RESUMO
OBJECTIVE: To develop a simple scoring system to predict 30 day in-hospital mortality of in-patients excluding those from intensive care units based on easily obtainable demographic, disease and nutrition related patient data. METHODS: Score development with general estimation equation methodology and model selection by P-value thresholding based on a cross-sectional sample of 52 risk indicators with 123 item classes collected with questionnaires and stored in an multilingual online database. SETTING: Worldwide prospective cross-sectional cohort with 30 day in-hospital mortality from the nutritionDay 2006-2009 and an external validation sample from 2012. RESULTS: We included 43894 patients from 2480 units in 32 countries. 1631(3.72%) patients died within 30 days in hospital. The Patient- And Nutrition-Derived Outcome Risk Assessment (PANDORA) score predicts 30-day hospital mortality based on 7 indicators with 31 item classes on a scale from 0 to 75 points. The indicators are age (0 to 17 points), nutrient intake on nutritionDay (0 to 12 points), mobility (0 to 11 points), fluid status (0 to 10 points), BMI (0 to 9 points), cancer (9 points) and main patient group (0 to 7 points). An appropriate model fit has been achieved. The area under the receiver operating characteristic curve for mortality prediction was 0.82 in the development sample and 0.79 in the external validation sample. CONCLUSIONS: The PANDORA score is a simple, robust scoring system for a general population of hospitalised patients to be used for risk stratification and benchmarking.
