Welcome to the CJU website » LOG IN


Estimating the risk of chronic kidney disease after nephrectomy
Stanford University School of Medicine, Stanford, California, USA
Dec  2013 (Vol.  20, Issue  6, Pages( 7035 - 7041)
PMID: 24331345


Text-Size + 


    To identify factors associated with the development of chronic kidney disease (CKD) after nephrectomy and to create a clinical model to predict CKD after nephrectomy for kidney cancer for clinical use.


    We identified 144 patients who had normal renal function (eGFR > 60) prior to undergoing nephrectomy for kidney cancer. Selected cases occurred between 2007 and 2010 and had at least 30 days follow up. Sixty-six percent (n = 95) underwent radical nephrectomy and 62.5% (n = 90) developed CKD (stage 3 or higher) postoperatively. We used univariable analysis to screen for predictors of CKD and multivariable logistic regression to identify independent predictors of CKD and their corresponding odds ratios. Interaction terms were introduced to test for effect modification. To protect against over-fitting, we used 10-fold cross-validation technique to evaluate model performance in multiple training and testing datasets. Validation against an independent external cohort was also performed.


    Of the variables associated with CKD in univariable analysis, the only independent predictors in multivariable logistic regression were patient age (OR = 1.27 per 5 years, 95% CI: 1.07-1.51), preoperative glomerular filtration rate (GFR), (OR = 0.70 per 10 mL/min, 95% CI: 0.56-0.89), and receipt of radical nephrectomy (OR = 4.78, 95% CI: 2.08-10.99). There were no significant interaction terms. The resulting model had an area under the curve (AUC) of 0.798. A 10-fold cross-validation slightly attenuated the AUC to 0.774 and external validation yielded an AUC of 0.930, confirming excellent model discrimination.


    Patient age, preoperative GFR, and receipt of a radical nephrectomy independently predicted the development of CKD in patients undergoing nephrectomy for kidney cancer in a validated predictive model.