Introduction:
Dry weight is defined as the lowest tolerated post dialysis weight achieved at which there are minimal signs or symptoms of hypovolemia or hypervolemia. Maintaining the desired dry weight is an effective strategy in managing the complications related to CKD and to maintain normal blood pressure among hypertensive patients on dialysis. On initiation, finding the dry weight may take a few sessions. Machine learning has been increasingly being used in diagnosing and predicting illnesses trends based on evaluating large samples of multiple parameters and background characteristics of patients. We aimed to use these skills to predict the dry weight change trend in Hemodialysis patients without the constant need of a Nephrologist supervising each dialysis session in a countries where the doctor patient ratio is yet to strike a balance.
Methods:
All medical data are taken from the electronic medical record of 20,800 dialysis sessions of 172 Asian patients in a single dialysis centre in India between July’2023 to June’2024 meeting the inclusion criteria. Several predictive models were developed using forest classifier to predict the trend of dry weight change in each dialysis session. The training model was trained using 75% of dataset obtained from 129 patients and remaining dataset from 43 patients were used as the test model to test the predictability of dry weight trend.
Results:
The 2 models of actual dry weight change and the predicted model for upward and downward trends were compared and showed the upward predictability of dry weight by machine learning with an accuracy of 0.60, precision of 0.04 and recall as 0.78. Downward trend showed accuracy of 0.68, precision of 0.06 and recall of 0.75. The major predictors of dry weight changes were hypotension, muscular cramps and edema amongst many others.
Conclusions:
The random forest classifier could prove to be helpful in predicting the dry weight change trend and help in the decision making in clinical settings in dialysis centres. As we live in an era of advanced computations and artificial intelligence, the use of these powerful methods can be applied to medicine in a productive way and may lead to great advances in healthcare practices and management skills
I have no potential conflict of interest to disclose.
I did not use generative AI and AI-assisted technologies in the writing process.