Abstract: PUB043
Dialysis Facilities Analysis Using Machine Learning
Session Information
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Author
- Ashkar, Ziad Maurice, University of Louisiana at Lafayette, Lafayette, Louisiana, United States
Background
Despite advancements in dialysis, outcomes such as mortality remain poor, highlighting the interplay of clinical, demographic and socioeconomic factors. One approach to understand these complexities is unsupervised machine learning .The objective of this study is to analyze dialysis facilties using cluster analysis, with a particular focus on sociodemographic disparities .
Methods
Dialysis facilities dataset was accessed from https://data.cms.gov. Sociodemographic data at the ZIP Code Tabulation Area (ZCTA) level were obtained from the 2021 American Community Survey via the Census Bureau API. These data were linked to dialysis facilities by ZIP code. PCA transformation was applied to the standardized data, then K means clustering was applied . Python 3.11.6 with pandas, and scikit-learn were used for analysis.
Results
2 distinct clusters were identified . Cluster 0 includes 4609 clinics and cluster 1 includes 2857 clinics. Cluster 1 facilities have higher mean mortality rate,readmission rate, hospitalization rate, anemia rate ,catheter rate, and higher percent with phosphorus >7 mg/dl. These facilties also have lower fistula rates and lower percent with KT/V>=1.2. Facilities in cluster 1 are located in regions with lower mean household income, higher percent of African Americans and Hispanics, higher unemployment rate, and lower percent with college education. Moreover, a higher percent of cluster 1 clinics are for profit( 91.4% vs 88.5%)
Conclusion
Using data from dialysis compare we have identified using unclassified machine a cluster with facilities with lower outcomes and lower quality metrics. Moreover, these facilities are located in areas with higher percentage of Blacks and Hispanics and lower socioeconomic status. Addressing theses disparities requires multifaceted approach and should explore other factors such patient level comorbitidies, staffing models and regional policy differences.