Application of zero-inflated negative binomial regression model in study of the impacting factors about multimorbidity

Objective To study the application of zero-inflated negative binomial regression model in the residents' multimorbidity and its impacting factors. Methods Poisson distribution, negative binomial distribution and zero-inflated model were used to fit the number of multimorbidity and the aggregation was analyzed, then the main factors were screened out in multimorbidity. Results The number of multimorbidity did not accord with Poisson distribution (χ²=196.419, P<0.001) and meet the negative binomial distribution (χ²=6.677, P=0.154); the aggregation index K=1.779, over-dispersion test O=15.18> 1.96, so the data was clustered. Zero expansion test Vuong=6.58, P<0.001, zero-inflate model was better than Poisson or negative binomial model.The negative binomial part of the results suggest thatthe number of multimorbidity will increase, when the residents had risk factors including the older, high intensity exercise, the higher the degree of anxiety, the higher the body mass index, the higher the hemoglobin A1c (HbA1c) level, a family history of diabetes, high blood pressure history, high systolic blood pressure and high levels of cholesterol; In the Logit part of the results:residents had a higher risk of developing chronic diseases, who had risk factors which included the older, the higher the degree of anxiety, the higher the body mass index, the higher the level of triglycerides, the higher the fasting blood glucose (FPG), the family history of hypertension and high systolic blood pressure. Conclusion Multimorbidity is characterized by aggregation and zero-inflated count. Zero-inflated negative binomial regression model has obvious advantages in fitting the data with such characteristics.