Evaluation of a wearable biosensor to monitor potassium imbalance in patients receiving hemodialysis

A non-invasive method capable of promptly detecting clinically important blood potassium changes could benefit care and safety for significant patient populations, including those with end-stage kidney disease. A total of 96 patients receiving maintenance hemodialysis participated in service evaluations of a wearable biosensor across four renal centers (two in UK, one in US and one in Saudi Arabia). All the patients had standard blood tests taken before and after their routine hemodialysis sessions and the results were used as reference potassium measurements for simultaneous, photoplethysmography-based, non-invasive digital samples obtained by the wearable biosensor. These digital samples were subsequently analyzed utilizing a machine learning model designed to identify excursions in serum potassium concentration by quantifying changes across a ternary classification strategy— hyperkalemia (K+ > 5.2 mEq/L), normokalaemia (K+ 3.5–5.2 mEq/L) or hypokalemia (K+ < 3.5 mEq/L). Reference serum potassium results ranged from 2.5 to 6.4 mEq/L. an XGBoost proprietary machine learning model enabled classification of non-invasive digital samples into one of three categories: Hypokalemia (K+ < 3.5 mEq/L), hyperkalemia (K+ > 5.2 mEq/L) or normokalemia (3.5 ≥ K+ ≤ 5.2). The total weighted recall of the biosensor and model was 86%. The overall weighted precision of the model was 86% with an F1-score of 0.86 indicating that the model achieved both high sensitivity and a low rate of false positives This evaluation demonstrates wearable technology capable of identifying important blood potassium changes outside of the normal reference range, in a group of patients receiving hemodialysis. • First evaluation of a non-invasive wearable technology used for potassium classification in hemodialysis patients. • Novel use of a photoplethysmography-based multi-modal sensor for non-invasive, patient-to-provider classification system. • Proprietary XGBoost machine learning model trained, validated and tested on the co-morbid population of interest. • Future potential for paradigm-shifting, ground-breaking changes to the clinical treatment of patients on hemodialysis.