pith. sign in

arxiv: 2605.18839 · v1 · pith:IJCKMZJDnew · submitted 2026-05-13 · 💻 cs.LG · cs.AI

An Integrated Forecasting Prototype for Emergency Department Boarding Time to Support Proactive Operational Decision Making

Orhun Vural , Abdulaziz Ahmed , Ferhat Zengul , James Booth , Bunyamin Ozaydin This is my paper

Pith reviewed 2026-05-20 21:58 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords emergency departmentboarding timetime series forecastingDLinearNLinearMLOpshospital overcrowdingoperational decision making
0
0 comments X

The pith

DLinear and NLinear models accurately forecast emergency department boarding times at multiple future horizons using integrated data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper builds a forecasting system that predicts how long patients will remain boarded in the emergency department while awaiting beds at 6, 8, 10, 12, and 24 hour intervals. It combines records from one university hospital with external details on weather, holidays, and local events to train the models. DLinear and NLinear approaches deliver better accuracy than alternatives tested, and this holds even when boarding times are unusually long due to severe crowding. A web-based MLOps prototype demonstrates how the forecasts can be ingested, visualized, and updated in daily hospital operations to support earlier decisions that ease congestion.

Core claim

The authors establish that DLinear and NLinear time series models trained on hospital data augmented with weather, holiday, and event information achieve superior performance for multi-horizon prediction of ED boarding time and retain accuracy under extreme congestion scenarios, with the results implemented in a practical MLOps web application for operational use.

What carries the argument

Multi-horizon time series forecasting framework that applies DLinear and NLinear deep learning models to hospital records integrated with external contextual data sources.

If this is right

  • Staff can anticipate boarding delays 6 to 24 hours ahead and adjust bed assignments or staffing before congestion builds.
  • Forecasts remain useful during periods of unusually high boarding times, supporting decisions when they matter most.
  • The MLOps prototype enables continuous data ingestion, visual review of predictions, and model retraining in a live setting.
  • Adding external factors such as weather and events measurably improves forecast quality over hospital data alone.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same linear modeling approach could be tested on related hospital metrics such as overall patient length of stay or discharge timing.
  • Running the framework on data from multiple hospitals of varying sizes would clarify how much retraining is needed for new sites.
  • Incorporating additional real-time inputs like current ambulance arrivals might tighten predictions at the shortest horizons.

Load-bearing premise

Patterns learned from data at one university-affiliated urban hospital will continue to produce reliable forecasts when applied at other hospitals or under future conditions.

What would settle it

Retraining the models on boarding time data from a different hospital and finding substantially higher error rates at the same horizons would show the results do not generalize.

Figures

Figures reproduced from arXiv: 2605.18839 by Abdulaziz Ahmed, Bunyamin Ozaydin, Ferhat Zengul, James Booth, Orhun Vural.

Figure 1
Figure 1. Figure 1: summarizes boarding, waiting, treatment, hospital census, and total patient counts including ESI stratified categories, along with associated time-based metrics, temperature, weather categories, and the extreme case indicator. Histograms illustrate empirical distributions, with mean, standard deviation (SD), minimum, and maximum values reported for each feature. Training and Evaluation For training, the la… view at source ↗
read the original abstract

Overcrowding in emergency departments (ED) remains a persistent operational challenge worldwide, causing delays in care delivery and downstream congestion. ED boarding time, defined as the duration admitted patients remain in the ED while awaiting inpatient bed placement, is a key indicator of this congestion. Predicting ED boarding time in advance enables proactive operational decision making before congestion escalates. We developed and evaluated a multi-horizon time series forecasting framework to predict ED boarding time at 6, 8, 10, 12, and 24-hour horizons. Real-world data from a university-affiliated urban hospital in the United States were utilized and integrated with external contextual data sources, including weather, holidays, and major local events. Decomposition-based Linear (DLinear) and Normalization-based Linear (NLinear) time series forecasting deep learning models showed superior performance across multiple horizons. Models were also evaluated under extreme congestion scenarios characterized by elevated boarding times. In addition, a Machine Learning Operations (MLOps) web application prototype was developed to support translation of the forecasting framework into practice through integrated data ingestion, forecast visualization, experimentation, and retraining.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper develops and evaluates a multi-horizon time series forecasting framework to predict emergency department boarding times at 6-, 8-, 10-, 12-, and 24-hour horizons. Real-world data from a single university-affiliated urban hospital are integrated with external sources (weather, holidays, local events). DLinear and NLinear models are reported to show superior performance across horizons, including under extreme congestion scenarios, and the work includes an MLOps web application prototype for data ingestion, visualization, experimentation, and retraining to support proactive operational decisions.

