A Prediction Model for Adverse Outcome in Hospitalized Patients With Diabetes

OBJECTIVE: There are no formal prognostic models predicting adverse outcomes (excessive length of stay or mortality) in hospitalized patients with diabetes. In this study, we aimed to develop a prediction model that will help identify patients with diabetes who are most likely to have an adverse event during their hospital stay. RESEARCH DESIGN AND METHODS: Analysis was based on 25,118 admissions with diabetes to University Hospital Birmingham, Birmingham, U.K., over 4 years (2007–2010). Adverse events are defined as either excessive length of stay or inpatient mortality. Key predictors were variables that are often available in the first 72 h of admission and included demographic characteristics, clinical pathological test results, and use of insulin. Models were constructed using logistic regression, discrimination and calibration was assessed, and internal validation was carried out. RESULTS: The model performed well with an area under the curve (AUC) of 0.802 with only a mild reduction being noted in the internal validation (AUC 0.798). At a cutoff value of 25% probability of having an adverse outcome the sensitivity was 76%, specificity was 70%, and the positive predictive value was 49%. If it is used for a case-finding approach limiting to noncritical care settings, then at the same cutoff value, two-thirds (sensitivity 69%) of the admissions with adverse outcomes could potentially be identified. CONCLUSIONS: Once externally validated, we suggest that our model will be a useful tool for identifying diabetic patients who are at risk for poor outcomes when admitted to hospital.