Determinants of candidemia and candidemia-related death in cardiothoracic ICU patients

Study objectives: To develop and prospectively validate models of independent predictors of candidemia and candidemia-related death in cardiothoracic ICU (CICU) patients. Design: (1) An initial, prospective, one-center, case-control, independent predictor-model determining study; and (2) a prospective, two-center, model-validation study. Setting: The initial study was performed at the 14-bed CICU of the Onassis Cardiac Surgery Center, Athens, Greece; the model-validation study was performed at the Onassis Cardiac Surgery Center CICU and the 12-bed CICU of Henry Dunant General Hospital, Athens, Greece. Patients: In the initial study, 4,312 patients admitted to the Onassis Center CICU between March 1997 and October 1999 were considered for enrollment; 30 candidemic and 120 control patients (case/control ratio, 1/4) matched according to potential confounders were ultimately enrolled. In the model-validation study, 2,087 patients admitted to the Onassis and Henry Dunant CICUs between November 1999 and May 2002 were prospectively enrolled. Measurements and results: Models of predictors of candidemia and associated death were constructed with stepwise logistic regression and subsequently validated. Independent candidemia predictors were ongoing invasive mechanical ventilation (IMV) greater than or equal to 10 days, hospital-acquired bacterial infection and/or bacteremia, cardiopulmonary bypass duration > 120 min, and diabetes mellitus. Model performance was as follows: sensitivity, 53.3%/57.9%; specificity, 100%/100%; positive predictive value (PPV), 100%/100%; negative predictive value (NPV), 88.9%/99.6%; and accuracy, 90.1%/99.6% (initial/model-validation study values, respectively). IMV greater than or equal to 10 days and hospital-acquired bacterial infection/bacteremia were the two strongest candidemia predictors. APACHE (acute physiology and chronic health evaluation) II score greater than or equal to30 at candidemia onset independently predicted candidemia-related death with 80.0%/85.7% sensitivity, 80%/75% specificity, 66.7%/66.7% PPV, 88.9%/88.9% NPV, and 80.0%/78.9% accuracy (initial/model-validation study values, respectively). Conclusions: We provided a set of easily determinable independent predictors of the occurrence of candidemia in CICU patients. Our results provide a rationale for implementing preventive measures in the form of independent predictor control, and initiating antifungal prophylaxis in high-risk CICU patients.