{"publication":"Preventative Health Bench","url":"https://preventativehealthbench.com","publisher":"Arcophos","updated":"2026-09-28","provenance":"Independent analytical publication. Benchmark creation and experimental results belong to their cited authors. Reported results are source-version snapshots, not new Arcophos runs or a live leaderboard.","benchmarks":[{"slug":"ukb-survival","name":"UK Biobank survival-model benchmark by Oexner et al.","shortName":"UK Biobank survival study","version":"2026 accepted journal manuscript","creators":"R. R. Oexner, R. Schmitt, H. Ahn and collaborators / King’s College London and UCL","paperDate":"2026-08-17","headline":"Ranking future risk is only one part of a prevention decision.","summary":"This published benchmarking study compares survival models across incident cardiovascular disease, breast cancer and Alzheimer’s disease, using several feature regimes. It supplies a concrete prevention-related evaluation rather than a general screening checklist. Our analysis asks how the endpoint, available information and validation design shape a model comparison. The reported outcome is risk discrimination: which people are ranked ahead of others. It does not by itself establish calibrated absolute probabilities or the benefit of an intervention. We retain the study’s name, conditions and uncertainty rather than branding its data as a new benchmark of our own.","task":{"input":"Baseline participant predictors under a specified feature regime","output":"A survival-risk score for a defined incident-disease endpoint","unit":"One participant with event/censoring time","setting":"Nested five-fold cross-validation; no single fixed public test set"},"dataOrigin":"Longitudinal UK Biobank participant information and linked health outcomes; access-controlled research data.","facts":[{"label":"Disease endpoints","value":"3","detail":"Cardiovascular disease, breast cancer and Alzheimer’s disease.","sourceIds":["ukb-methods"],"locator":"§2.2"},{"label":"Feature regimes","value":"5","detail":"Age/sex; clinical risk; PRS/metabolomics; medical history; complete.","sourceIds":["ukb-methods"],"locator":"§2.3"},{"label":"Primary model families","value":"8","detail":"Cox, Lasso, Ridge, Elastic Net, random forest, LightGBM, XGBoost and deep learning.","sourceIds":["ukb-methods"],"locator":"Model implementations"},{"label":"Validation","value":"Nested 5-fold CV","detail":"Inner tuning is separate from outer evaluation; approximate cohort counts vary after endpoint exclusions.","sourceIds":["ukb-methods"],"locator":"§2.7"},{"label":"Analysis population","value":"Approximately 240k","detail":"Selected for <20% missingness across predictor matrices, largely requiring metabolomics coverage; before endpoint-specific baseline-disease exclusions.","sourceIds":["ukb-methods"],"locator":"§§2.4–2.5"}],"metric":{"name":"Harrell’s concordance index","description":"Measures ordering of risk against observable event order among comparable pairs; the paper also reports Uno’s C.","formula":"C = concordant comparable pairs / comparable pairs, with the implementation’s tie handling","direction":"Higher is better","comparability":"Compare identical endpoint, predictor regime and validation folds. C is not percent of patients correctly diagnosed.","sourceIds":["ukb-methods"]},"workflow":[{"label":"Define incident outcomes","detail":"Exclude baseline occurrence of the corresponding endpoint.","sourceIds":["ukb-methods"]},{"label":"Specify available predictors","detail":"Fix one of the five feature regimes before comparison.","sourceIds":["ukb-methods"]},{"label":"Tune inside training folds","detail":"Use nested cross-validation and fitted preprocessing.","sourceIds":["ukb-methods"]},{"label":"Evaluate held-out ordering","detail":"Report fold-level discrimination and uncertainty for that endpoint.","sourceIds":["ukb-results","ukb-methods"]}],"slices":[],"sliceTitle":"No additive patient chart","sliceNote":"The endpoints and feature regimes reuse participants. Counting their combinations does not create independent patient populations.","results":[{"id":"cvd-clinical","title":"Cardiovascular disease with clinical-risk features","metric":"Harrell’s C","unit":"C-index","lower":0,"upper":1,"scope":"2026 accepted manuscript and Supplementary Table 3; nested CV test performance for the Clinical Risk regime.","sourceIds":["ukb-results","ukb-methods"],"locator":"Cardiovascular Disease worksheet; Clinical Risk rows","rows":[{"label":"Minimally penalised Cox-PH","value":0.719,"display":"0.719","detail":"95% CI 0.717–0.721."