Abstract: FR-PO395
Distinguishing among Causes of Death in Patients with Kidney Failure on Hemodialysis
Session Information
- Hemodialysis Epidemiology and Outcomes
October 25, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Dialysis
- 801 Dialysis: Hemodialysis and Frequent Dialysis
Authors
- Tran, Michelle, UVA Health, Charlottesville, Virginia, United States
- Xu, Chun, Duke University, Durham, North Carolina, United States
- Wilson, Jonathan A., Duke University, Durham, North Carolina, United States
- Ephraim, Patti, Northwell Health Feinstein Institutes for Medical Research, Manhasset, New York, United States
- Shafi, Tariq, Houston Methodist Hospital, Houston, Texas, United States
- Weiner, Daniel E., Tufts Medical Center, Boston, Massachusetts, United States
- Scialla, Julia J., UVA Health, Charlottesville, Virginia, United States
- Goldstein, Benjamin A., Duke University, Durham, North Carolina, United States
Background
Patients treated with maintenance hemodialysis (HD) are at high risk of death from various causes. While predicting all-cause mortality has been shown to be feasible, predicting cause-specific mortality risk could promote personalized treatment. We assessed whether clinical markers can differentiate among distinct causes of death.
Methods
We used electronic health record data from a not-for-profit dialysis provider linked to United States Renal Data System (USRDS) Files for a cohort of adults treated with maintenance in-center HD who died between 2003-2016. We classified USRDS-reported causes of death into five categories: sudden cardiac death (SCD), non-SCD cardiovascular death, infection, other, and unknown. A subcohort was linked to National Death Index (NDI) with similar categories defined. We used gradient boosting trees to discriminate among causes with demographics, vital signs, laboratory measures, health service utilization and comorbidity claims from 30 days prior to death. We created nested case-control populations for each cause of death and used ridge logistic regression to assess clinical factors that associate with distinct causes.
Results
Classification models for USRDS and NDI outcomes had area under the receiver operating curves (AUROCs) between 0.59-0.70 suggesting minimal ability to distinguish among causes of death. In case-control ridge regression models, coefficients were similar across models for distinct causes of death. Correlation of the coefficients across models was very high (Figure) suggesting that most clinical risk factors are shared and do not distinguish among causes.
Conclusion
Using patient-level clinical data from the 30 days prior to death, we were not able to distinguish among causes of death in maintenance HD patients, suggesting that different causes of death in kidney failure either share similar risk factors or that stated causes of death in USRDS or NDI forms are imprecise.
Funding
- NIDDK Support