Objectives: Large amounts of healthcare data are now generated via patient health records, records of diagnosis and treatment, smart devices, and wearables. Extracting insights from such data can transform healthcare from a traditional, symptom-driven practice into precisely personalized medicine. Dialysis treatments generate a vast amount of data, with more than 100 parameters that must be regulated for ideal treatment outcomes. When complications occur, understanding electrolyte parameters and predicting their outcomes to deliver the optimal dialysis dosing for each patient is a challenge. This study focused on refining dialysis dosing by utilizing emerging data from the growing number of dialysis patients to improve patients' quality of life and well-being.
Methods: Exploratory data analysis and data prediction approaches were performed to gather insights from patients' vital electrolytes on how to improve the patients' dialysis dosing. Four predictive models were constructed to predict electrolyte levels through various dialysis parameters.
Results: The decision tree model showed excellent performance and more accurate results than the support vector machine, linear regression, and neural network models.
Conclusions: The predictive models identified that pre-dialysis blood urea nitrogen, pre-weight, dry weight, anticoagulation, and sex had the most significant effects on electrolyte concentrations. Such models could fine-tune dialysis dosing levels for the growing number of dialysis patients to improve each patient's quality of life, life expectancy, and well-being, and to reduce costs, efforts, and time consumption for both patients and physicians. The study's results need to be validated on a larger scale.
Keywords: Chronic Kidney Disease; Data Science; Machine Learning; Renal Dialysis; Statistical Data Analysis.