Although socioeconomic, behavioural, psychological, and biological factors have been individually linked to multimorbidity, data on the importance of these factors are limited. Our study aimed to determine the leading predictors for multimorbidity of chronic conditions in middle-aged Australian adults using machine learning methods. We included 53,867 participants aged 45-64 years from the 45 and Up Study who were free of eleven predefined chronic conditions at baseline (2006-2009) in the analysis. Incident multimorbidity was defined by the co-existence of ≥2, ≥3, or ≥ 4 conditions during follow-up until December 31, 2016. The five leading predictors for multimorbidity in men were age (7.2-20.5% of total variance), body mass index (6.5-15.4%), smoking (4.0-8.3%), chicken intake (3.6-7.5%), and red meat intake (4.6-6.3%) across the three definitions. Leading predictors varied across the three definitions in women, but the four common ones were body mass index (6.3-20.1%), age (6.2-16.4%), chicken intake (4.1-8.3%), and red meat intake (4.2-4.7%). The ten leading modifiable health factors accounted for 39.4-46.1% of total variance across the three definitions. Men with 6-10 health factors had 46-54% lower risks for multimorbidity compared with those reporting ≤2. The corresponding percentage for women was 45-52%. Non-behavioural factors including psychological distress, low education and income and high relative economic disadvantage were among the leading risk factors for multimorbidity. In conclusion, modifications on behavioural factors including diets, physical activity, smoking, alcohol consumption may reduce the risk of multimorbidity in middle-aged adults, whereas individuals with low socioeconomic status or psychological distress are at the highest priority for intervention.
Keywords: Biological factors; Leading determinants; Lifestyle factors; Machine learning; Multimorbidity; Psychological factors; Socioeconomic factors.
Copyright © 2020 Elsevier Inc. All rights reserved.