},{"label":"Deep learning","value":0.721,"display":"0.721","detail":"95% CI 0.719–0.722."},{"label":"XGBoost","value":0.564,"display":"0.564","detail":"95% CI 0.560–0.569."}],"note":"Study-specific implementations and tuning grids. These are discrimination values, not clinical accuracy percentages or universal algorithm rankings."}],"analysis":[{"heading":"The endpoint belongs in the result name","evidence":"The study changes endpoints and predictor sets explicitly.","interpretation":"Our inference: an algorithm ranking without its endpoint and available features is underspecified.","sourceIds":["ukb-methods"]},{"heading":"Discrimination leaves an unanswered prevention question","evidence":"The study focuses on discrimination rather than calibration.","interpretation":"Our inference: an ordering result cannot alone support an absolute-risk treatment threshold.","sourceIds":["ukb-methods"]},{"heading":"Validation units are participants, not pair counts","evidence":"The C-index compares observable pairs constructed from participant outcomes.","interpretation":"Our inference: the number of pairs should not be used as though it were the number of independent patients when discussing uncertainty.","sourceIds":["ukb-methods"]},{"heading":"A small score gap is context dependent","evidence":"The selected clinical CVD estimates differ by 0.002 between deep learning and Cox-PH.","interpretation":"Our arithmetic: 0.721 minus 0.719 equals 0.002; that difference is not a measured reduction in disease incidence.","sourceIds":["ukb-results"]}],"limitations":[{"title":"Selected source cohort and assay coverage","detail":"UK Biobank recruitment is selective. The analysis further filters for predictor availability and missingness, largely removing participants without metabolomics. Transport to another population needs separate evidence.","sourceIds":["ukb-methods"]},{"title":"No intervention endpoint","detail":"The comparison predicts disease events; it does not randomize or test a prevention policy.","sourceIds":["ukb-methods"]},{"title":"Controlled data access","detail":"The paper is open, but participant records require UK Biobank researcher approval.","sourceIds":["ukb-methods"]},{"title":"Accepted manuscript version","detail":"The journal labels this an article in press; the cited supplement is the 2026 version, not the 2025 preprint.","sourceIds":["ukb-paper"]}],"access":{"status":"Paper and aggregate supplements public; participant data restricted","license":"Article CC BY 4.0; participant data under UK Biobank access terms","restrictions":"Approved-researcher access is required for participant data. Article licensing does not grant dataset access.","url":"https://link.springer.com/article/10.1186/s40537-026-01533-2","sourceIds":["ukb-paper","ukb-methods"]},"sourceIds":["ukb-paper","ukb-methods","ukb-results"]}],"explorer":{"kind":"coverage","title":"Explore endpoint and predictor conditions","intro":"Filter selected published comparison conditions by disease endpoint. Rows describe tasks; they are not independent cohorts or new model measurements.","caution":"This is our independent analytical map of published tasks. It does not execute an evaluation, predict a model’s performance or establish clinical benefit.","sourceIds":["ukb-paper","ukb-methods","ukb-results"],"rows":[{"label":"Cardiovascular disease · Clinical Risk","category":"Cardiovascular disease","input":"Baseline Clinical Risk predictors","output":"Incident-event risk ordering","metric":"Harrell’s C; Uno’s C","constraint":"Nested CV within UK Biobank; no intervention effect or calibrated threshold claim.","benchmarkSlug":"ukb-survival","sourceIds":["ukb-methods","ukb-results"]},{"label":"Cardiovascular disease · Complete","category":"Cardiovascular