Significance. If the performance claims are substantiated with rigorous metrics and the framework generalizes, this could provide a practical tool for reducing ED overcrowding through proactive bed management. The integration of contextual external data and the emphasis on an operational prototype are strengths that aid translation. The single-site retrospective evaluation, however, limits immediate claims about broader utility.

major comments (2)
  1. Results section: The abstract asserts superior performance of DLinear and NLinear across multiple horizons and under extreme congestion but provides no quantitative metrics, baselines, error bars, data-split details, or statistical tests; this information is load-bearing for the central claim of outperformance and must be presented explicitly with tables or figures.
  2. Evaluation or Discussion section: All reported results derive from retrospective data at one university-affiliated urban hospital; no external validation set, multi-center cohort, or domain-adaptation experiment is described. This is load-bearing for the operational claim that the framework supports proactive decision making at other sites with differing case mix, capacity, and policies.
minor comments (2)
  1. Abstract: Consider adding one or two key performance numbers (e.g., MAE or RMSE at the 12-hour horizon) to allow readers to gauge the magnitude of improvement without reading the full results.
  2. Methods: Clarify the exact train/validation/test split ratios and any temporal blocking used to prevent leakage in the time-series setting.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. We address each major comment below and indicate the revisions made.

read point-by-point responses

  1. Referee: Results section: The abstract asserts superior performance of DLinear and NLinear across multiple horizons and under extreme congestion but provides no quantitative metrics, baselines, error bars, data-split details, or statistical tests; this information is load-bearing for the central claim of outperformance and must be presented explicitly with tables or figures.

    Authors: We agree that explicit quantitative support is required for the performance claims. Although the full Results section contains model evaluations, we have revised the abstract to include key metrics (e.g., MAE and RMSE values for each horizon) and added a dedicated table reporting MAE, RMSE, and MAPE for DLinear, NLinear, and all baselines across the 6-, 8-, 10-, 12-, and 24-hour horizons. Error bars have been added to the performance figures, the chronological train/validation/test split (70/15/15) is now stated explicitly, and we have included statistical tests (paired t-tests with p-values) comparing DLinear/NLinear against baselines. These changes directly substantiate the central claims. revision: yes

  2. Referee: Evaluation or Discussion section: All reported results derive from retrospective data at one university-affiliated urban hospital; no external validation set, multi-center cohort, or domain-adaptation experiment is described. This is load-bearing for the operational claim that the framework supports proactive decision making at other sites with differing case mix, capacity, and policies.

    Authors: We acknowledge that single-site retrospective evaluation limits strong claims of generalizability. We have expanded the Discussion section to explicitly address this limitation, noting potential differences in patient case mix, bed capacity, and hospital policies that could affect performance at other sites. We now discuss domain adaptation as a future direction and state that prospective multi-center validation would be required to support broader operational deployment. We cannot add new external or multi-center data within the scope of this revision. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical forecasting evaluation on held-out data

full rationale

The paper describes a standard empirical machine learning pipeline for multi-horizon time series forecasting of ED boarding times. Models (DLinear, NLinear) are trained on integrated real-world hospital data plus external covariates and evaluated on retrospective held-out periods, including extreme congestion subsets. No equations, derivations, or claims reduce a reported prediction to a fitted parameter by construction, nor does the central performance result depend on self-citation chains, imported uniqueness theorems, or ansatz smuggling. The evaluation protocol is externally falsifiable via standard train-test splits and does not tautologically reproduce its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based on abstract only; no explicit free parameters, axioms, or invented entities are stated. Standard machine-learning assumptions (i.i.d. data splits, stationarity properties of time series, external covariates being predictive) are implicitly used but not enumerated.

