Abstract: PO0907
Prediction of Severe Gastrointestinal Bleeding Events in Hemodialysis: Collaborative Development of Machine Learning Model Within INSPIRE
Session Information
- Leveraging Technology and Innovation to Predict Events and Improve Dialysis Delivery
November 04, 2021 | Location: On-Demand, Virtual Only
Abstract Time: 10:00 AM - 12:00 PM
Category: Dialysis
- 701 Dialysis: Hemodialysis and Frequent Dialysis
Authors
- Lama, Suman Kumar, Fresenius Medical Care, Global Medical Office, Waltham, Massachusetts, United States
- Chaudhuri, Sheetal, Fresenius Medical Care, Global Medical Office, Waltham, Massachusetts, United States
- Willetts, Joanna, Fresenius Medical Care, Global Medical Office, Waltham, Massachusetts, United States
- Larkin, John W., Fresenius Medical Care, Global Medical Office, Waltham, Massachusetts, United States
- Winter, Anke, Fresenius Medical Care, Global Medical Office, Bad Homburg, Germany
- Stauss-Grabo, Manuela, Fresenius Medical Care, Global Medical Office, Bad Homburg, Germany
- Usvyat, Len A., Fresenius Medical Care, Global Medical Office, Waltham, Massachusetts, United States
- Hymes, Jeffrey L., Fresenius Medical Care, Global Medical Office, Waltham, Massachusetts, United States
- Maddux, Franklin W., Fresenius Medical Care AG und Co KGaA, Bad Homburg, Hessen, Germany
- Wheeler, David C., University College London, London, United Kingdom
- Stenvinkel, Peter, Karolinska Institutet, Stockholm, Stockholm, Sweden
- Floege, Jürgen, RWTH Aachen University Hospital, Aachen, Germany
Group or Team Name
- On behalf of the INSPIRE Core Group
Background
INitiativeS on advancing Patients’ outcomes In REnal disease (INSPIRE) is an academia and industry collaboration set forth to identify critical investigations/models needed to advance the practice of nephrology. As an inaugural effort, INSPIRE group aims to develop a machine learning (ML) model that can identify a hemodialysis (HD) patient’s 30-day risk for hospitalization due to gastrointestinal (GI) bleeding.
Methods
We used data from adult (age ≥18 years) HD patients (Jan 2017-Dec 2020) in the United States to build a XGBoost model considering 2,292 variables for classification of 30-day GI bleed hospitalization risk. Data were randomly split in 50%:20%:30% ratio for model training, validation, and testing. Unseen data by model (testing) was used for assessing performance via area under the curve (AUC) and feature importance of predictors via Shapley (SHAP) values.
Results
Among 58,187 HD patients included in the testing dataset, 1150 had a GI bleed hospitalization. ML model showed AUC=0.67 and top predictors of a GI bleed hospitalization in 30 days were the minimum hemoglobin level in prior 180 days, time since prior GI bleed hospitalization, and higher vitamin D levels (Figure 1).
Conclusion
ML model appears to have suitable performance for identifying a patient’s 30-day risk for GI bleed hospitalization. Albeit further model iterations/tuning are needed, ML techniques that account for collinearity and missingness hold promise for early detection of potentially avoidable GI bleeding admissions. Model identified an important association between higher vitamin D levels and GI bleeding events, which is consistent with the increasing evidence suggesting antithrombotic and anticoagulant actions of vitamin D derivatives.
Funding
- Commercial Support – Fresenius Medical Care