disease","input":"Baseline Complete predictors","output":"Incident-event risk ordering","metric":"Harrell’s C; Uno’s C","constraint":"Nested CV within UK Biobank; no intervention effect or calibrated threshold claim.","benchmarkSlug":"ukb-survival","sourceIds":["ukb-methods","ukb-results"]},{"label":"Cardiovascular disease · Age & Sex","category":"Cardiovascular disease","input":"Baseline Age & Sex predictors","output":"Incident-event risk ordering","metric":"Harrell’s C; Uno’s C","constraint":"Nested CV within UK Biobank; no intervention effect or calibrated threshold claim.","benchmarkSlug":"ukb-survival","sourceIds":["ukb-methods","ukb-results"]},{"label":"Breast cancer · Clinical Risk","category":"Breast cancer","input":"Baseline Clinical Risk predictors","output":"Incident-event risk ordering","metric":"Harrell’s C; Uno’s C","constraint":"Nested CV within UK Biobank; no intervention effect or calibrated threshold claim.","benchmarkSlug":"ukb-survival","sourceIds":["ukb-methods","ukb-results"]},{"label":"Breast cancer · Complete","category":"Breast cancer","input":"Baseline Complete predictors","output":"Incident-event risk ordering","metric":"Harrell’s C; Uno’s C","constraint":"Nested CV within UK Biobank; no intervention effect or calibrated threshold claim.","benchmarkSlug":"ukb-survival","sourceIds":["ukb-methods","ukb-results"]},{"label":"Breast cancer · Age & Sex","category":"Breast cancer","input":"Baseline Age & Sex predictors","output":"Incident-event risk ordering","metric":"Harrell’s C; Uno’s C","constraint":"Nested CV within UK Biobank; no intervention effect or calibrated threshold claim.","benchmarkSlug":"ukb-survival","sourceIds":["ukb-methods","ukb-results"]},{"label":"Alzheimer’s disease · Clinical Risk","category":"Alzheimer’s disease","input":"Baseline Clinical Risk predictors","output":"Incident-event risk ordering","metric":"Harrell’s C; Uno’s C","constraint":"Nested CV within UK Biobank; no intervention effect or calibrated threshold claim.","benchmarkSlug":"ukb-survival","sourceIds":["ukb-methods","ukb-results"]},{"label":"Alzheimer’s disease · Complete","category":"Alzheimer’s disease","input":"Baseline Complete predictors","output":"Incident-event risk ordering","metric":"Harrell’s C; Uno’s C","constraint":"Nested CV within UK Biobank; no intervention effect or calibrated threshold claim.","benchmarkSlug":"ukb-survival","sourceIds":["ukb-methods","ukb-results"]},{"label":"Alzheimer’s disease · Age & Sex","category":"Alzheimer’s disease","input":"Baseline Age & Sex predictors","output":"Incident-event risk ordering","metric":"Harrell’s C; Uno’s C","constraint":"Nested CV within UK Biobank; no intervention effect or calibrated threshold claim.","benchmarkSlug":"ukb-survival","sourceIds":["ukb-methods","ukb-results"]}],"parameters":[]},"references":[{"id":"ukb-paper","title":"Comprehensive benchmarking of machine learning methods for risk prediction modelling from large-scale survival data: a UK Biobank study","organization":"Oexner et al. / King’s College London and UCL","url":"https://link.springer.com/article/10.1186/s40537-026-01533-2","note":"Accepted peer-reviewed article; results differ in some uncertainty estimates from the 2025 preprint.","locator":"Abstract; publication information","version":"2026-08-17; article in press"},{"id":"ukb-methods","title":"UK Biobank survival benchmark: accepted manuscript","organization":"Oexner et al. / Journal of Big Data","url":"https://link.springer.com/content/pdf/10.1186/s40537-026-01533-2_reference.pdf","note":"Complete methods, endpoints, nested cross-validation and limitations.","locator":"§2.2–2.9; §3.1; Discussion","version":"2026-08-17 accepted manuscript"},{"id":"ukb-results","title":"UK Biobank survival benchmark: Supplementary Table 3","organization":"Oexner et al. / Journal of Big Data","url":"https://media.springernature.com/original/springer-static/esm/art%3A10.1186%2Fs40537-026-01533-2/MediaObjects/40537_2026_1533_MOESM3_ESM.xlsx","note":"Published workbook containing condition-specific model results and confidence intervals.","locator":"Cardiovascular Disease sheet; Clinical Risk rows","version":"2026 journal supplement"}]}