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: TH-PO058

Actionable Forecasting of Kidney Function and Serum Potassium in Hospitalized Adults

Session Information

Category: Acute Kidney Injury

  • 102 AKI: Clinical‚ Outcomes‚ and Trials

Authors

  • Walther, Carl P., Baylor College of Medicine, Houston, Texas, United States
  • Philp, Jamie, Baylor College of Medicine, Houston, Texas, United States
  • Navaneethan, Sankar D., Baylor College of Medicine, Houston, Texas, United States
Background

Derangements in kidney function and electrolytes are common in hospitalizations. Preemptive risk identification for these derangements could enable intervention. We developed forecasting models to inform risk identification using deep learning.

Methods

We obtained data for all adults hospitalized at an academic hospital 1/1/20-1/20/22. We divided hospitalizations into training (70%), validation (15%), and test (15%) sets. From each hospitalization, multiple input/target pairs were created, from which all prior information was used to predict each subsequent day’s creatinine and potassium. For initial modeling, 12 laboratory tests were used. Long short term memory (LSTM) layers were used, and multiple architectures and hyperparameters were assessed with training/validation data. The best performing model was evaluated using test data. Results were compared to naïve baseline predictions (last value carry forward and linear regression).

Results

A total of 219,087 input/target pairs were available. The final model used 4 LSTM layers (64 nodes each), followed by 4 densely connected layers (32 nodes each), a linear layer, and dropout regularization. On test data, the mean absolute error for creatinine was 0.24 mg/dl and for potassium was 0.29 mEq/L. Both outperformed naïve baseline predictions. Interrogation of modeling results in individual hospitalizations from the test set revealed possible patterns that the model learned in forecasting values.

Conclusion

A deep learning model outperformed baseline predictions in forecasting creatinine and potassium levels in hospitalized adults. Next steps will incorporate additional predictive variables and translate the predicted levels to risk estimates for clinically actionable derangements.

Forecasting models to applied to two examples from test set. Predicted values for each day (starting the day after admission) were calculated using all available lab values from all prior days.

Funding

  • NIDDK Support