RESUMEN
BACKGROUND: Accurate prediction of cardiac surgery-associated acute kidney injury (AKI) would improve clinical decision making and facilitate timely diagnosis and treatment. The aim of the study was to develop predictive models for cardiac surgery-associated AKI using presurgical and combined pre- and intrasurgical variables. STUDY DESIGN: Prospective observational cohort. SETTINGS & PARTICIPANTS: 25,898 patients who underwent cardiac surgery at Cleveland Clinic in 2000-2008. PREDICTOR: Presurgical and combined pre- and intrasurgical variables were used to develop predictive models. OUTCOMES: Dialysis therapy and a composite of doubling of serum creatinine level or dialysis therapy within 2 weeks (or discharge if sooner) after cardiac surgery. RESULTS: Incidences of dialysis therapy and the composite of doubling of serum creatinine level or dialysis therapy were 1.7% and 4.3%, respectively. Kidney function parameters were strong independent predictors in all 4 models. Surgical complexity reflected by type and history of previous cardiac surgery were robust predictors in models based on presurgical variables. However, the inclusion of intrasurgical variables accounted for all explained variance by procedure-related information. Models predictive of dialysis therapy showed good calibration and superb discrimination; a combined (pre- and intrasurgical) model performed better than the presurgical model alone (C statistics, 0.910 and 0.875, respectively). Models predictive of the composite end point also had excellent discrimination with both presurgical and combined (pre- and intrasurgical) variables (C statistics, 0.797 and 0.825, respectively). However, the presurgical model predictive of the composite end point showed suboptimal calibration (P < 0.001). LIMITATIONS: External validation of these predictive models in other cohorts is required before wide-scale application. CONCLUSIONS: We developed and internally validated 4 new models that accurately predict cardiac surgery-associated AKI. These models are based on readily available clinical information and can be used for patient counseling, clinical management, risk adjustment, and enrichment of clinical trials with high-risk participants.
