Deep Learning–Based Recurrent Delirium Prediction in Critically Ill Patients
Bibliographic record
Abstract
OBJECTIVES: To predict impending delirium in ICU patients using recurrent deep learning. DESIGN: Retrospective cohort study. SETTING: Fifteen medical-surgical ICUs across Alberta, Canada, between January 1, 2014, and January 24, 2020. PATIENTS: Forty-three thousand five hundred ten ICU admissions from 38,426 patients. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We used ICU and administrative health data to train deep learning models to predict delirium episodes in the next two 12-hour periods (0-12 and 12-24 hr), starting at 24 hours after ICU admission, and to generate new predictions every 12 hours. We used a comprehensive set of 3,643 features, capturing patient history, early ICU admission information (first 24 hr), and the temporal dynamics of various clinical variables throughout the ICU admission. Our deep learning architecture consisted of a feature embedding, a recurrent, and a prediction module. Our best model based on gated recurrent units yielded a sensitivity of 0.810, a specificity of 0.848, a precision (positive predictive value) of 0.704, and an area under the receiver operating characteristic curve (AUROC) of 0.909 in the hold-out test set for the 0-12-hour prediction horizon. For the 12-24-hour prediction horizon, the same model achieved a sensitivity of 0.791, a specificity of 0.807, a precision of 0.637, and an AUROC of 0.895 in the test set. CONCLUSIONS: Our delirium prediction model achieved strong performance by applying deep learning to a dataset that is at least one order of magnitude larger than those used in previous studies. Another novel aspect of our study is the temporal nature of our features and predictions. Our model enables accurate prediction of impending delirium in the ICU, which can potentially lead to early intervention, more efficient allocation of ICU resources, and improved patient outcomes.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.202 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".