Statistical analyses

The data was input and checked by EpiData3.1, and the diseases status of old adults with multimorbidity was described by the number of diseases (N) and percentage (%). Meanwhile, LCA was used to identify the multimorbidity patterns, and hierarchical logistic regression analysis was applied to determine the multi-layered factors associated with various multimorbidity patterns for old adults. Specifically, multi-layered predictors were tested step-by-step for their prediction of the outcome variables, and five models were established in each multimorbidity pattern respectively: model I incorporated the first-layer factor (innate personal traits); the second-layer factor (behavioral lifestyles) was added in model II on the basis of the model I; similarly, model III, model IV and V introduced the third-layer factor (interpersonal networks), the forth-layer factor (socio-economic status) and the fifth-layer factor (macro-environmental) individually on the basis of model II, model III and model IV. We applied Mplus 7.4 to perform LCA and Stata15.1 to perform hierarchical logistic regression analysis, and statistically significant level was set at 0.05.