Background: There is a concern about the accuracy of the available prognostic indexes when applying them to the emergent population of polypathological patients (PP).
Methods: To develop a 1-year mortality predictive index on PP, we developed a multicenter prospective cohort-study recruiting 1.632 PP after hospital discharge, outpatient clinics, or home hospitalization, from 33 hospitals. Potential risk factors were obtained in the 1.525 PP who completed follow-up. Each factor independently associated with mortality in the derivation cohort (757 PP from western hospitals) was assigned a weight, and risk scores were calculated by adding the points of each factor. Accuracy was assessed in the validation cohort (768 PP from eastern hospitals) by risk quartiles calibration, and discrimination power, by ROC curves. Finally, accuracy of the index was compared with that of the Charlson index.
Results: Mortality in the derivation/validation cohorts was 35%/39.5%, respectively. Nine independent mortality predictors were identified to create the index (age ≥85 years, 3 points; No caregiver or caregiver other than spouse, 2 points; active neoplasia, 6 points; dementia, 3 points; III-IV functional class on NYHA and/or MRC, 3 points; delirium during last hospital admission, 3 points; hemoglobinemia <10 g/dl, 3 points; Barthel index <60 points, 4 points; ≥4 hospital admissions in last 12 months, 3 points). Mortality in the derivation/validation cohorts was 12.1%/14.6% for patients with 0-2 points; 21.5%/31.5% for those with 3-6 points; 45%/50% for those with 7-10 points; and 68%/61.3% for those with ≥11 points, respectively. Calibration was good in derivation/validation cohorts, and discrimination power by area under the curve was 0.77/0.7. Calibration of the Charlson index was good, but discrimination power was suboptimal (area under the curve, 0.59).
Conclusions: This prognostic index provides an accurate and transportable method of stratifying 1-year death risk in PP.
Copyright © 2010 European Federation of Internal Medicine. Published by Elsevier B.V. All rights reserved.