MétaCan
Menu
← Back to cohort
Record W7020970830

Modèles d'apprentissage automatique pour prédire et expliquer la congestion à l'urgence

2024· other· fr· W7020970830 on OpenAlexaboutno aff

Bibliographic record

VenuePolyPublie (École Polytechnique de Montréal) · 2024
Typeother
Languagefr
Field
Topic
Canadian institutionsnot available
Fundersnot available
KeywordsLimitingNucleofectionContext (archaeology)Filter (signal processing)
DOInot available

Abstract

fetched live from OpenAlex

RÉSUMÉ: La congestion des services d’urgence est un problème mondial pressant exacerbé par une demande croissante de soins de santé et des ressources limitées, surtout dans le contexte de la pandémie de COVID-19. Générée par un déséquilibre entre les besoins et les ressources, la congestion accentue à son tour la demande en augmentation la morbidité et la mortalité chez les patients, en plus de dépléter les ressources en épuisant le personnel soignant. C’est dans ce contexte qu’est née la collaboration avec l’équipe du Centre Hospitalier de l’Université de Montréal. Avec l’objectif principal de prédire la congestion 24 heures et 48 heures d’avance, nous introduisons une approche novatrice en utilisant le score EDWIN comme mesure objective de la congestion et comme variable dépendante à prédire. Les approches traditionnelles pour évaluer la congestion se basent souvent sur des proxy, tels que l’achalandage ou le nombre d’admissions, plutôt que d’évaluer directement le phénomène de congestion lui-même. En intégrant les notions de temporalité et de logistique hospitalière, notre étude s’est concentrée sur les délais logistiques pour obtenir les services nécessaires à l’urgence, tels les examens diagnostiques et délais de visite médicales, comme variables indépendantes. Reconnaissant la nature dynamique dans le temps de ces variables, des délais spécifiques à chaque variable ont été ajustés à travers une fonction de corrélation croisée pour prédire la congestion. En exploitant des modèles d’apprentissage automatique et d’optimisation, notamment des arbres de décision, forêts aléatoires, arbres boostés par gradient et arbres de classification optimaux, nous prédisons la congestion dans 24 et 48 heures, atteignant des taux de précision élevés de 90,1 % et 89.9 %, respectivement ainsi que des R2 ajustés de 84.9$. De plus, l’intégration de délais spécifiques aux variables à travers une fonction de corrélation croisée améliore significativement les performances des modèles. Notre analyse révèle le rôle pivot des score de congestion passés dans la prédiction des occurrences futures et met en lumière la nature dynamique des étiologies de la congestion. Grâce aux valeurs SHAP et aux valeurs contrefactuelles, nous montrons une variabilité entre les variables importantes à un niveau global et local ainsi qu’une variabilité dynamique dans le temps. Nous présentons un outil prédictif couplé à des méthodes d’intelligence artificielle explicables, offrant des aperçus sur les causes quotidiennes de congestion et aidant les administrateurs dans l’allocation et la planification des ressources. En comblant un vide critique dans la prédiction directe de la congestion et en fournissant des insights exploitables, notre étude contribue à améliorer les pratiques de gestion des services d’urgence et, ultimement, les soins aux patients. ABSTRACT: Emergency department overcrowding is a pressing global problem exacerbated by growing healthcare demand and limited resources, especially in the aftermath of the COVID-19 pande-mic. Generated by an imbalance between needs and resources, overcrowding, in turn, exacer-bates demand by increasing patient morbidity and mortality, as well as depleting resources by exhausting healthcare personnel. Against this backdrop, collaboration with the Centre Hospitalier de l’Université de Montréal team was born. With the primary objective of predicting overcrowding 24 and 48 hours in advance, we are introducing an innovative approach using the EDWIN score as an objective measure of over-crowding and as the dependent variable to be predicted. Traditional approaches to assessing overcrowding often rely on proxies, such as ridership or number of admissions, rather than directly assessing the overcrowding phenomenon itself. By integrating temporality and hospital logistics notions, our study focused on logistical delays in obtaining necessary emergency services, such as diagnostic tests and medical visit times, as independent variables. Recognizing the time-dynamic nature of these variables, variable-specific delays were fitted through a crosscorrelation function to predict overcrow-ding. By exploiting machine learning and optimization models, including decision trees, random forests, gradient-boosted trees, and optimal classification trees, we predict overcrowding in 24 and 48 hours, achieving high accuracy rates of 90.1% and 89.9%, respectively, as well as adjusted R2 of 84.9%. In addition, the integration of variable-specific lags through a crosscorrelation function was found. Our analysis reveals the pivotal role of past overcrowding scores in predicting future occur-rences and highlights the dynamic nature of overcrowding etiologies. Using SHAP and counterfactual values, we show variability between important variables at a global and local level, as well as dynamic variability over time. We present a predictive tool coupled with explainable artificial intelligence methods, offering insights into the day-to-day causes of overcrowding and assisting administrators in resource allocation and planning. By filling a critical gap in the direct prediction of overcrowding and providing actionable insights, our study contributes to improving emergency department management practices and, ultimately, patient care.

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 machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.003
metaresearch head score (Gemma)0.008
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Simulation or modeling · Consensus signal: Simulation or modeling
GenreCandidate signal: Empirical · Consensus signal: none
Teacher disagreement score0.056
Threshold uncertainty score0.112

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0030.008
Meta-epidemiology (narrow)0.0010.001
Meta-epidemiology (broad)0.0010.002
Bibliometrics0.0010.001
Science and technology studies0.0010.001
Scholarly communication0.0030.002
Open science0.0020.001
Research integrity0.0020.002
Insufficient payload (model declined to judge)0.0120.002

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.

Opus teacher head0.014
GPT teacher head0.259
Teacher spread0.245 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designSimulation or modeling
Domainnot available
GenreEmpirical

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".

Quick stats

Citations0
Published2024
Admission routes1
Has abstractyes

Explore more

Same venuePolyPublie (École Polytechnique de Montréal)→French-language works237,207→