Preventative Health BenchDossier01
2026 accepted journal manuscript
UK Biobank survival-model benchmark by Oexner et al. ↗
Ranking future risk is only one part of a prevention decision.
UnitOne participant with event/censoring timeMeasureHarrell’s concordance index
Prevention-related risk prediction
Measure future risk in context.
Independent analysis of Oexner et al.’s UK Biobank survival-model benchmark: disease endpoints, feature regimes, nested validation, C-index results and calibration limits.
One evaluation for each incident endpoint
[1][2][3]Baseline
Participant predictors
5 feature regimes
01Cardiovascular diseaseEvent / censoring
02Breast cancerEvent / censoring
03Alzheimer’s diseaseEvent / censoring
Nested 5-fold CV
Tuning stays inside training folds.
- Outer 1Tune → evaluate
- Outer 2Tune → evaluate
- Outer 3Tune → evaluate
- Outer 4Tune → evaluate
- Outer 5Tune → evaluate
Study-design schematic; line length does not represent follow-up time. Endpoint-specific exclusions define each analysis population. [2]
01 /
Defined endpoints
Keep cardiovascular disease, breast cancer and Alzheimer’s disease separate.
02 /
Available predictors
A comparison depends on the feature regime as well as the model.
03 /
Decision boundary
Risk discrimination alone does not establish calibrated thresholds or intervention benefit.
Our analytical question
We examine a specific published benchmark of survival models for incident disease risk in UK Biobank. Its three endpoints and five feature regimes make the comparison concrete. Our original guides and condition explorer explain what discrimination measures and what remains unknown about calibration, transport and prevention decisions. This is analysis of Oexner and colleagues’ study, not a new dataset, a risk calculator or an intervention trial.
Read the original evidence closely
The benchmark, unpacked.
An original analytical tool
Explore endpoint and predictor conditions
Filter selected published comparison conditions by disease endpoint. Rows describe tasks; they are not independent cohorts or new model measurements.
9 of 9 evidence entries shown
- Input
- Baseline Clinical Risk predictors
- Output
- Incident-event risk ordering
- Measure
- Harrell’s C; Uno’s C
- Input
- Baseline Complete predictors
- Output
- Incident-event risk ordering
- Measure
- Harrell’s C; Uno’s C
- Input
- Baseline Age & Sex predictors
- Output
- Incident-event risk ordering
- Measure
- Harrell’s C; Uno’s C
- Input
- Baseline Clinical Risk predictors
- Output
- Incident-event risk ordering
- Measure
- Harrell’s C; Uno’s C
- Input
- Baseline Complete predictors
- Output
- Incident-event risk ordering
- Measure
- Harrell’s C; Uno’s C
- Input
- Baseline Age & Sex predictors
- Output
- Incident-event risk ordering
- Measure
- Harrell’s C; Uno’s C
- Input
- Baseline Clinical Risk predictors
- Output
- Incident-event risk ordering
- Measure
- Harrell’s C; Uno’s C
- Input
- Baseline Complete predictors
- Output
- Incident-event risk ordering
- Measure
- Harrell’s C; Uno’s C
- Input
- Baseline Age & Sex predictors
- Output
- Incident-event risk ordering
- Measure
- Harrell’s C; Uno’s C
This is our independent analytical map of published tasks. It does not execute an evaluation, predict a model’s performance or establish clinical benefit. [1][2][3]
Original analysis / methods and interpretation
What the score leaves unsaid.
Benchmark analysis
A concrete prevention benchmark: the UK Biobank survival-model study
Analyze a published risk-prediction comparison without inventing a new prevention dataset.
Benchmark analysis
Why the endpoint and feature set belong beside a survival-model ranking
Use the UK Biobank study to make model comparisons conditional and reproducible.
Benchmark analysis
A C-index is not a calibrated prevention threshold
Explain rank discrimination, censoring and absolute-risk calibration using the published survival benchmark.
Questions, answered
Read the result in context.
Specific tasks. Stated conditions.
Inspect every source.
Which prevention benchmark does this site analyze?
The published UK Biobank survival-model benchmarking study by Oexner and colleagues, using its 2026 accepted manuscript and matched supplements. We do not invent a separate PreventiveBench dataset.
Is Harrell’s C a patient-level accuracy percentage?
No. It measures risk-order concordance among comparable participant pairs under survival-data rules. Censoring and ties affect how it is calculated.
Does the study prove a prevention program works?
No. It compares risk prediction and computation under defined conditions. It does not test the effect of acting on predictions in an intervention trial.
Can I download the participant data from this site?
No. Public papers and aggregate supplements are linked. Participant-level UK Biobank access requires approval under its research access process.
Working tool / saved on this device
Prepare a survival-study comparison brief
Use this secondary checklist to document a run or literature comparison after inspecting the named benchmark conditions. Completion records documentation, not performance.
Evidence you can inspect. Benchmark dossiers distinguish published facts from our interpretation, with source versions and access notes attached.
Download the evidence ↗