pith-pipeline@v0.9.0 · 5740 in / 1054 out tokens · 66983 ms · 2026-05-20T21:58:04.426394+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

46 extracted references · 46 canonical work pages

  1. [1]

    Improving safety and quality with an emergency department overcrowding plan,

    A. Watson and W. P. Stuart, "Improving safety and quality with an emergency department overcrowding plan," Journal of Emergency Nursing, vol. 49, no. 5, pp. 680-693, 2023

    work page 2023

  2. [2]

    Emergency department contributors to ambulance diversion: a quantitative analysis,

    M. J. Schull, K. Lazier, M. Vermeulen, S. Mawhinney, and L. J. Morrison, "Emergency department contributors to ambulance diversion: a quantitative analysis," Annals of emergency medicine, vol. 41, no. 4, pp. 467-476, 2003

    work page 2003

  3. [3]

    Overcrowding in emergency department: causes, consequences, and solutions—a narrative review,

    M. Sartini et al., "Overcrowding in emergency department: causes, consequences, and solutions—a narrative review," in Healthcare, 2022, vol. 10, no. 9: MDPI, p. 1625

    work page 2022

  4. [4]

    Simulation of emergency department operations: A comprehensive review of KPIs and operational improvements,

    L. Vanbrabant, K. Braekers, K. Ramaekers, and I. Van Nieuwenhuyse, "Simulation of emergency department operations: A comprehensive review of KPIs and operational improvements," Computers & Industrial Engineering, vol. 131, pp. 356-381, 2019

    work page 2019

  5. [5]

    Key performance indicators in emergency department simulation: a scoping review,

    M. H. Mehrolhassani, A. Behzadi, and E. Asadipour, "Key performance indicators in emergency department simulation: a scoping review," Scandinavian Journal of Trauma, Resuscitation and Emergency Medicine, vol. 33, no. 1, p. 15, 2025

    work page 2025

  6. [6]

    Emergency department overcrowding: understanding the factors to find corresponding solutions,

    G. Savioli et al. , "Emergency department overcrowding: understanding the factors to find corresponding solutions," Journal of personalized medicine, vol. 12, no. 2, p. 279, 2022

    work page 2022

  7. [7]

    The impact of hospital boarding on the emergency department waiting room,

    C. M. Smalley et al., "The impact of hospital boarding on the emergency department waiting room," JACEP Open, vol. 1, no. 5, pp. 1052-1059, 2020

    work page 2020

  8. [8]

    Strategies to mitigate emergency department crowding and its impact on cardiovascular patients,

    C. W. Baugh, Y . Freund, P. G. Steg, R. Body, D. J. Maron, and M. Y . A. Yiadom, "Strategies to mitigate emergency department crowding and its impact on cardiovascular patients," European Heart Journal: Acute Cardiovascular Care, vol. 12, no. 9, pp. 633-643, 2023

    work page 2023

  9. [9]

    The association between emergency department length of stay and in -hospital mortality in older patients using machine learning: an observational cohort study,

    L. Wu et al., "The association between emergency department length of stay and in -hospital mortality in older patients using machine learning: an observational cohort study," Journal of Clinical Medicine, vol. 12, no. 14, p. 4750, 2023

    work page 2023

  10. [10]

    Measurement of cost of boarding in the emergency department using time -driven activity -based costing,

    M. M. Canellas, M. Jewell, J. L. Edwards, D. Olivier, A. H. Jun-O’Connell, and M. A. Reznek, "Measurement of cost of boarding in the emergency department using time -driven activity -based costing," Annals of Emergency Medicine, vol. 84, no. 4, pp. 376-385, 2024

    work page 2024

  11. [11]

