APPLICATION OF ZERO-INFLATED POISSON MODEL TO DISABILITY IN TYPE 2 DIABETIC DATA

Abstract

Disability in type – 2 diabetic data is a discrete quantitative data which is mostly modeled by the Poisson regression. This type of data sometimes has the variance greater than the mean which is called over-dispersion; and Poisson regression may not be able to give significant estimation to the data. Therefore, this study aimed at identifying appropriate model to analyze type 2 diabetic patients' disability. Methods: This was cross-sectional data of patients with type 2 diabetic attending a University Teaching Hospital in South-West Nigeria. Disability was the outcome variable while explanatory variables were: age, sex, marital status, disability duration, education, occupation, diastolic blood pressure, fasting blood glucose, body mass index, and glycosylated haemoglobin (HBA1 ). Poisson and Zero-Inflated Poisson (ZIP) models were applied to the data and compared. C Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were used to select a better model. P-value of 5% was considered to be significant. Results: ZIPmodel had the smaller AIC and BIC of 2349.27 and 2407.82 respectively. Therefore, ZIP model seems to be a better model. Age (IRR = 0.997, p < 0.05), marital status (single, IRR= 1.31, p<0.01), BMI (IRR= 0.991, p<0.01), Teaching occupation (IRR= 1.705, p<0.01) were some of the associated factors of disability. Conclusion: ZIP model was the better model in the presence of excess zeros in identifying factors associated with disability in type 2 diabetic patients