Why the study?
Intradialytic hypotension is associated with increased morbidity and mortality, leading to recent development of machine learning algorithms whose performance, methodological integrity, and clinical applicability required systematic evaluation.
Do machine learning models accurately predict intradialytic hypotension in adult patients on hemodialysis?
Do machine learning models accurately predict intradialytic hypotension in adult patients on hemodialysis?
While machine learning models show high predictive performance for intradialytic hypotension, their clinical applicability is currently limited by a high risk of bias and lack of external validation.
This is the authors' abstract. We don't add key points for this paper.
BACKGROUND: Intradialytic hypotension is associated with increased morbidity, and mortality. Several machine learning (ML) algorithms have been recently developed to predict intradialytic hypotension. We systematically reviewed ML models employed to predict intradialytic hypotension, their performance, methodological integrity, and clinical applicability. METHODS: We conducted this systematic review with a pre-established protocol registered at the International Prospective Register of Systematic Reviews (PROSPERO ID: CRD42022362194). Six databases, from their inception to July 20, 2023, were comprehensively searched. Two independent investigators reviewed the articles, extracted data, and evaluated the risk of bias using the Prediction model Risk of Bias Assessment Tool (PROBAST). RESULTS: Out of 84 screened articles, 16 studies with 14,500 adult patients on hemodialysis were included in the review. Fourteen studies (87.5%) were found to have a high risk of bias. The intradialytic hypotension prevalence in the population investigated was between 1.2 and 51%. A diverse range of predictive ML tools were used to predict intradialytic hypotension, with various neural networking models being the most frequent, appearing in 13 studies (AUROC ranges: 0.684-0.978). One study performed both internal and external validation. CONCLUSIONS: Researchers have made a concerted effort to develop ML tools to predict intradialytic hypotension. Despite their significant efforts, the lack of thorough external and clinical validation, and heterogeneity among the models and settings have resulted in a substantial challenge to offering ML tools as a global intradialytic hypotension prevention and management solution. Future studies should focus on external and clinical validation of these models to enhance the chances of clinically relevant changes in clinical practices.
High bias precludes routine clinical use of ML models for intradialytic hypotension; leaves open need for validated, low-bias prospective trials.
No takes yet. Share an insight, caveat, or question.
Ninan et al. (2025) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: