Original Investigation
Nephrology
January 3, 2025
Mortality Risk Prediction Models for People With Kidney Failure: A Systematic Review
Faisal Jarrar, Meghann Pasternak, Tyrone G. Harrison, et al
JAMA Netw Open. 2025;8(1):e2453190. doi:10.1001/jamanetworkopen.2024.53190
Question What are the quality and clinical applicability of existing mortality prediction models for people with kidney failure?
Findings This systematic review identified and evaluated 50 studies with more than 2.9 million participants reporting mortality prediction models for people with kidney failure. Models were found to be at high risk of bias and to have applicability concerns for clinical practice, and none demonstrated promise in terms of clinical usability or incorporation into guidelines.
Meaning These findings suggest that new mortality prediction models are needed to inform treatment decisions in people with kidney failure.
Abstract
Importance People with kidney failure have a high risk of death and poor quality of life. Mortality risk prediction models may help them decide which form of treatment they prefer.
Objective To systematically review the quality of existing mortality prediction models for people with kidney failure and assess whether they can be applied in clinical practice.
Evidence Review MEDLINE, Embase, and the Cochrane Library were searched for studies published between January 1, 2004, and September 30, 2024. Studies were included if they created or evaluated mortality prediction models for people who developed kidney failure, whether treated or not treated with kidney replacement with hemodialysis or peritoneal dialysis. Studies including exclusively kidney transplant recipients were excluded. Two reviewers independently extracted data and graded each study at low, high, or unclear risk of bias and applicability using recommended checklists and tools. Reviewers used the Prediction Model Risk of Bias Assessment Tool and followed prespecified questions about study design, prediction framework, modeling algorithm, performance evaluation, and model deployment. Analyses were completed between January and October 2024.
Findings A total of 7184 unique abstracts were screened for eligibility. Of these, 77 were selected for full-text review, and 50 studies that created all-cause mortality prediction models were included, with 2 963 157 total participants, who had a median (range) age of 64 (52-81) years. Studies had a median (range) proportion of women of 42% (2%-54%). Included studies were at high risk of bias due to inadequate selection of study population (27 studies [54%]), shortcomings in methods of measurement of predictors (15 [30%]) and outcome (12 [24%]), and flaws in the analysis strategy (50 [100%]). Concerns for applicability were also high, as study participants (31 [62%]), predictors (17 [34%]), and outcome (5 [10%]) did not fit the intended target clinical setting. One study (2%) reported decision curve analysis, and 15 (30%) included a tool to enhance model usability.
Conclusions and Relevance According to this systematic review of 50 studies, published mortality prediction models were at high risk of bias and had applicability concerns for clinical practice. New mortality prediction models are needed to inform treatment decisions in people with kidney failure.
