Predicting Significant Symptom Burden Following Dialysis Initiation: Two-Year Risk Model

Journal of the American Society of Nephrology, vol. 36

Abstract

Background: Predictive research in nephrology has neglected patient-centered outcomes. We developed a novel prediction model to identify patients at risk of experiencing significant symptom burden following initiation of maintenance dialysis, which could identify patients for treatment and aid in shared decision making. Methods: Data from the CRIC study was used to develop a tool for use in patients with advanced CKD planning to start dialysis. Baseline was the study visit preceding dialysis initiation. Symptom burden was the average response to 12 symptom specific questions on the Kidney Disease Quality of Life (KDQOL) instrument. The composite outcome, significant symptom burden, was a ≥5 point decrease (worsening) from baseline in the KDQOL symptom score or a post-dialysis KDQOL symptom score ≥5 points below the national average. Predictors were chosen based on prior association with symptoms and included age, sex, BMI, eGFR, albumin, hemoglobin, diabetes, CVD, depression, polypharmacy, opioid use, quality of life and pre-dialysis symptom burden. The model was developed using competing risks regression accounting for death, and performance was assessed by discrimination and calibration. Results: The analysis included 1,057 CRIC participants who initiated dialysis. Of these, 438 (41%) reported significant symptom burden within two years of dialysis initiation and 161 (15%) died. Discrimination of the model was limited with a time-dependent AUC of 0.62 (95% CI: 0.58-0.65). Overall model calibration was excellent, with a ratio of observed and expected outcomes (O/E ratio) of 1.0 (95% CI: 0.9 – 1.1). The calibration plot is shown in Figure 1. Conclusion: We developed a parsimonious prediction model for significant symptom burden following dialysis initiation with excellent calibration but only modest discrimination. Next steps involve testing of additional predictors to improve performance, external validation, and exploration of alternative patient-reported measures as outcomes.

Authors 4

1. Kendra E. Wulczyn

   Massachusetts General Hospital

   Affiliation as printed

   Massachusetts General Hospital Department of Medicine, Boston, Massachusetts, United States;

2. James Phillip Lash

   University of Illinois Chicago

   Affiliation as printed

   University of Illinois Chicago, Chicago, Illinois, United States;

3. Sahir Kalim

   Massachusetts General Hospital

   Affiliation as printed

   Massachusetts General Hospital Department of Medicine, Boston, Massachusetts, United States;

4. Merel van Diepen Aachen

   Leiden University Medical Center

   Affiliation as printed

   Leids Universitair Medisch Centrum, Leiden, ZH, Netherlands