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 Twitter

Kidney Week

Abstract: FR-PO090

Continuous Prediction of Mortality During AKI Renal Replacement Therapy (RRT): A Deep Learning Approach

Session Information

Category: Acute Kidney Injury

  • 101 AKI: Epidemiology‚ Risk Factors‚ and Prevention

Authors

  • Liu, Lucas Jing, University of Kentucky, Lexington, Kentucky, United States
  • Ortiz-Soriano, Victor M., University of Kentucky Medical Center, Lexington, Kentucky, United States
  • Chen, Jin, University of Kentucky, Lexington, Kentucky, United States
  • Neyra, Javier A., University of Alabama at Birmingham, Birmingham, Alabama, United States
Background

AKI-RRT is associated with high risk of mortality. Continuous accurate prediction of mortality in these patients could assist with resource utilization and transition to palliative care when needed. Deep learning (DL) models have shown promise in risk-prediction using temporal EHR data. However, most existing methods fall short in clinical settings where the real-world EHR data are often asynchronous and irregular.

Methods

We developed a novel DL model based on Long Short Term Memory (LSTM) to continuously predict 24-hour mortality risk during AKI-RRT. Our model extends LSTM with two time-aware gates to handle irregular and asynchronous data, and a knowledge-aware gate that uses medical ontology to guide attention between multiple variables at each time step. We used data from 570 adult patients with AKI-RRT admitted to the ICU, excluding patients with ESRD, kidney transplant or those in whom RRT lasted <72h. We utilized 12 temporal features including vital signs and biochemical parameters and 6 static features including demographics, BMI and Charlson score. We assessed our model based on subpopulations according to static features and compared it with four existing models.

Results

Hospital mortality rate was 58.4%. Our model outperformed all others at almost all levels of subpopulations and all metrics for the continuous prediction of mortality during AKI-RRT (Table).

Conclusion

We developed a novel DL model with three additional gates for the continuous prediction of mortality during AKI-RRT. With the new gates, our model achieved better performance for the asynchronous and irregular EHR data. This model can be further enhanced and validated for augmenting bedside clinical decisions.

Funding

  • NIDDK Support