How can we best predict acute kidney injury following cardiac surgery?: a prospective observational study.

BACKGROUND

Several models for predicting acute kidney injury following cardiac surgery have been published, and various end-point definitions have been used.

OBJECTIVES

Our aim was to investigate how acute kidney injury following cardiac surgery could be most accurately predicted.

DESIGN

Single-centre prospective observational study.

SETTING

St Olav's University Hospital, Trondheim, Norway, from 2000 to 2007.

PATIENTS

All 5029 adult patients undergoing cardiac surgery were considered eligible for participation. Patients who required preoperative dialysis and patients with missing information on preoperative or maximum postoperative serum creatinine concentration were excluded (n=51). A total of 4978 patients were entered into the statistical analyses.

MAIN OUTCOME MEASURES

Logistic regression with bootstrapping methods was applied for model development and validation, together with the area under the receiver operating characteristic curve and Hosmer-Lemeshow test. We tested different end-points, exchanged serum creatinine concentration with creatinine clearance or estimated glomerular filtration rate and added intraoperative variables. The main end-point was at least 50% increase in serum creatinine concentration, an increase in concentration by at least 26.4 μmol l(-1) (0.3 mg dl(-1)) or a new requirement for dialysis after surgery.

RESULTS

The final model consisted of 11 preoperative predictors of acute kidney injury: age, BMI, lipid-lowering treatment, hypertension, peripheral vascular disease, chronic pulmonary disease, haemoglobin concentration, serum creatinine concentration, previous cardiac surgery, emergency operation and operation type. The area under the receiver operating characteristic curve was 0.819 (95% confidence interval 0.801 to 0.837), and the Hosmer-Lemeshow test P value was 0.17. Exchanging serum creatinine concentration with glomerular filtration rate or creatinine clearance slightly reduced model discrimination and the addition of intraoperative variables improved discrimination somewhat. Slight end-point definition changes had little impact.

CONCLUSION

The risk of acute kidney injury can be accurately predicted using preoperative variables. Serum creatinine concentration was more accurate than estimated glomerular filtration rate or creatinine clearance. Intraoperative variables slightly improved the model, but did not seem to outweigh the advantages of a preoperative model.