    What is full capacity protocol, and how is it implemented successfully?,

    A. Alishahi Tabriz, S. A. Birken, C. M. Shea, B. J. Fried, and P. Viccellio, "What is full capacity protocol, and how is it implemented successfully?," Implementation Science, vol. 14, no. 1, p. 73, 2019

    work page 2019

  12. [12]

    Impact of artificial intelligence on hospital admission prediction and flow optimization in health services: a systematic review,

    A. L. Nunes, T. Lisboa, B. N. da Rosa, and C. R. Blatt, "Impact of artificial intelligence on hospital admission prediction and flow optimization in health services: a systematic review," International Journal of Medical Informatics, vol. 204, p. 106057, 2025

    work page 2025

  13. [13]

    Artificial intelligence-driven forecasting and shift optimization for pediatric emergency department crowding,

    I. T. Akbasli, A. Z. Birbilen, and O. Teksam, "Artificial intelligence-driven forecasting and shift optimization for pediatric emergency department crowding," JAMIA open, vol. 8, no. 2, 2025

    work page 2025

  14. [14]

    Deep Learning-Based Forecasting of Boarding Patient Counts to Address Emergency Department Overcrowding,

    O. Vural, B. Ozaydin, J. Booth, B. F. Lindsey, and A. Ahmed, "Deep Learning-Based Forecasting of Boarding Patient Counts to Address Emergency Department Overcrowding," in Informatics, 2025, vol. 12, no. 3: MDPI, p. 95

    work page 2025

  15. [15]

    AI-powered models for overcrowding prediction at TUMS hospitals,

    M. J. Shaibani, A. A. Fazaeli, R. Daroudi, M. Radinmanesh, and S. Emamgholipour Sefiddashti, "AI-powered models for overcrowding prediction at TUMS hospitals," Scientific Reports, 2025

    work page 2025

  16. [16]

    Leveraging a Large Language Model (LLM) to Predict Hospital Admissions of Emergency Department Patients,

    V . Pasquadibisceglie, A. Appice, D. Malerba, and G. Fiameni, "Leveraging a Large Language Model (LLM) to Predict Hospital Admissions of Emergency Department Patients," Expert Systems with Applications, p. 128224, 2025

    work page 2025

  17. [17]

    Work systems analysis of emergency nurse patient flow management using the systems engineering initiative for patient safety model: Applying findings from a grounded theory study,

    E. Benjamin and K. K. Giuliano, "Work systems analysis of emergency nurse patient flow management using the systems engineering initiative for patient safety model: Applying findings from a grounded theory study," JMIR Human Factors, vol. 11, no. 1, p. e60176, 2024

    work page 2024

  18. [18]

    Forecasting Pediatric Emergency Department Arrivals: Evaluating the Role of Exogenous Variables Using Deep Learning Models,

    E.-E. Etu et al., "Forecasting Pediatric Emergency Department Arrivals: Evaluating the Role of Exogenous Variables Using Deep Learning Models," Intelligence-Based Medicine, p. 100313, 2025

    work page 2025

  19. [19]

    Enhanced forecasting of emergency department patient arrivals using feature engineering approach and machine learning,

    B. M. Porto and F. S. Fogliatto, "Enhanced forecasting of emergency department patient arrivals using feature engineering approach and machine learning," BMC Medical Informatics and Decision Making, vol. 24, no. 1, p. 377, 2024

    work page 2024

  20. [20]

    Predicting patient waiting time in the queue system using deep learning algorithms in the emergency room,

    H. Hijry and R. Olawoyin, "Predicting patient waiting time in the queue system using deep learning algorithms in the emergency room," International Journal of Industrial Engineering, vol. 3, no. 1, pp. 33-45, 2021

    work page 2021

  21. [21]

    Predicting Inpatient Admissions From Emergency Department Triage Using Machine Learning: A Systematic Review,

