Cancer as a main contributor to multimorbidity in the UK Biobank: analysis of risk factors using conventional statistical modelling and machine learning

Cancer frequently co-occurs with other chronic conditions, contributing substantially to disease burden. Multimorbidity is an increasing public health concern, resulting from complex interactions between sociodemographic, lifestyle, and environmental factors. The study aimed to identify the main predictors of multimorbidity using both conventional statistical models and machine learning approaches. We analysed data from 138 126 UK Biobank participants aged 39-73 years. Twenty potential risk factors were assessed using Fine and Gray competing risk regression and random survival forests (RSF). Over the five-year follow-up period, 4384 individuals developed multimorbidity, with cancer present in 52.3% of cases. Five-year cumulative incidence was 3.9% in males and 2.6% in females. Increased waist circumference [males - hazard ratio: 1.15, 95% confidence interval (CI): 1.00-1.31; females - hazard ratio: 1.29, 95% CI: 1.09-1.54] and smoking (per 5 pack-years: males - hazard ratio: 1.06, 95% CI: 1.05-1.07; females - hazard ratio: 1.08, 95% CI: 1.06-1.10) significantly increased risk. Both insufficient and prolonged sleep were linked to a higher risk, especially among females (hazard ratio for prolonged sleep: 1.68, 95% CI: 1.17-2.40). Compared with moderate drinkers, former drinkers, lifelong abstainers, and heavy drinkers showed elevated risks. RSF found smoking, waist circumference, and sleep duration as key predictors in men, while alcohol use, smoking, and waist circumference were most important in women. Available dietary information, physical activity, and air pollution were not major predictors. Smoking, obesity, alcohol, and sleep duration are key risk factors to target in midlife to reduce the future burden of multimorbidity.