Predicting delays in antibiotic administration in the emergency department: a machine learning approach incorporating nursing workload and crowding factors

Background: Emergency department (ED) crowding is a well-documented issue that significantly contributes to delays in critical care interventions, including antibiotic administration. Although previous studies have explored the effects of crowding, the specific role of nursing workload in such delays remains underexplored. This study aimed to develop a machine learning (ML) model to predict delays in antibiotic administration by integrating nursing workload data from electronic health records (EHRs) alongside ED crowding metrics. Methods: We conducted a retrospective analysis of EHR data from a single-center ED, focusing on nursing-specific workload indicators such as the frequency of nursing procedures. Models were developed using three variable groups (National Emergency Department Overcrowding Scale (NEDOCS)-only, workload-only, and combined NEDOCS/workload) across three ML algorithms (Poisson regression, Random Forest (RF) and eXtreme Gradient Boosting (XGBoost)). Each developed model was evaluated on an unseen test dataset using performance metrics, including root mean square error (RMSE), adjusted R2 and mean absolute error (MAE). Results: A total of 63,831 ED visits were recorded during the study period, with an average of 0.83 instances of delayed antibiotic administration occurring per hour (approximately once every 50 minutes). Models incorporating workload-related variables consistently outperformed those using NEDOCS-only variables. The combined NEDOCS/workload models demonstrated the best performance, with both the RF and XGBoost models yielding RMSE = 0.907, adjusted R2 = 0.120 and MAE = 0.712 on the test dataset. XGBoost was selected as the best model owing to its computational efficiency and interpretability. Conclusions: To the best of our knowledge, this is the first study to integrate nursing workload data into an ML model to predict delays in antibiotic administration in the ED. The study findings underscore the significant effect of nursing workload on timely care delivery, suggesting that alleviating nursing workload could reduce delays in antibiotic administration and improve patient outcomes.

Crowding; Nursing workload; Delays in treatment; Delays in antibiotic administration; Machine learning; Emergency department; NEDOCS; NASA-TLX

[1] Chang H, Ko E, Lee JH, Kim M, Kim T, Shin TG, et al. Emergency department crowding: a national data report. Clinical and Experimental Emergency Medicine. 2024; 11: 331–334.

[2] Lee E, Lim JY, Lee DH, Lee J II. Correlations between modified early warning scores in emergency departments and predictions of prognosis in South Korea. Signa Vitae. 2024; 20: 49–60.

[3] Eidstø A, Ylä-Mattila J, Tuominen J, Huhtala H, Palomäki A, Koivistoinen T. Emergency department crowding increases 10-day mortality for non-critical patients: a retrospective observational study. Internal and Emergency Medicine. 2024; 19: 175–181.

[4] Park S, Chang H, Jung W, Lee SU, Hwang SY, Yoon H, et al. Impact of the 24-hour time target policy for emergency departments in South Korea: a mixed method study in a single medical center. BMC Health Services Research. 2022; 22: 1510.

[5] Bisarya R, Song X, Salle J, Liu M, Patel A, Simpson SQ. Antibiotic timing and progression to septic shock among patients in the ED with suspected infection. Chest. 2022; 161: 112–120.

[6] Leung LY, Huang HL, Hung KK, Leung CY, Lam CC, Lo RS, et al. Door-to-antibiotic time and mortality in patients with sepsis: systematic review and meta-analysis. European Journal of Internal Medicine. 2024; 129: 48–61.

[7] Tang F, Yuan H, Li X, Qiao L. Effect of delayed antibiotic use on mortality outcomes in patients with sepsis or septic shock: a systematic review and meta-analysis. International Immunopharmacology. 2024; 129: 111616.

[8] Seymour CW, Gesten F, Prescott HC, Friedrich ME, Iwashyna TJ, Phillips GS, et al. Time to treatment and mortality during mandated emergency care for sepsis. The New England Journal of Medicine. 2017; 376: 2235–2244.

[9] Darraj A, Hudays A, Hazazi A, Hobani A, Alghamdi A. The association between emergency department overcrowding and delay in treatment: a systematic review. Healthcare. 2023; 11: 385.

[10] Clopton EL, Hyrkäs EK. Modeling emergency department nursing workload in real time: an exploratory study. International Emergency Nursing. 2020; 48: 100793.

[11] Wretborn J, Khoshnood A, Wieloch M, Ekelund UE. Skåne emergency department assessment of patient load (SEAL)—a model to estimate crowding based on workload in swedish emergency departments. PLOS ONE. 2015; 10: e0130020.

[12] Bernstein SL, Verghese V, Leung W, Lunney AT, Perez I. Development and validation of a new index to measure emergency department crowding. Academic Emergency Medicine. 2003; 10: 938–942.

[13] Weiss SJ, Derlet R, Arndahl J, Ernst AA, Richards J, Fernández-Frackelton M, et al. Estimating the degree of emergency department overcrowding in academic medical centers: results of the national ED overcrowding study (NEDOCS). Academic Emergency Medicine. 2004; 11: 38–50.

[14] Ilhan B, Kunt MM, Damarsoy FF, Demir MC, Aksu NM. NEDOCS: Is it really useful for detecting emergency department overcrowding today? Medicine. 2020; 99: E20478.

[15] Song Y, Zhang X, Luo D, Shi J, Zang Q, Wang Y, et al. Predicting nursing workload in digestive wards based on machine learning: a prospective study. BMC Nursing. 2024; 23: 1–10.

[16] Ivziku D, Ferramosca FM, Filomeno L, Gualandi R, De Maria M, Tartaglini D. Defining nursing workload predictors: a pilot study. Journal of Nursing Management. 2022; 30: 473–481.

