Predicting emergency department length of stay using quantile regression

Ru Ding, Melissa L. McCarthy, Jennifer Lee Levy, Jeffrey S. Desmond, Scott Zeger, Dominik Aronsky

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Objectives: Length of stay (LOS) is an important emergency department (ED) performance measure. The objective of this study was to predict the 10th, 50th and 90th percentile of ED LOS using demographic, clinical and temporal characteristics in order to better inform patients and ED staff. Methods: A retrospective cohort study was conducted on one year ED visit data from an academic ED (N=50,824). We estimated the 10th, 50th and 90th percentile for three different phases of ED care: waiting time, treatment time and boarding time. We used multivariate quantile regression to model the three phases of ED care separately as a function of patients' arrival day and time, age, gender, mode of arrival, insurance status, acuity level and chief complaint. Results: The median waiting time was 14 minutes, 191 minutes for median treatment time and 154 minutes for median boarding time. Patients at the 90% waited 7 times longer (98 minutes), took 2.5 times longer to be treated (487 minutes) and boarded 7 times longer (1,122 minutes) compared to patients at the median. Patients' chief complaint and acuity level were the most important predictors of the three phases of LOS. The adjusted median treatment times for patients with a cardiovascular symptom (202 to 328 minutes depending on acuity level) were longer than patients with any other complaint, regardless of acuity. Of all chief complaints, the longest median boarding times were experienced by patients with a skin problem (177 to 291 minutes depending on acuity level). Day and time of arrival were important predictors of wait time and boarding time as well. The adjusted boarding times were 30% to 100% longer on Mondays compared to Sundays. Conclusions: ED LOS varied significantly among patients in a predictable manner that is largely explained by information available at triage. Providing patients with an expected LOS at triage may result in increased patient satisfaction.

Original languageEnglish (US)
Title of host publicationProceedings - International Conference on Management and Service Science, MASS 2009
DOIs
StatePublished - 2009
EventInternational Conference on Management and Service Science, MASS 2009 - Wuhan, China
Duration: Sep 20 2009Sep 22 2009

Other

OtherInternational Conference on Management and Service Science, MASS 2009
CountryChina
CityWuhan
Period9/20/099/22/09

Fingerprint

Emergency department
Length of stay
Quantile regression
Median
Complaints
Waiting time
Predictors
Performance measures
Patient satisfaction
Insurance
Cohort study
Staff
Demographics

Keywords

  • Emergency department
  • Length of stay
  • Prediction
  • Quantile regression

ASJC Scopus subject areas

  • Information Systems and Management
  • Management Science and Operations Research

Cite this

Ding, R., McCarthy, M. L., Levy, J. L., Desmond, J. S., Zeger, S., & Aronsky, D. (2009). Predicting emergency department length of stay using quantile regression. In Proceedings - International Conference on Management and Service Science, MASS 2009 [5300861] https://doi.org/10.1109/ICMSS.2009.5300861

Predicting emergency department length of stay using quantile regression. / Ding, Ru; McCarthy, Melissa L.; Levy, Jennifer Lee; Desmond, Jeffrey S.; Zeger, Scott; Aronsky, Dominik.

Proceedings - International Conference on Management and Service Science, MASS 2009. 2009. 5300861.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Ding, R, McCarthy, ML, Levy, JL, Desmond, JS, Zeger, S & Aronsky, D 2009, Predicting emergency department length of stay using quantile regression. in Proceedings - International Conference on Management and Service Science, MASS 2009., 5300861, International Conference on Management and Service Science, MASS 2009, Wuhan, China, 9/20/09. https://doi.org/10.1109/ICMSS.2009.5300861
Ding R, McCarthy ML, Levy JL, Desmond JS, Zeger S, Aronsky D. Predicting emergency department length of stay using quantile regression. In Proceedings - International Conference on Management and Service Science, MASS 2009. 2009. 5300861 https://doi.org/10.1109/ICMSS.2009.5300861
Ding, Ru ; McCarthy, Melissa L. ; Levy, Jennifer Lee ; Desmond, Jeffrey S. ; Zeger, Scott ; Aronsky, Dominik. / Predicting emergency department length of stay using quantile regression. Proceedings - International Conference on Management and Service Science, MASS 2009. 2009.
@inproceedings{c6a77d99fa1343ab92d6677a55610c4a,
title = "Predicting emergency department length of stay using quantile regression",
abstract = "Objectives: Length of stay (LOS) is an important emergency department (ED) performance measure. The objective of this study was to predict the 10th, 50th and 90th percentile of ED LOS using demographic, clinical and temporal characteristics in order to better inform patients and ED staff. Methods: A retrospective cohort study was conducted on one year ED visit data from an academic ED (N=50,824). We estimated the 10th, 50th and 90th percentile for three different phases of ED care: waiting time, treatment time and boarding time. We used multivariate quantile regression to model the three phases of ED care separately as a function of patients' arrival day and time, age, gender, mode of arrival, insurance status, acuity level and chief complaint. Results: The median waiting time was 14 minutes, 191 minutes for median treatment time and 154 minutes for median boarding time. Patients at the 90{\%} waited 7 times longer (98 minutes), took 2.5 times longer to be treated (487 minutes) and boarded 7 times longer (1,122 minutes) compared to patients at the median. Patients' chief complaint and acuity level were the most important predictors of the three phases of LOS. The adjusted median treatment times for patients with a cardiovascular symptom (202 to 328 minutes depending on acuity level) were longer than patients with any other complaint, regardless of acuity. Of all chief complaints, the longest median boarding times were experienced by patients with a skin problem (177 to 291 minutes depending on acuity level). Day and time of arrival were important predictors of wait time and boarding time as well. The adjusted boarding times were 30{\%} to 100{\%} longer on Mondays compared to Sundays. Conclusions: ED LOS varied significantly among patients in a predictable manner that is largely explained by information available at triage. Providing patients with an expected LOS at triage may result in increased patient satisfaction.",
keywords = "Emergency department, Length of stay, Prediction, Quantile regression",
author = "Ru Ding and McCarthy, {Melissa L.} and Levy, {Jennifer Lee} and Desmond, {Jeffrey S.} and Scott Zeger and Dominik Aronsky",
year = "2009",
doi = "10.1109/ICMSS.2009.5300861",
language = "English (US)",
isbn = "9781424446391",
booktitle = "Proceedings - International Conference on Management and Service Science, MASS 2009",

