[Paper Review] Continual Prediction from EHR Data for Inpatient Acute Kidney Injury

Rohit J. Kate, Noah Pearce+2 more|arXiv (Cornell University)|Feb 26, 2019

Acute Kidney Injury ResearchMedicine26 references3 citations

TL;DR

This paper proposes a continual prediction framework for inpatient acute kidney injury (AKI) using electronic health record (EHR) data, updating predictions in real time as new patient data becomes available. By leveraging the latest values of AKI-relevant variables at every data update, the model achieves an AUC of 0.724—significantly outperforming traditional one-time prediction models that fix predictions at specific time points.

ABSTRACT

Acute kidney injury (AKI) commonly occurs in hospitalized patients and can lead to serious medical complications. In order to optimally predict AKI before it develops at any time during a hospital stay, we present a novel framework in which AKI is continually predicted automatically from EHR data over the entire hospital stay instead of at only one particular time. The continual model predicts AKI every time a patients AKI-relevant variable changes in the EHR. Thus the model is not only independent of a particular time for making predictions, but it can also leverage the latest values of all the AKI-relevant patient variables for making predictions. Using data of 44,691 hospital stays of duration longer than 24 hours we evaluated our continual prediction model and compared it with the traditional one-time prediction models. Excluding hospitals stays in which AKI occurred within 24 hours from admission, the one-time prediction model predicting at 24 hours from admission obtained area under ROC curve (AUC) of 0.653 while the continual prediction model obtained AUC of 0.724. The one-time prediction model that predicts at 24 hours obviously cannot predict AKI incidences that occur within 24 hours of admission which when included in the evaluation reduced its AUC to 0.57. In comparison, the continual prediction model had AUC of 0.709. The continual prediction model also did better than all other one-time prediction models predicting at other fixed times. By being able to take into account the latest values of AKI-relevant patient variables and by not being limited to a particular time of prediction, the continual prediction model out-performed one-time prediction models in predicting AKI.

Motivation & Objective

• To address the limitation of one-time prediction models that fix predictions at a single time point during hospitalization.
• To improve early detection of acute kidney injury (AKI) by enabling continuous, real-time prediction throughout the entire hospital stay.
• To leverage the most up-to-date clinical data from EHRs at every patient data update for dynamic risk assessment.
• To evaluate the performance of continual prediction against traditional one-time models across varying time windows.

Proposed method

• The framework continuously updates AKI risk predictions every time a new AKI-relevant variable is recorded in the EHR.
• It uses a machine learning model trained on 44,691 inpatient stays with durations longer than 24 hours.
• The model dynamically incorporates the latest values of vital signs, lab results, and other clinical variables relevant to AKI.
• Prediction is not constrained to a fixed time (e.g., 24 hours post-admission), allowing early detection of AKI onset.
• The approach avoids the bias of fixed-timing models by evaluating performance across the entire hospitalization timeline.
• Model performance is evaluated using the area under the ROC curve (AUC), comparing continual prediction to one-time predictions at various fixed time points.

Experimental results

Research questions

• RQ1 Can continual prediction from EHR data improve early detection of inpatient acute kidney injury compared to one-time prediction models?
• RQ2 How does the performance of a continually updated model compare to fixed-time prediction models when AKI occurs within the first 24 hours of hospitalization?
• RQ3 Does leveraging the most recent clinical data at every update lead to better predictive performance than static predictions at a single time point?
• RQ4 How does the continual model perform across different time windows of hospitalization, especially when early AKI cases are included?

Key findings

• The continual prediction model achieved an AUC of 0.724 when evaluating AKI prediction over the entire hospital stay, outperforming one-time models.
• The one-time model predicting at 24 hours post-admission achieved an AUC of 0.653, but dropped to 0.57 when early AKI cases (within 24 hours) were included.
• The continual model maintained strong performance, achieving an AUC of 0.709 even when including AKI cases that occurred within the first 24 hours of admission.
• The continual model outperformed all one-time prediction models that fixed predictions at other time points.
• The framework successfully captured early AKI onset by updating predictions with every new data point, reducing the risk of missing early-onset cases.
• The results demonstrate that continual prediction using real-time EHR updates significantly enhances predictive accuracy for inpatient AKI.