    E. L. Williams, D. Huynh, M. Estai, T. Sinha, M. Summerscales, and Y . Kanagasingam, "Predicting Inpatient Admissions From Emergency Department Triage Using Machine Learning: A Systematic Review," Mayo Clinic Proceedings: Digital Health, p. 100197, 2025

    work page 2025

  22. [22]

    Prospective comparison of econometric, machine learning, and foundation models for forecasting emergency department boarding patients,

    L. Poursoltan et al., "Prospective comparison of econometric, machine learning, and foundation models for forecasting emergency department boarding patients," npj Health Systems, vol. 2, no. 1, p. 49, 2025

    work page 2025

  23. [23]

    Models to predict length of stay in the emergency department: a systematic literature review and appraisal,

    R. M. Farimani et al., "Models to predict length of stay in the emergency department: a systematic literature review and appraisal," BMC Emergency Medicine, vol. 24, no. 1, p. 54, 2024

    work page 2024

  24. [24]

    Leveraging machine learning and rule extraction for enhanced transparency in emergency department length of stay prediction,

    W. A. Sulaiman et al. , "Leveraging machine learning and rule extraction for enhanced transparency in emergency department length of stay prediction," Frontiers in Digital Health, vol. 6, p. 1498939, 2025

    work page 2025

  25. [25]

    An integrated approach of machine learning and systems thinking for waiting time prediction in an emergency department,

    Y.-H. Kuo et al. , "An integrated approach of machine learning and systems thinking for waiting time prediction in an emergency department," International journal of medical informatics, vol. 139, p. 104143, 2020

    work page 2020

  26. [26]

    Using artificial intelligence to predict patient wait times in the emergency department: A scoping review,

    T. Gloyn et al., "Using artificial intelligence to predict patient wait times in the emergency department: A scoping review," Artificial Intelligence in Medicine, p. 103316, 2025

    work page 2025

  27. [27]

    Using Long Short -Term Memory (LSTM) neural networks to predict emergency department wait time,

    N. Cheng and A. Kuo, "Using Long Short -Term Memory (LSTM) neural networks to predict emergency department wait time," in The Importance of Health Informatics in Public Health during a Pandemic : IOS Press, 2020, pp. 199-202

    work page 2020

  28. [28]

    Emergency department boarding: a descriptive analysis and measurement of impact on outcomes,

    L. Salehi et al. , "Emergency department boarding: a descriptive analysis and measurement of impact on outcomes," Canadian Journal of Emergency Medicine, vol. 20, no. 6, pp. 929-937, 2018

    work page 2018

  29. [29]

    Patterns in Duration of Emergency Department Boarding and Variation by Sociodemographic Factors,

    C. K. Prucnal et al. , "Patterns in Duration of Emergency Department Boarding and Variation by Sociodemographic Factors," Western Journal of Emergency Medicine, vol. 26, no. 6, p. 1640, 2025

    work page 2025

  30. [30]

    A rapid admission protocol to reduce emergency department boarding times,

    R. Amarasingham, T. Swanson, D. Treichler, S. Amarasingham, and W. Reed, "A rapid admission protocol to reduce emergency department boarding times," BMJ Quality & Safety, vol. 19, no. 3, pp. 200-204, 2010

    work page 2010

  31. [31]

    Models and insights for hospital inpatient operations: Time-dependent ED boarding time,

    P. Shi, M. C. Chou, J. G. Dai, D. Ding, and J. Sim, "Models and insights for hospital inpatient operations: Time-dependent ED boarding time," Management Science, vol. 62, no. 1, pp. 1-28, 2016

    work page 2016

  32. [32]

    Multisite evaluation of prediction models for emergency department crowding before and during the COVID-19 pandemic,

    A. J. Smith et al., "Multisite evaluation of prediction models for emergency department crowding before and during the COVID-19 pandemic," Journal of the American Medical Informatics Association, vol. 30, no. 2, pp. 292-300, 2023

    work page 2023

  33. [33]

