) are rarely used as end-to-end survival models because censoring prevents straightforward supervised fine-tuning. Here we present LLMSurvival, a framework that enables censoring-aware survival analysis with unmodified LLMs operating directly on tabular clinical data. Materials and Methods: LLMSurvival reformulates time-to-event prediction as pairwise ranking among comparable subjects, and derives test-time risk by aggregating comparisons against anchor individuals from the training cohort. Results: Across two clinical tasks (ICU mortality prediction in MIMIC-IV and fragility fracture prediction in a NewYork-Presbyterian/Weill Cornell Medicine cohort), LLMSurvival improves overall concordance over Cox proportional hazards modeling by 3.1% for ICU mortality and 0.5% for fracture risk, 2.1% on average for ICU mortality and 2.8% for fracture risk over three established deep learning survival models. Discussion: The results show that survival modeling with censoring can be made compatible with LLM fine-tuning through comparison-based reformulation. The framework demonstrates high portability and superior performance over expert curated scores like SAPS-II and FRAX scores across diverse clinical context. Furthermore, the framework supports local deployment, as compact, publicly available base models provide sufficient performance. Conclusion: The LLMSurvival framework serves as a proof of concept for an integrated, censoring-conscious approach to survival analysis via LLMs.

Whole-slide survival models commonly provide risk rankings without calibrated statements about individual event times. We evaluate fixed-cutoff drcosarc, a post-hoc conformal wrapper for discrete-time multiple-instance learning survival heads using frozen UNI2-h representations, in an internal 18-configuration sweep across five TCGA cohorts and an external five-configuration evaluation across three CPTAC cohorts. We distinguish configuration--fold--split summaries of the inverse-probability-of-censoring-weighted (IPCW) estimate and median lower predictive bound (LPB) from a hierarchy-aware patient-ensemble estimand of the mean drcosarc--naive LPB difference. At $α=0.1$, the drcosarc IPCW estimate was nearest 0.90 in KIRC, LUAD, and STAD. Patient-ensemble drcosarc--naive intervals excluded zero in KIRC, KIRP, STAD, UCEC, and CPTAC-CCRCC, but included zero in internal LUAD, CPTAC-LUAD, CPTAC-UCEC, and the internal LUSC extension. In a 20-replicate low-censoring semi-synthetic setting with known event times, drcosarc empirical coverage was 0.9129 [0.9053, 0.9207]. An exploratory analysis supported a head-error-by-censoring interaction within that data-generating process. In a two-cohort ABMIL sensitivity analysis, increasing the hazard grid to $K=16$ raised localized marginal IPCW estimates above the prespecified 0.87 threshold and yielded positive paired LPB differences, although worst-group estimates remained below 0.87. Overall, performance was cohort dependent, and its interpretation changed with the patient-level unit, estimand, and censoring assumptions.