Abstract: FR-OR03
Development and Validation of a Convolutional Neural Network Model for Intensive Care Unit AKI Prediction
Session Information
- AKI Predictors and Outcomes: Research Abstracts
October 23, 2020 | Location: Simulive
Abstract Time: 05:00 PM - 07:00 PM
Category: Acute Kidney Injury
- 102 AKI: Clinical, Outcomes, and Trials
Authors
- Calvert, Jacob, Dascena, Oakland, California, United States
- Allen, Angier O., Dascena, Oakland, California, United States
- Le, Sidney Hoang, Dascena, Oakland, California, United States
- Palevsky, Paul M., VA Pittsburgh Healthcare System, Pittsburgh, Pennsylvania, United States
- Braden, Gregory Lee, Baystate Medical Center, Springfield, Massachusetts, United States
- Patel, Sharad, Cooper University Health Care, Camden, New Jersey, United States
- Pellegrini, Emily, Dascena, Oakland, California, United States
- Green-Saxena, Abigail, Dascena, Oakland, California, United States
- Hoffman, Jana, Dascena, Oakland, California, United States
- Das, Ritankar, Dascena, Oakland, California, United States
Background
Acute kidney injury (AKI) is common among hospitalized patients and has a significant impact on morbidity and mortality. While early prediction of AKI has the potential to reduce adverse patient outcomes, it remains a difficult condition to predict and diagnose. The purpose of this study was to evaluate the ability of a machine learning algorithm to predict for AKI KDIGO Stage 2 or 3 up to 72 hours in advance of onset using convolutional recurrent neural nets (CNN) and patient Electronic Health Record (EHR) data.
Methods
A CNN prediction system was developed to continuously and automatically monitor for incipient AKI. 7122 patient encounters were retrospectively analyzed from the Medical Information Mart for Intensive Care III (MIMIC-III) database. The CNN machine learning-based AKI prediction model was compared to an established XGBoost AKI prediction model and the Sequential Organ Failure Assessment (SOFA) scoring system. AKI onset was used for the outcome. The model was trained on routinely-collected patient EHR data.
Results
On a hold-out test set, the algorithm attained an Area Under the Receiver Operating Characteristic (AUROC) of 0.85 and PPV of 0.25, relative to a cohort AKI prevalence of 5.21%, for long-horizon AKI prediction at a 72-hour window prior to onset. The ROC curve comparison of 72-hour prediction on the 10% hold-out test set is shown in Figure 1. The CNN model, which was provided text data through Doc2Vec input, outperformed the XGBoost model and the SOFA score.
Conclusion
A CNN machine learning-based AKI prediction model outperforms XGBoost and the SOFA scoring system, demonstrating superior performance in predicting acute kidney injury 72 hours prior to onset, without reliance on changes in serum creatinine.
ROC curve comparison of prediction performance using a CNN classifier, an XGB classifier, and the SOFA score, 72 hours prior to AKI onset on the MIMIC III ICU hold out data set.
Funding
- Other NIH Support