Development and validation of a multi-cancer risk identification model in 42,666 individuals: a population-based prospective study

Background

Identifying high-risk individuals is crucial for effective cancer screening. However, developing a practical risk prediction model with proper validation for multiple cancer types presents significant challenges.

Methods

We initialized the FuSion cohort study by recruiting 42,666 participants from Taizhou, China, between 2011 and 2021. Among these participants, 16,340 were recruited from 2011 to 2014 and were designated as the discovery cohort, while 26,308 participants enrolled between 2018 and 2021 were utilized as the validation cohort. In the discovery phase, we developed a multi-cancer risk prediction model for five common cancers, including lung, esophageal, liver, gastric, and colorectal cancer, using a comprehensive variable selection framework based on five machine learning methods. The predictors were selected from 74 epidemiological risk factors and blood biomarkers. The participants from the validation cohort were classified into high-, intermediate-, and low-risk groups based on the established model. We followed up the different risk groups and conducted clinical medical examinations, such as CT scans and endoscopic examinations, to evaluate the model's effectiveness in predicting cancer risk.

Findings

In the discovery phase, we developed a multi-cancer risk prediction model based on four biomarkers, AFP, CEA, CYFRA-211 and HBsAg, as well as age, sex, and smoking intensity. The model exhibited an AUROC of 0.767 (95%CI: 0.723-0.814) for five-year incidence prediction of multiple cancers, and the high-risk group exhibited a 15.19-fold (95%CI: 5.97-38.64) and 4.13-fold (95%CI: 2.67-6.39) increased risk compared to the low-risk population and intermediate-risk population, respectively. Among 17.19% of participants in the validation cohort that were identified as high risk, 50.41% of all new cancer cases were expected. In the face-to-face follow-up of 2,941 high-risk individuals, 9.64% were newly diagnosed with cancer or precancerous lesions. It was 5.02 times and 1.74 times as high as that in the low- and intermediate-risk group, respectively. In particular, the incidence of esophageal cancers in the high-risk group was 16.84 times as high as that in the low-risk group.

Interpretation

This is the first population-based multi-cancer risk prediction study conducted on a large Chinese cohort. The effective risk stratification model developed in this study would facilitate targeted prevention strategies and enhance early screening efforts for high-risk populations, ultimately optimizing healthcare resources.