Risk stratification models in elderly patients: recalibrating or remodeling?

ABSTRACT

OBJECTIVE: Risk stratification models perform poorly regarding elderly patients. This study aims to evaluate the performance of the Italian Coronary Artery Bypass Graft Outcome Project (ItCABG) model and the logistic European System for Cardiac Operative Risk Evaluation (LogEuroSCORE) model by age group to build a model specifically for elderly patients. METHODS AND RESULTS: Data from the ItCABG and Mattone Outcome-BYPASS study (MO-BYPASS) have been used. ItCABG and LogEuroSCORE models were applied to the ItCABG population (n = 34,310) stratified by four age classes < 60, 60-69, 70-79, > or = 80 years. Each model's ability to predict 30-day mortality was assessed for accuracy (Brier score and pseudo-R2), calibration (Hosmer-Lemeshow test, X(H-L)) and discrimination (area under the receiver operating characteristic curve, AUC) in age classes. To estimate the elderly risk function a logistic regression was performed on 2,255 octogenarian patients from ItCABG and MO-BYPASS. Elderly model's performance was tested. Model accuracy is fair in all age classes although the explained variance is poor. ItCABG and LogEuroSCORE models revealed good discrimination power in patients aged < (60, 60-69, and 70-79 years, but not in patients aged > or = 80 years (AUCs 0.82, 0.77, 0.76, 0.64, and 0.78, 0.75, 0.74, 0.65, respectively). Calibration of both models is poor in patients 80 years (ItCABG X(H-L) = 18.1, P = 0.05; LogEuroSCORE X(H-L) = 129.7, P < 0.001). When a new model specific to octogenarian patients was built, discrimination power remained poor (AUC = 0.66), although calibration power improved (X(H-L) = 3.93, P = 0.86). CONCLUSIONS: ItCABG and LogEuroSCORE models were poor predictors of mortality in octogenarian patients. Elderly-specific risk factors must be assessed to improve risk stratification in patients aged 80 years and older.