Prospective silent deployment and evaluation of an intelligent machine learning model for prediction of emergency department visits during cancer treatment.
Bibliographic record
Abstract
407 Background: Patients undergoing cancer treatment often need to visit the emergency department (ED), straining the healthcare system. We aim to use long-term electronic health record (EHR) data to deploy and evaluate a previously built warning system designed to accurately identify those at risk of ED visits. This system, once validated, will enable clinicians to take early, personalized actions to prevent these ED visits, saving resources and enhancing the quality of life for cancer patients. Methods: Machine learning models were trained usinglongitudinal retrospective EHR data from patients receiving intravenous systemic therapies for gastrointestinal cancers at the Princess Margaret Cancer Centre, aiming to predict ED visits within 30 days of each treatment. Treatments followed by ED visits within one day were excluded to ensure the system focuses on detecting early warning signs rather than imminent ED visits. The models, including tree-based methods and neural networks, were tuned with Bayesian hyperparameter optimization and calibrated using isotonic regression. A temporal split was applied to establish a held-out retrospective test cohort. The best model was silently deployed for prospective validation in patients with gastrointestinal cancer through our internally developed 'MIRA' platform that supports clinical integration. Within MIRA, the patients' EHR data with treatments scheduled the next day are extracted from the EHR system and forwarded to the model for analysis. Results: In the retrospective cohort from January 1, 2014, to December 31, 2019, 1,997 patients underwent 24,350 treatments, with 2,219 (9.11%) leading to ED visits within 30 days. The top-performing model, an extreme gradient boosting tree, achieved an area under the receiver operating characteristic curve (AUROC) of 0.68 and an area under the precision-recall curve (AUPRC) of 0.19 in the held-out test set. Although the evaluation of the system through prospective silent deployment is ongoing, here we report on patients with treatments during March 2024, with adequate 30-day follow up by April 30th, 2024. During this period, 357 patients received 676 treatments, with ED visits within 30-days following 60 (8.88%) treatments. The deployed system achieved an AUROC of 0.66 (confidence interval: 0.60-0.72) and an AUPRC of 0.22 (confidence interval: 0.15-0.32), which closely aligns with those observed during its retrospective testing. At a 10% alarm rate, model has a positive predictive value of 0.33 and sensitivity of 0.22. Conclusions: During a silent prospective deployment, our system predicted ED visits in cancer patients undergoing medical treatment. These findings indicate that the system should be integrated into the clinical workflow and combined with interventions to prevent ED visits.
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 | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 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".