Predicting the Risk of End-stage Renal Disease in the Population-based Setting

A Retrospective Case-control Study

Eric S Johnson; David H Smith; Micah L Thorp; Xiuhai Yang; Juhaeri Juhaeri


BMC Nephrology. 2011;11 

In This Article


Most observational studies that predicted ESRD using patient characteristics enrolled patients with clinically recognized CKD. We identified a broad range of clinical characteristics--many of them easily measured in the primary care setting--that could help identify patients at risk of progressing to ESRD, possibly through the development of a prognostic risk score.

Our baseline characteristics predict RRT and which patients lived long enough to start RRT. For example, among KPNW patients with stage 3 or worse CKD, the competing risk of mortality was seven times more common than RRT.[8] Although it is only one of the relevant CKD-related outcomes, our model provides a first step toward predicting which patients will start RRT in usual care; patients with those baseline characteristics may be priority candidates for clinical interventions that prevent or slow the rate of progression.

The rate of ESRD in the cohort from which we sampled cases and controls was markedly lower than the overall US rate, in part because our HMO and the metropolitan area that it serves under-represented higher-risk racial and ethnic groups compared with the entire US. Although the HMO did not collect race and ethnicity data consistently during our study period, unpublished surveys of our members have suggested that most (88%) were white, non-Hispanic. Even among white residents, the Portland metropolitan area ranked low (25th percentile) based on its age- and sex-adjusted rate of ESRD (compared with the largest 25 metropolitan statistical areas in the US).[14] Oregon and Washington belong to Network 16, a collection of five Northwestern states tracked by the US Renal Data Systems (USRDS) Annual Data Report. Network 16 states experienced the lowest age- and sex-adjusted rate of dialysis of the 18 Networks tracked by USRDS.[14]

We need to consider biases that might have distorted the size of the odds ratios in this case-control study. To the extent that our final model omitted other important predictors of ESRD, the odds ratios reported here may appear larger than they would in a more complete model. Many of the characteristics were not measured systematically, especially among control patients, as revealed by the pattern of missing data: The ratio of controls to cases drops from 10 to 1 to 6 to 1 after excluding patients with missing data. Physicians ordered certain laboratory tests in part because other clinical signs raised their suspicion of CKD. For example, universal screening for proteinuria at KPNW would probably exhibit a lower odds ratio than the estimate reported in our case-control study because higher risk patients would not be tested preferentially. Our odds ratios reflect the combination of the characteristics' intrinsic predictive values and the physicians' insights for ordering the tests or recording the diagnoses. But reducing the measurement error could also strengthen the predictive power of other characteristics that were measured more systematically. For example, we only required one serum creatinine value at baseline to estimate patients' GFR, which inevitably increases the probability of false-positive classifications attributable to acute renal disease; a second serum creatinine repeated 90 or more days after the initial serum creatinine would reduce measurement error in our classification of GFR and increase the odds ratio for predicting RRT. Perneger and colleagues provide a more thorough review of potential biases in conducting epidemiologic studies of ESRD.[15]

Our retrospectively measured clinical characteristics were largely consistent with the size of estimates from the most similar cohort study, conducted in patients from Kaiser Permanente Northern California. For example, our adjusted odds ratio for diabetes (OR = 2.1) was similar to the adjusted hazard ratio (HR = 2.5) reported by Hsu and colleagues even though their final model included many additional predictors and they followed patients five times longer than we did.[4] Our odds ratio for hypertension (OR = 4.5) was larger than the hazard ratio (HR = 2.9) for severe (stage 2) hypertension reported by Hsu and colleagues.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.