ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

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

email@asn-online.org

202-640-4660

The Latest on X

Kidney Week

Abstract: SA-PO405

Real-Time Forecasting of Intradialytic Hypotension Using Deep Learning and Multimodal Data Integration

Session Information

Category: Dialysis

  • 801 Dialysis: Hemodialysis and Frequent Dialysis

Authors

  • Luo, Yunfei, University of California San Diego, La Jolla, California, United States
  • Zhao, Siwei, Sanderling Renal Services, Nashville, Tennessee, United States
  • Dasgupta, Subhasis, University of California San Diego, La Jolla, California, United States
  • Rahman, Tauhidur, University of California San Diego, La Jolla, California, United States
  • Malhotra, Rakesh, University of California San Diego, La Jolla, California, United States
Background

Intradialytic hypotension (IDH) is a common and serious complication during hemodialysis and is linked to high morbidity and mortality. Early prediction of IDH can facilitate timely interventions and potentially reduce IDH rates.

Methods

We developed a transformer-based deep learning model to predict IDH events 30-60 minutes before their occurrence. IDH was defined using multiple criteria: Nadir 90, Nadir 100, Fall 20, Fall 30, Fall 20 plus Nadir 90, Fall 30 plus Nadir 90, KDOQI (SBP reduction ≥20 mmHg and associated symptoms), and HEMO (any drop in SBP resulting in an intervention). We utilized real-time physiological data from dialysis sessions and electronic health records (EHR), including demographics, past dialysis records, labs, and comorbidities. The model employed multimodal fusion methods and was trained using a multitask learning approach to account for different IDH definitions. We performed a 5-fold cross-validation by splitting the patients into five disjoint groups. Model performance was assessed using the area under the receiver operating characteristic curve AUROC.

Results

Our study included data from 1,452 patients, encompassing 21,129 dialysis sessions. The mean age was 65 years. The model achieved AUROC scores ranging from 0.85 to 0.94 across different IDH definitions as shown in Fig 1. The model performance was similar with or without integration of EHR module.

Conclusion

We successfully developed a rmodel using available clinical data to predict IDH during dialysis sessions. Future research should focus on validating our model within independent datasets. Implementing this predictive system could facilitate timely therapeutic interventions, and mitiogate the severe consequences of IDH.