Abstract
Background/Aims Chronic kidney disease (CKD) is common, but only progresses to end-stage renal disease (ESRD) in a minority of patients. Developing a tool for early identification of patients likely to progress to ESRD would enable clinicians and medical systems to target resources to patients likely to benefit from interventions.
Methods We conducted a cohort study by linking data among 38,483 patients with stage 3 CKD. We measured patient characteristics during the year before patients became eligible because of their poor kidney function. The patient characteristics were known from previous studies to predict ESRD. We followed patients for up to one year. Seven routinely measured patient characteristics accurately predicted the risk of ESRD. By combining those characteristics with numeric weights for their importance, the risk score identified the subgroup (top decile or tenth) of patients at the highest risk.
Results We observed 461 patients who developed ESRD, a one-year risk of 1.6 per 100 patients. We judged the risk score’s effectiveness by dividing the cohort into ten equal groups or deciles, which included approximately 3,849 patients. Patients at or above the 90th percentile of predicted risk (top decile) were 60 times more likely to suffer a deep infection when compared with patients below the 10th percentile (bottom decile): 11.0 per 100 patients (top decile) versus 0.2 per 100 (bottom decile). The c-statistic was 0.89.
Conclusions This pragmatic risk score appears to be the first of its kind for predicting ESRD in patients with stage 3 CKD. Previous risk scores in the KPNW population and other populations have predicted ESRD in both earlier and later stages of CKD and may not be sufficiently accurate for KPNW’s intended application. Our risk score can predict a patient’s absolute risk (e.g., 11 per 100 patients). The risk score can also reveal where a patient’s absolute risk ranks as a percentile. The same seven characteristics are more useful to decision-makers when the hazard ratios (numeric weights) can be combined in a risk score to predict the absolute risk.
