Abstract: SA-PO953
Using Machine Learning to Predict Optimal Renal Replacement Therapy Starts in Patients with Advanced Renal Function Loss
Session Information
- Dialysis: Vascular Access - II
October 27, 2018 | Location: Exhibit Hall, San Diego Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Dialysis
- 704 Dialysis: Vascular Access
Authors
- Fielding, Ollie, pulseData, New York, New York, United States
- Kipers, Chris, pulseData, New York, New York, United States
- Son, Jung Hoon, pulseData, New York, New York, United States
- Lee, Edward, pulseData, New York, New York, United States
- Levine, Daniel, The Rogosin Institute, New York, New York, United States
- Parker, Thomas, The Rogosin Institute, New York, New York, United States
- Smith, Barry H., The Rogosin Institute, New York, New York, United States
- Silberzweig, Jeffrey I., The Rogosin Institute, New York, New York, United States
Background
Building on our previous work using machine learning techniques to identify patients at risk of progression to End Stage Renal Disease (ESRD), we focused this model more precisely on identifying patients with advanced kidney function loss who should plan for optimal renal replacement therapy. The most recent USRDS data indicates that more than 80% of patients begin hemodialysis with a catheter and only 2.5% receive preemptive renal transplants.
Methods
Using longitudinal patient data of 109,028 patients from The Rogosin Institute, we identified a cohort of patients with advanced kidney function loss, defined as an eGFR <20, and built a machine learning model to predict which patients would need to begin renal replacement therapy in the next six months, based on progression to an eGFR <10. Information in the model included patient demographics, vital signs, comorbidities, laboratory values, and medications. We used an algorithm to remove measurements taken during an AKI episode. We evaluated whether the model could identify the need for renal replacement therapy prior to clinical need.
Results
Between 2014 and 2016, only 17 of 214 patients who progressed within a six month period received an AV fistula prior to their decline to an eGFR <10. The model identified 181 patients as the top quintile of risk who would benefit from preparation for renal replacement therapy. Our model has an AUC of 0.93, a sensitivity of 0.81 and a specificity of 0.89 at the top quintile.
Conclusion
We demonstrate improved ability to identify patients who will need renal replacement therapy using an advanced machine learning model incorporating longitudinal data commonly available in EHRs. We plan to augment clinical decision making with machine learning tools.
Numbers for 2014-2016
| Total patient cohort | 109,028 |
| Patients in dataset who progress to an eGFR <20 | 2,416 |
| Patients who continue on to an eGFR <10 | 241 |
| Patients who received an AV fistula in the six months prior to their decline to an eGFR <10 | 17 |
| Patients identified by the model (at the top risk quintile) in the six months prior to a decline to an eGFR <10 | 181 |
Funding
- Commercial Support – pulseData