[17] Benjamin E. Innovations in emergency nursing: adapting patient flow management to emergency department overcrowding. Journal of Emergency Nursing. 2025; 51: 261–268.

[18] Park S, Yoo J, Lee Y, DeGuzman PB, Kang MJ, Dykes PC, et al. Quantifying emergency department nursing workload at the task level using NASA-TLX: an exploratory descriptive study. International Emergency Nursing. 2024; 74: 101424.

[19] Signorini F, Nattino G, Rossi C, Ageno W, Catania F, Cortellaro F, et al. Measuring the crowding of emergency departments: an assessment of the NEDOCS in Lombardy, Italy, and the development of a new objective indicator based on the waiting time for the first clinical assessment. BMC Emergency Medicine. 2024; 24: 196.

[20] Altun M, Kudu E, Demir O, Karacabey S, Sanri E, Onur OE, et al. Effect of access block on emergency department crowding calculated by NEDOCS score. American Journal of Emergency Medicine. 2024; 82: 136–141.

[21] Getachew M, Musa I, Degefu N, Beza L, Hawlte B, Asefa F. Emergency department overcrowding and its associated factors at HARME medical emergency center in eastern Ethiopia. African Journal of Emergency Medicine. 2024; 14: 26–32.

[22] Fabbri C, Lombardi M, Malaguti E, Monaci M. On-line strategy selection for reducing overcrowding in an emergency department. Omega. 2024; 127: 103098.

[23] Matthew H, Melinda H. Understanding poisson regression. Journal of Nursing Education. 2014; 53: 207–215.

[24] Gao J, Lu Y, Ashrafi N, Domingo I, Alaei K, Pishgar M. Prediction of sepsis mortality in ICU patients using machine learning methods. BMC Medical Informatics and Decision Making. 2024; 24: 228.

[25] Oh N, Cha WC, Seo JH, Choi SG, Kim JM, Chung CR, et al. ChatGPT predicts in-hospital all-cause mortality for sepsis: in-context learning with the korean sepsis alliance database. Healthcare Informatics Research. 2024; 30: 266–276.

[26] Li X, Wu R, Zhao W, Shi R, Zhu Y, Wang Z, et al. Machine learning algorithm to predict mortality in critically ill patients with sepsis-associated acute kidney injury. Scientific Reports. 2023; 13: 5223.

[27] Breiman L. Random forests. Machine Learning. 2001; 45: 5–32.

[28] Chen T, Guestrin C. XGBoost: a scalable tree boosting system. 2016. Available at: http://dx.doi.org/10.1145/2939672.2939785 (Accessed: 18 June 2025).

[29] Lundberg SM, Lee SI. A unified approach to interpreting model predictions. 2017. Available at: https://proceedings.neurips.cc/paper_files/paper/2017/file/8a20a8621978632d76c43dfd28b67767-Paper.pdf (Accessed: 18 June 2025).

[30] Van Rossum G, Drake FL. The Python Language Reference. 2021. Available at: https://docs.python.org/3.10/reference/ (Accessed: 18 June 2025).

[31] Choi H, Ok JS, An SY. Evaluation of validity of the Korean triage and acuity scale. Journal of Korean Academy of Nursing. 2019; 49: 26–35.

[32] Chung H, Kwak DS, Park J. Comparison of validity of Korean triage and acuity scale levels assessed by triage nurse in real-time and by retrospectively applying strict systemic inflammatory response syndrome criteria in patients with fever. Signa Vitae. 2024; 20: 39–45.

[33] Chang H, Yu JY, Lee GH, Heo S, Lee SU, Hwang SY, et al. Clinical support system for triage based on federated learning for the Korea triage and acuity scale. Heliyon. 2023; 9: e19210.

[34] Bomrah S, Uddin M, Upadhyay U, Komorowski M, Priya J, Dhar E, et al. A scoping review of machine learning for sepsis prediction- feature engineering strategies and model performance: a step towards explainability. Critical Care. 2024; 28: 180.

[35] Lee JH, Cho SH. Effect of crowding and nurse staffing on time to antibiotic administration for patients with pneumonia in an emergency department. Journal of Korean Academy of Nursing Administration. 2018; 24: 107–117.

[36] Phend C. Emergency department volume numbers don’t tell the whole story: the rising toll of boarding and high-acuity care. Annals of Emergency Medicine. 2023; 82: A15–A18.

[37] Fee C, Weber EJ, Bacchetti P, Maak CA. Effect of emergency department crowding on pneumonia admission care components. The American Journal of Managed Care. 2011; 17: 269–278.

[38] Roberts RJ, Alhammad AM, Crossley L, Anketell E, Wood L, Schumaker G, et al. A survey of critical care nurses’ practices and perceptions surrounding early intravenous antibiotic initiation during septic shock. Intensive and Critical Care Nursing. 2017; 41: 90–97.

[39] Mittal Y, Sankar J, Dhochak N, Gupta S, Lodha R, Kabra SK. Decreasing the Time to administration of first dose of antibiotics in children with severe sepsis. Journal for Healthcare Quality. 2019; 41: 32–38.

[40] Mitzkewich M. Sepsis screening in triage to decrease door-to-antibiotic time. Journal of Emergency Nursing. 2019; 45: 254–256.

[41] Bray C, Kennedy C. Improving timely sepsis care using the surviving sepsis campaign one-hour bundle in a rural emergency department. Journal of the American Association of Nurse Practitioners. 2020; 33: 246–253.

[42] Roman CP, Poole SG, Dooley MJ, Smit DV, Mitra B. Implementation of hospital‐wide reform at improving access and flow: impact on time to antibiotics in the emergency department. Emergency Medicine Australasia. 2016; 28: 133–137.