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

Please note that you are viewing an archived section from 2025 and some content may be unavailable. To unlock all content for 2025, please visit the archives.

Abstract: FR-PO0481

Dual Attention Transformer for Real-Time Prediction of Intradialytic Hypotension and Hypertension

Session Information

Category: Dialysis

  • 801 Dialysis: Hemodialysis and Frequent Dialysis

Author

  • Guan, Xilin, The Third Affiliated Hospital of Sun Yet-sun University Department of Nephrology, Guangzhou, Guangdong, China
Background

Intradialytic hypotension (IDH) and intradialytic hypertension (IDHTN) are common hemodialysis complications, each independently associated with increased cardiovascular risk and all-cause mortality. Existing predictive models often fail to capture the complex temporal dynamics of real-world hemodialysis data, underscoring the need for more advanced machine learning approaches.

Methods

We conducted a retrospective analysis of 1,301 hemodialysis patients, comprising 177,211 dialysis sessions and 1,165,811 time-stamped data points after preprocessing. The dataset was randomly divided into training (80%), validation (10%), and testing (10%) subsets. Intradialytic hypotension (IDH) and intradialytic hypertension (IDHTN) were defined using standard criteria: IDH-20 as a systolic blood pressure drop of ≥20 mmHg or a mean arterial pressure (MAP) reduction of ≥10 mmHg; IDH-30 as a ≥30 mmHg systolic drop or the same MAP threshold; IDHTN-10 as a systolic increase of ≥10 mmHg; and IDHTN-20 as a ≥20 mmHg increase. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), and F1 score. All patient data were anonymized to ensure privacy.

Results

In the test cohort, which included 130 patients and 70,280 time-stamped data points, the incidences of IDH-20, IDH-30, IDHTN-10, and IDHTN-20 were 15.24%, 9.94%, 18.48%, and 8.03%, respectively. The Transformer model demonstrated strong predictive performance, achieving AUROC values of 0.91 for IDH-20, 0.93 for IDH-30, and 0.89 for both IDHTN-10 and IDHTN-20 (all P < 0.01). Less stringent definitions, such as IDH-20 and IDHTN-10, yielded higher AUPRCs, suggesting that the model effectively captured early or moderate events without compromising classification accuracy. To complement these findings, we trained a LightGBM model using data from 1,024 patients and 63,976 dialysis sessions to predict symptoms or clinical interventions associated with IDH and IDHTN during individual dialysis sessions. This model achieved an AUC of 0.824.

Conclusion

Our Transformer model with a dual attention mechanism demonstrates high accuracy in predicting intradialytic blood pressure abnormalities in real time. This approach holds promise for proactive clinical decision-making and personalized dialysis management.

Digital Object Identifier (DOI)