Improvement of Dialysis Dosing Using Big Data Analytics.

Area of Science:

• Nephrology
• Biomedical Informatics
• Data Science

Background:

• Healthcare data, including patient records and wearable device information, offers potential for personalized medicine.
• Dialysis treatments generate extensive data, requiring careful management of over 100 parameters for optimal outcomes.
• Predicting electrolyte levels and outcomes is crucial for effective dialysis dosing, especially when complications arise.

Purpose of the Study:

• To refine dialysis dosing strategies by leveraging data from a growing dialysis patient population.
• To enhance patient quality of life and well-being through improved dialysis treatment.
• To explore predictive modeling for electrolyte management in dialysis patients.

Main Methods:

• Exploratory data analysis and predictive modeling were employed to analyze vital electrolyte data.
• Four distinct predictive models were developed to forecast electrolyte levels based on dialysis parameters.
• Model performance was evaluated to identify the most accurate approach for predicting electrolyte concentrations.

Main Results:

• A decision tree model demonstrated superior performance and accuracy compared to support vector machine, linear regression, and neural network models.
• Key factors influencing electrolyte concentrations were identified as pre-dialysis blood urea nitrogen, pre-weight, dry weight, anticoagulation, and sex.
• The predictive models offer a pathway to fine-tune dialysis dosing for individual patients.

Conclusions:

• The developed predictive models can personalize dialysis dosing, leading to improved patient quality of life and longevity.
• These models have the potential to reduce healthcare costs and resource utilization for both patients and providers.
• Further validation of these findings on a larger scale is recommended to confirm their efficacy.