Abstract: FR-PO506
A New Workflow Using Artificial Intelligence to Reduce Peritonitis for Patients Treated with Peritoneal Dialysis
Session Information
- Peritoneal Dialysis: Modality, Catheter, Infections
November 08, 2019 | Location: Exhibit Hall, Walter E. Washington Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Dialysis
- 703 Dialysis: Peritoneal Dialysis
Authors
- Long, Andrew, Fresenius Medical Care, Waltham, Massachusetts, United States
- Han, Hao, Fresenius Medical Care, Waltham, Massachusetts, United States
- Willetts, Joanna, Fresenius Medical Care, Waltham, Massachusetts, United States
- Chaudhuri, Sheetal, Fresenius Medical Care, Waltham, Massachusetts, United States
- Usvyat, Len A., Fresenius Medical Care, Waltham, Massachusetts, United States
- Cameron, Kylee K., Fresenius Medical Care, Waltham, Massachusetts, United States
- Ellison, Brian Christopher, Fresenius Medical Care, Waltham, Massachusetts, United States
- Moran, Judith, Fresenius Medical Care, Waltham, Massachusetts, United States
- Hymes, Jeffrey L., Fresenius Medical Care, Waltham, Massachusetts, United States
- Chatoth, Dinesh K., Fresenius Medical Care, Waltham, Massachusetts, United States
- Maddux, Franklin W., Fresenius Medical Care, Waltham, Massachusetts, United States
Background
~10% of patients with kidney failure are treated with peritoneal dialysis (PD). To reduce peritonitis risk (a major complication of PD), we trained a model to predict which patients were at risk of peritonitis in the next month. The model was trained with an XGBoost classifier and had an area under the receiver operating characteristic curve of 0.66. This abstract outlines a new workflow implemented by a large dialysis provider.
Methods
Starting in Dec 2018, we used the peritonitis predictive model to generate risk scores on a monthly basis for the entire PD population treated by Fresenius Medical Care North America. We scheduled a home visit for patients within the first 30 days and a phone call for patients on dialysis between 30 and 90 days. For the remaining patients, we used the model to segment into three groups: high risk (~10%), medium risk (~10%) and low risk. Those patients with high risk were scheduled for a home visit and those with medium risk were scheduled for a phone call. An assessment was created in Feb 2018 to track specific interventions.
Results
We have performed over 14000 phone calls and over 7000 home visits due to this personalized care effort. Many of the home visits resulted in reviews of infection control, home environment or equipment. More of the phone calls resulted in no intervention than the home visits, which is logical given that those patients have lower risk scores. Analysis of outcomes such as peritonitis rates and modality changes is underway.
Conclusion
Integrating artificial intelligence into clinical decision support allows us to intervene with the right patient at the right time for the right reason. Future work must be conducted to improve the model to include additional reasons for intervention such as psychosocial or educational needs.
Funding
- Commercial Support –