}

TY - GEN

T1 - Predicting emergency department length of stay using quantile regression

AU - Ding, Ru

AU - McCarthy, Melissa L.

AU - Levy, Jennifer Lee

AU - Desmond, Jeffrey S.

AU - Zeger, Scott

AU - Aronsky, Dominik

PY - 2009

Y1 - 2009

N2 - Objectives: Length of stay (LOS) is an important emergency department (ED) performance measure. The objective of this study was to predict the 10th, 50th and 90th percentile of ED LOS using demographic, clinical and temporal characteristics in order to better inform patients and ED staff. Methods: A retrospective cohort study was conducted on one year ED visit data from an academic ED (N=50,824). We estimated the 10th, 50th and 90th percentile for three different phases of ED care: waiting time, treatment time and boarding time. We used multivariate quantile regression to model the three phases of ED care separately as a function of patients' arrival day and time, age, gender, mode of arrival, insurance status, acuity level and chief complaint. Results: The median waiting time was 14 minutes, 191 minutes for median treatment time and 154 minutes for median boarding time. Patients at the 90% waited 7 times longer (98 minutes), took 2.5 times longer to be treated (487 minutes) and boarded 7 times longer (1,122 minutes) compared to patients at the median. Patients' chief complaint and acuity level were the most important predictors of the three phases of LOS. The adjusted median treatment times for patients with a cardiovascular symptom (202 to 328 minutes depending on acuity level) were longer than patients with any other complaint, regardless of acuity. Of all chief complaints, the longest median boarding times were experienced by patients with a skin problem (177 to 291 minutes depending on acuity level). Day and time of arrival were important predictors of wait time and boarding time as well. The adjusted boarding times were 30% to 100% longer on Mondays compared to Sundays. Conclusions: ED LOS varied significantly among patients in a predictable manner that is largely explained by information available at triage. Providing patients with an expected LOS at triage may result in increased patient satisfaction.

AB - Objectives: Length of stay (LOS) is an important emergency department (ED) performance measure. The objective of this study was to predict the 10th, 50th and 90th percentile of ED LOS using demographic, clinical and temporal characteristics in order to better inform patients and ED staff. Methods: A retrospective cohort study was conducted on one year ED visit data from an academic ED (N=50,824). We estimated the 10th, 50th and 90th percentile for three different phases of ED care: waiting time, treatment time and boarding time. We used multivariate quantile regression to model the three phases of ED care separately as a function of patients' arrival day and time, age, gender, mode of arrival, insurance status, acuity level and chief complaint. Results: The median waiting time was 14 minutes, 191 minutes for median treatment time and 154 minutes for median boarding time. Patients at the 90% waited 7 times longer (98 minutes), took 2.5 times longer to be treated (487 minutes) and boarded 7 times longer (1,122 minutes) compared to patients at the median. Patients' chief complaint and acuity level were the most important predictors of the three phases of LOS. The adjusted median treatment times for patients with a cardiovascular symptom (202 to 328 minutes depending on acuity level) were longer than patients with any other complaint, regardless of acuity. Of all chief complaints, the longest median boarding times were experienced by patients with a skin problem (177 to 291 minutes depending on acuity level). Day and time of arrival were important predictors of wait time and boarding time as well. The adjusted boarding times were 30% to 100% longer on Mondays compared to Sundays. Conclusions: ED LOS varied significantly among patients in a predictable manner that is largely explained by information available at triage. Providing patients with an expected LOS at triage may result in increased patient satisfaction.

KW - Emergency department

KW - Length of stay

KW - Prediction

KW - Quantile regression

UR - http://www.scopus.com/inward/record.url?scp=73849084843&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=73849084843&partnerID=8YFLogxK

U2 - 10.1109/ICMSS.2009.5300861

DO - 10.1109/ICMSS.2009.5300861

M3 - Conference contribution

AN - SCOPUS:73849084843

SN - 9781424446391

BT - Proceedings - International Conference on Management and Service Science, MASS 2009

ER -