This is a prospective, single-arm observational study that aims to assess the validity and reproducibility of an algorithm for assessing fluid status in a cohort of dialysis patients. The study will externally validate an existing algorithm for dry weight prediction in real-time in a cohort of dialysis patients.
Volume Overload is a contributing factor to the high rates of cardiovascular and all-cause mortality demonstrated in haemodialysis patients. At present, no method exists that can consistently refine volume status and provide patients with feedback to allow adjustments to their fluid intake. Current standards used to assess volume are either poorly predictive of fluid status, cumbersome to use, or lack an adequate patient interface. An automated, accurate and periodic assessment of dry weight would be clinically useful, low-cost, and rapidly scalable. Machine learning methods have been widely studied in nephrology. Large amounts of precise haemodialysis data, collected and stored electronically at regular intervals, have the potential to be leveraged in the prediction of patients' extracellular volume or ideal fluid status. A number of proof-of-concept machine-learning models for the prediction of dry weight in haemodialysis data have been created using retrospective data. This study will evaluate the usability of the machine learning models in managing fluid volume in haemodialysis patients while also assessing their validity and reproducibility against validated measurements; in this instance the Body Composition Monitor (BCM) by Fresenius. As the machine learning model for assessing fluid status was trained and tested on retrospective data, there is sufficient justification for testing the model's performance, acceptability and usability in a controlled, observational prospective study. This will be an 8-week trial with a 2-week run-in period conducted in a single centre in Beaumont, Dublin, Ireland. Bioimpedance measurements using the Fresenius BCM will be performed every 2 weeks. Haemodialysis data will be processed continuously throughout the trial. The algorithm will use haemodialysis data to predict the BCM output. The algorithm prediction will be compared to the BCM prediction to assess its usability.
Study Type
OBSERVATIONAL
Enrollment
24
Beaumont Hospital
Dublin, Leinster, Ireland
Beaumont Hospital
Dublin, Leinster, Ireland
The primary objective is to determine the validity of the machine learning model in estimating bioimpedance-determined dry weight in haemodialysis patients.
Dry weight (kg) estimated by the machine learning estimation model will be compared with the bioimpedance normohydration weight in kg.
Time frame: 8 weeks
Acceptability
The acceptability of the machine learning model's outputs from a clinical healthcare perspective will be assessed. * The patient's clinical fluid status will be assessed via a survey administered to haemodialysis nursing staff every 2 weeks pre-dialysis. The survey will ask the haemodialysis nursing staff to define a patient's fluid overload in kg. * The mean difference between fluid overload in kg as defined by healthcare staff and fluid overload in kg as defined by the algorithmic output will be assessed.
Time frame: 8 weeks
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.