A mortality prediction model in diabetic ketoacidosis

AIM To assess the value of clinical and laboratory parameters in predicting mortality in patients presenting with diabetic ketoacidosis (DKA). METHODS The records of all DKA admissions within 10 years were reviewed. Eighteen variables were evaluated at initial presentation and 20 variables at 4, 12 and 24 h from admission. A scoring system derived from these variables was compared to the APACHE III scoring system. RESULTS Among 154 patients (52 males, mean age 58 +/- 12 years), 20 (13%) died in hospital. Multivariate analysis yielded six variables as significant independent predictors ( P < 0.05) of mortality: severe coexisting diseases (SCD) and pH < 7.0, at presentation; units of regular insulin required in the first 12 h > 50 and serum glucose > 16.7 mmo/l, after 12 h; depressed mental state and fever, after 24 h. An integer-based scoring system was derived, as follows: number of points = 6 (SCD at presentation) + 4 (pH < 7.0 at presentation) + 4 (regular insulin required > 50 IU after 12 h) + 4 (serum glucose > 16.7 mmo/l after 12 h) + 4 (depressed mental state after 24 h) + 3 (fever after 24 h). Patients with 0-14 points had 0.86% risk of death, whereas for those with 19-25 points the risk was 93.3%. Median APACHE III scores differed significantly ( P < 0.001) among groups of patients stratified according to the above scoring system. CONCLUSIONS Risk stratification of patients with diabetic ketoacidosis is possible from simple clinical and laboratory variables available during the first day of hospitalization.