ASN's Mission

ASN leads the fight to prevent, treat, and cure kidney diseases throughout the world by educating health professionals and scientists, advancing research and innovation, communicating new knowledge, and advocating for the highest quality care for patients.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: FR-PO791

Using Artificial Intelligence to Help Predict Imminent Hospitalizations in Patients with ESRD

Session Information

Category: Dialysis

  • 701 Dialysis: Hemodialysis and Frequent Dialysis

Authors

  • Long, Andrew, Fresenius Medical Care North America, Waltham, Massachusetts, United States
  • Blanchard, Tommy C., Fresenius Medical Care North America, Waltham, Massachusetts, United States
  • Willetts, Joanna, Fresenius Medical Care North America, Waltham, Massachusetts, United States
  • Han, Hao, Fresenius Medical Care North America, Waltham, Massachusetts, United States
  • O'Connell, Michael Ryan, Fresenius Medical Care North America, Waltham, Massachusetts, United States
  • Ward, Amanda K., Fresenius Health Partners, Austin, Texas, United States
  • Reighter, Lisa, Fresenius Health Partners, Austin, Texas, United States
  • Braverman, Scott, Fresenius Health Partners, Austin, Texas, United States
  • Smith, Kendra N., Fresenius Health Partners, Austin, Texas, United States
  • Baker, Annette L., Fresenius Health Partners, Austin, Texas, United States
  • Brown, Gamaela Dee, Fresenius Health Partners, Austin, Texas, United States
  • Garza, Greg S., Fresenius Health Partners, Austin, Texas, United States
  • Gallagher, Cara, Fresenius Health Partners, Austin, Texas, United States
  • Conti, Jodi, Fresenius Health Partners, Austin, Texas, United States
  • Chaudhuri, Sheetal, Fresenius Medical Care North America, Waltham, Massachusetts, United States
  • Usvyat, Len A., Fresenius Medical Care North America, Waltham, Massachusetts, United States
  • Ketchersid, Terry L., Fresenius Medical Care North America, Waltham, Massachusetts, United States
  • Maddux, Franklin W., Fresenius Medical Care North America, Waltham, Massachusetts, United States
Background

Patients with end-stage renal disease are hospitalized two times per year on average; approximately 35% have a re-admission within 30 days of discharge (USRDS 2017). The cost of hospitalizations represents about one-third of the total Medicare spending for patients on dialysis. We developed a model to predict which patients treated at a large dialysis provider are at imminent risk of hospitalization to highlight patients who might benefit from additional interventions.

Methods

We built a machine learning model using over 1500 variables to predict the probability that a patient would be admitted within 7 days of the current outpatient dialysis treatment. Some of the variables in the model include treatment vital signs, administered medications, lab values, prior hospitalizations, demographics, comorbidities, lifestyle, and free-text clinical notes. Training data was extracted from patients treated at a large dialysis provider between January 2016 through May 2017. The results shown below are from un-seen test data with 200,000 patients in June 2017.

Results

Approximately 2.7% of patients are hospitalized weekly. Results from the test data show the model has an area under the receiver operating curve (AUROC) of 0.78, sensitivity of 69%, and specificity of 72%. The confusion matrix for test data is shown in Table 1. Top variables for predicting hospitalization are related to prior hospitalizations, the content of free-text clinical notes, serum albumin, blood pressure, interdialytic weight gain, and hemoglobin.

Conclusion

This work demonstrates that data routinely collected during dialysis treatments can be used to predict imminent hospitalization. We are currently pilot-testing this model to determine if surfacing the results of these real time artificial intelligence-based machine learning algorithms to the local care team can help patients avoid hospitalizations and lead to improved patient outcomes.

Confusion matrix for test data (n = 200000)
  Actual admit in 7 days
  PositiveNegative
Predicted admit in 7 daysPositive1.8%27.2%
Negative0.8%70.2%

Funding

  • Commercial Support