What question did this study set out to answer?

May 5, 2026

Development and internal validation of a clinical risk prediction model for pancreatic ductal adenocarcinoma in the UK Biobank.

Puntos clave

The study aims to develop a risk prediction model for pancreatic ductal adenocarcinoma (PDAC) using clinical risk factors.
Utilized data from the UK Biobank, involving 365,326 individuals aged 40-69 years.
Identified incident PDAC through cancer registries and hospital records, analyzing time-to-PDAC using Cox proportional hazards models.
Evaluated model performance through calibration and discrimination analysis, with internal validation via 5-fold cross-validation.
Among participants, 314 incident PDAC cases were identified within 5 years.
The model demonstrated good calibration (observed-to-expected ratio 0.994 [95% CI 0.890-1.111]) and discrimination (AUC 0.741).
Stable performance was confirmed through internal validation across 5-fold cross-validation.

Resumen

INTRODUCTION: Pancreatic ductal adenocarcinoma (PDAC) remains one of the most lethal cancers. Earlier identification through improved risk prediction modelling represents a promising strategy for improving PDAC outcomes. We developed a PDAC risk prediction model using routinely available clinical risk factors, including modifiable factors, to enable targeted screening. METHODS: Data from the UK Biobank, a prospective cohort of adults aged 40-69 years recruited between 2006 and 2010, were used to develop a PDAC risk prediction model. Incident PDAC was identified through linkage to cancer registries, hospital admissions, and death records. Candidate predictors included demographic, lifestyle, clinical, and laboratory variables routinely available in primary care. Time-to-PDAC was analyzed using Cox proportional hazards models with backward selection. Model performance at 5 years was evaluated calibration and discrimination, with internal validation via 5-fold cross-validation. RESULTS: Among 365,326 participants, 314 incident PDAC cases occurred within 5 years. Predictors included age, cigarette smoking, alcohol intake, pancreatitis, blood test alterations (alanine/aspartate aminotransferase, high-density cholesterol, glycated hemoglobin A1c). The model showed good calibration (observed-to-expected ratio was 0.994 95% CI 0.890-1.111; integrated calibration index = 2.402 × 10⁻⁵; E50 = 1.928 × 10⁻⁵; E90 = 3.283 × 10⁻⁵) and discrimination (time‑dependent area under the receiver operating characteristic curve 0.741). Internal 5-fold cross-validation confirmed stable performance. CONCLUSION: We developed a well-calibrated and discriminative PDAC risk prediction model using routinely available clinical and laboratory data. This models offers an actionable framework for risk stratification in primary care settings, though external validation in independent cohorts is warranted before clinical implementation.

Preguntar a la IA

Me gusta

Guardar