    History Bulk

    "History Bulk." OpenWeather. https://openweathermap.org/history-bulk (accessed 2025)

    work page 2025

  34. [34]

    Federal Holidays

    "Federal Holidays." United States Office of Personnel Management. https://www.opm.gov/policy-data- oversight/pay-leave/federal-holidays/ (accessed 2025)

    work page 2025

  35. [35]

    Football Schedule

    "Football Schedule." Alabama Athletics - Official Athletics Website. https://rolltide.com/sports/football/schedule (accessed 2025)

    work page 2025

  36. [36]

    Football Schedule

    "Football Schedule " Auburn Tigers - Official Athletics Website. https://auburntigers.com/sports/football/schedule (accessed 2025)

    work page 2025

  37. [37]

    An Artificial Intelligence–Based Framework for Predicting Emergency Department Overcrowding: Development and Evaluation Study,

    O. Vural, B. Ozaydin, K. Y . Aram, J. Booth, B. F. Lindsey, and A. Ahmed, "An Artificial Intelligence–Based Framework for Predicting Emergency Department Overcrowding: Development and Evaluation Study," JMIR Medical Informatics, vol. 13, p. e73960, 2025

    work page 2025

  38. [38]

    Long-term forecasting with tide: Time-series dense encoder,

    A. Das, W. Kong, A. Leach, S. Mathur, R. Sen, and R. Y u, "Long -term forecasting with tide: Time -series dense encoder," arXiv preprint arXiv:2304.08424, 2023

    work page arXiv 2023

  39. [39]

    Are transformers effective for time series forecasting?,

    A. Zeng, M. Chen, L. Zhang, and Q. Xu, "Are transformers effective for time series forecasting?," in Proceedings of the AAAI conference on artificial intelligence, 2023, vol. 37, no. 9, pp. 11121-11128

    work page 2023

  40. [40]

    Temporal fusion transformers for interpretable multi -horizon time series forecasting,

    B. Lim, S. Ö. Arık, N. Loeff, and T. Pfister, "Temporal fusion transformers for interpretable multi -horizon time series forecasting," International journal of forecasting, vol. 37, no. 4, pp. 1748-1764, 2021

    work page 2021

  41. [41]

    TSTPlus

    I. Oguiza. "TSTPlus." Github. https://timeseriesai.github.io/tsai/models.tstplus.html (accessed

    work page

  42. [42]

    An Artificial Intelligence-Based Framework for Predicting Emergency Department Overcrowding: Development and Evaluation Study,

    O. Vural, B. Ozaydin, K. Y . Aram, J. Booth, B. F. Lindsey, and A. Ahmed, "An Artificial Intelligence-Based Framework for Predicting Emergency Department Overcrowding: Development and Evaluation Study," arXiv preprint arXiv:2504.18578, 2025

    work page arXiv 2025

  43. [43]

    Scikit-learn: Machine learning in Python,

    F. Pedregosa et al., "Scikit-learn: Machine learning in Python," the Journal of machine Learning research, vol. 12, pp. 2825-2830, 2011

    work page 2011

  44. [44]

    Optuna: A next -generation hyperparameter optimization framework,

    T. Akiba, S. Sano, T. Yanase, T. Ohta, and M. Koyama, "Optuna: A next -generation hyperparameter optimization framework," in Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, 2019, pp. 2623-2631

    work page 2019

  45. [45]

    PostgreSQL 18.1 Documentation

    P. G. D. Group. "PostgreSQL 18.1 Documentation." https://www.postgresql.org/docs/18/index.html (accessed

    work page

  46. [46]

    RabbitMQ Documentation

    "RabbitMQ Documentation." RabbitMQ. https://www.rabbitmq.com/docs (accessed 2026). Supplementary Figure S1. Data Stream tab of the ED overcrowding system dashboard. Figure S2. Dashboard tab of the ED overcrowding system dashboard. Figure S3. Dashboard tab of the ED overcrowding system dashboard. Figure S4. Retraining tab of the ED overcrowding system dash...

    work page 2026