Qiao-Li Wang1, Eivind Ness-Jensen, Giola Santoni, Shao-Hua Xie, Jesper Lagergren. 1. 1Upper Gastrointestinal Surgery, Department of Molecular Medicine and Surgery, Karolinska Institutet, Karolinska University Hospital, Stockholm, Sweden; 2HUNT Research Centre, Department of Public Health and Nursing, NTNU, Norwegian University of Science and Technology, Levanger, Norway; 3Medical Department, Levanger Hospital, Nord-Trøndelag Hospital Trust, Levanger, Norway; 4School of Cancer and Pharmaceutical Sciences, King's College London, United Kingdom.
Abstract
INTRODUCTION: Esophageal squamous cell carcinoma (ESCC) carries a poor prognosis, but earlier tumor detection would improve survival. We aimed to develop and externally validate a risk prediction model based on exposure to readily available risk factors to identify high-risk individuals of ESCC. METHODS: Competing risk regression modeling was used to develop a risk prediction model. Individuals' absolute risk of ESCC during follow-up was computed with the cumulative incidence function. We used prospectively collected data from the Nord-Trøndelag Health Study (HUNT) for model derivation and the UK Biobank cohort for validation. Candidate predictors were age, sex, tobacco smoking, alcohol consumption, body mass index (BMI), education, cohabitation, physical exercise, and employment. Model performance was validated internally and externally by evaluating model discrimination using the area under the receiver-operating characteristic curve (AUC) and model calibration. RESULTS: The developed risk prediction model included age, sex, smoking, alcohol, and BMI. The AUC for 5-year risk of ESCC was 0.76 (95% confidence interval [CI], 0.58-0.93) in the derivation cohort and 0.70 (95% CI, 0.64-0.75) in the validation cohort. The calibration showed close agreement between the predicted cumulative risk and observed probabilities of developing ESCC. Higher net benefit was observed when applying the risk prediction model than considering all participants as being at high risk, indicating good clinical usefulness. A web tool for risk calculation was developed: https://sites.google.com/view/escc-ugis-ki. DISCUSSION: This ESCC risk prediction model showed good discrimination and calibration and validated well in an independent cohort. This readily available model can help select high-risk individuals for preventive interventions.
INTRODUCTION: Esophageal squamous cell carcinoma (ESCC) carries a poor prognosis, but earlier tumor detection would improve survival. We aimed to develop and externally validate a risk prediction model based on exposure to readily available risk factors to identify high-risk individuals of ESCC. METHODS: Competing risk regression modeling was used to develop a risk prediction model. Individuals' absolute risk of ESCC during follow-up was computed with the cumulative incidence function. We used prospectively collected data from the Nord-Trøndelag Health Study (HUNT) for model derivation and the UK Biobank cohort for validation. Candidate predictors were age, sex, tobacco smoking, alcohol consumption, body mass index (BMI), education, cohabitation, physical exercise, and employment. Model performance was validated internally and externally by evaluating model discrimination using the area under the receiver-operating characteristic curve (AUC) and model calibration. RESULTS: The developed risk prediction model included age, sex, smoking, alcohol, and BMI. The AUC for 5-year risk of ESCC was 0.76 (95% confidence interval [CI], 0.58-0.93) in the derivation cohort and 0.70 (95% CI, 0.64-0.75) in the validation cohort. The calibration showed close agreement between the predicted cumulative risk and observed probabilities of developing ESCC. Higher net benefit was observed when applying the risk prediction model than considering all participants as being at high risk, indicating good clinical usefulness. A web tool for risk calculation was developed: https://sites.google.com/view/escc-ugis-ki. DISCUSSION: This ESCC risk prediction model showed good discrimination and calibration and validated well in an independent cohort. This readily available model can help select high-risk individuals for preventive interventions.