Development and validation of a risk prediction model for incident liver cancer

Abstract

ObjectiveWe aimed to develop and validate a risk prediction model for liver cancer based on routinely available risk factors using the data from UK Biobank prospective cohort study.MethodsThis analysis included 359,489 participants (2,894,807 person-years) without a previous diagnosis of cancer. We used the Fine-Gray regression model to predict the incident risk of liver cancer, accounting for the competing risk of all-cause death. Model discrimination and calibration were validated internally. Decision curve analysis was conducted to quantify the clinical utility of the model. Nomogram was built based on regression coefficients.ResultsGood discrimination performance of the model was observed in both development and validation datasets, with an area under the curve (95% confidence interval) for 5-year risk of 0.782 (0.748–0.816) and 0.771 (0.702–0.840) respectively. The calibration showed fine agreement between observed and predicted risks. The model yielded higher positive net benefits in the decision curve analysis than considering either all participants as being at high or low risk, which indicated good clinical utility.ConclusionA new risk prediction model for liver cancer composed of routinely available risk factors was developed. The model had good discrimination, calibration and clinical utility, which may help with the screening and management of liver cancer for general population in the public health field.

Deep learning methods in liver diseases

“…Liu and colleagues devised and validated a predictive model for liver cancer risk utilizing routinely accessible risk factors. They employed prospective cohort data from the UK Biobank, encompassing 359,489 participants devoid of prior cancer diagnoses, employing fine-gray regression modeling techniques [8]. There have also been studies combined with a multigene risk score and a clinical fibrosis score.…”