![Knowledge](../images/knowledge_off.png)
![Community](../images/community_off.png)
![Influence](../images/influence_off.png)
Salon 2
Testing of Prediction Models for End Stage Kidney Disease Patient Nonadherence to Renal Replacement Treatment Regimens (303282)
Jane Brzozowski, Fresenius Medical Care North AmericaSheetal Chaudhuri, Fresenius Medical Care North America
Jodi Conti, Fresenius Medical Care North America
*Yue Jiao, Fresenius Medical Care North America
Zuwen Kuang, Fresenius Medical Care North America
Hanjie Zhang, Renal Research Institute
Keywords: Healthcare Information Technology, Predictive Modeling, Kidney Disease, Dialysis, Noncompliance
In patients with end stage kidney disease (ESKD), renal replacement via hemodialysis (HD) treatments three times per week assumes some functions of the diseased kidney by filtering the body’s toxins from the blood, and is required to sustain life. Nonadherence with HD regimens is known to be associated with increased morbidity and mortality. This endeavor aimed to utilize clinical and nonclinical data to develop and test predictive models (PMs) that identify patients with a high probability of not attending their HD treatments within the following week. Data from Fresenius Medical Care patients during 2014 was analyzed. Various PMs were investigated including generalized linear model, partitioning and regression trees, artificial neural networks, and generalized additive model (GAM). 1,554,833 clinical records stratified in weekly intervals on 60 variables from 172,854 patients was utilized for development and testing. The GAM had the highest performance, with an average area under the curve (AUC) of 0.87 for five multitier models utilizing a 30% test data set. This analysis demonstrates that PMs can assist in identifying patients with a high probability for missing HD treatments.