The predictors of hospital mortality among adult patients with diabetes

Background: Diabetes Mellitus is a lifelong chronic disease with higher risks of mortality and morbidity. The disease is associated with acute and chronic complications. Objective: To study of the risk factors, including the social determinants of the disease to help in mitigating the complications and reducing the deaths among patients with DM. Methods: The study is a secondary data analysis of existing dataset based on the Nationwide Inpatient Sample (NIS). The current study is based on the NIS data during the period 2007 to 2010 inclusive. The analysis will include only adult population (18 years age or older). My primary outcomes of interest will be the mortality (dead/alive) or (living status of adult diabetic subjects). Risk factors that will be investigated are Personal and demographic characteristics, socioeconomic factors, Medical factors, and health related factors. Descriptive (means and proportions/percentage) and bivariate analyses (chi-square and t-test) where appropriate. Regression models to evaluate the crude association between each potential predictor variable and all outcomes of interest. Then, a hierarchical generalized linear modeling (HGLM) approach will be used to assess the odds of changing death rate controlling for potential confounders. Results: A sample of 438838 participated in the 4 year included in the study. Death Rate among diabetic patients decreased from 2007-2010 significantly. Race, income, insurance, patient living location, admission source, admission type, other diagnosis, drugs, age and many other factors have statistical significant difference between rate of death among diabetic patients compared to non diabetic subjects. We found that age, total charge, no of diagnosis, no of procedures, drug risk mortality and severity, LOS, and comorbid conditions are statistically significant risk factors for higher mortality among diabetic subjects compared to non diabetic subjects controlling for the other factors and potential confounders. All significant relationships were tested at the alpha level of (P