A clinically applicable approach to continuous prediction of future acute kidney injury

Abstract

The early prediction of deterioration could have an important role in supporting healthcare professionals, as an estimated 11% of deaths in hospital follow a failure to promptly recognize and treat deteriorating patients1. To achieve this goal requires predictions of patient risk that are continuously updated and accurate, and delivered at an individual level with sufficient context and enough time to act. Here we develop a deep learning approach for the continuous risk prediction of future deterioration in patients, building on recent work that models adverse events from electronic health records2-17 and using acute kidney injury-a common and potentially life-threatening condition18-as an exemplar. Our model was developed on a large, longitudinal dataset of electronic health records that cover diverse clinical environments, comprising 703,782 adult patients across 172 inpatient and 1,062 outpatient sites. Our model predicts 55.8% of all inpatient episodes of acute kidney injury, and 90.2% of all acute kidney injuries that required subsequent administration of dialysis, with a lead time of up to 48 h and a ratio of 2 false alerts for every true alert. In addition to predicting future acute kidney injury, our model provides confidence assessments and a list of the clinical features that are most salient to each prediction, alongside predicted future trajectories for clinically relevant blood tests9. Although the recognition and prompt treatment of acute kidney injury is known to be challenging, our approach may offer opportunities for identifying patients at risk within a time window that enables early treatment.

Conflict of interest statement

G.R., H.M. and C.L. are paid contractors of DeepMind. The authors have no other competing interests to disclose.

Figures

Extended Data Figure 1 |. The sequential representation of EHR data.

All EHR data available for each patient was structured into a sequential history for both inpatient and outpatient events in six hourly blocks, shown here as circles. In each 24 hour period events without a recorded time were included in a fifth block. Apart from the data present at the current time step, the models optionally receive an embedding of the previous 48 hours and the longer history of 6 months or 5 years.

Extended Data Figure 2 |. The proposed model architecture.

The best performance was achieved by a multitask deep recurrent highway network architecture on top of an L1-regularised deep residual embedding component that learns the best data representation end-to-end without pre-training.

Extended Data Figure 3 |. Calibration.

a, b, The predictions were recalibrated using isotonic regression before (a) and after (b) calibration. Model predictions were grouped into 20 buckets, with a mean model risk prediction plotted against the percentage of positive labels in that bucket. The diagonal line demonstrates the ideal calibration.

Extended Data Figure 4 |. Analysis of false positive predictions.

a, For prediction of any AKI within 48 h at 33% precision, nearly half of all predictions are trailing, after the AKI has already occurred (orange bars) or early, more than 48 h prior (blue bars). The histogram shows the distribution of these trailing and early false positives for prediction. Incorrect predictions are mapped to their closest preceding or following episode of AKI (whichever is closer) if that episode occurs in an admission. For ±1 day, 15.2% of false positives correspond to observed AKI events within 1 day after the prediction (model reacted too early) and 2.9% correspond to observed AKI events within 1 day before the prediction (model reacted too late). b, Subgroup analysis for all false-positive alerts. In addition to the 49% of false-positive alerts that were made in admissions during which there was at least one episode of AKI, many of the remaining false-positive alerts were made in patients who had evidence of clinical risk factors present in their available electronic health record data. These risk factors are shown here for the proposed model that predicts any stage of AKI occurring within the next 48 h.

Figure 1 |. Illustrative example of risk prediction, uncertainty and predicted future laboratory values.

The first 8 days of admission for a male patient aged 65 with a history of chronic obstructive pulmonary disease. (a) Creatinine measurements showing AKI occurring on day 5. (b) Continuous risk predictions; the model predicted increased AKI risk 48 hours before it was observed. A risk above 0.2, corresponding to 33% precision, was the threshold above which AKI was predicted. Lighter green borders on the risk curve indicate uncertainty, taken as the range of 100 ensemble predictions once trimmed for highest and lowest 5 values. (c) Predictions of the maximum future observed values of creatinine, urea, and potassium.

Figure 2 |. Model performance illustrated by Receiver Operating Characteristic (ROC) and Precision/Recall (PR) curves.

(a) ROC and (b) PR curves for the risk that AKI of any severity will occur within 48 hours. Blue dots: different model operating points (A, 20% precision; C, 33% precision; E, 50% precision; see Extended Data Table 4). Grey shading: area corresponding to operating points with greater than four false positives for each true positive. Blue shading: performance in the more clinically applicable part of the operating space. The model significantly (p-value of <1e-6 outperformed the gradient-boosted tree baseline, shown in (b) for operating point C using two-sided Mann–Whitney U test on 200 samples per model (see Methods).

Figure 3 |. The time between model prediction and actual AKI event.

The models predict AKI risk within a particular time window. Within this the time in hours between prediction and AKI can vary (error bars: bootstrap pivotal 95% confidence intervals; n=200). a, b, Prediction performance for any AKI (a) and AKI stage 3 (b) 48 h ahead of time, shown for different precisions. A greater proportion were correctly predicted closer to the time step immediately prior to the AKI. The available time window for prediction is shortened in AKI events which occur <48 hours after admission; for each column the boxed area shows the upper limit on possible predictions.

Methods-only References