Duration of Patients' Visits to the Hospital Emergency Department

Zeynal Karaca; Herbert S Wong; Ryan L Mutter

Disclosures

BMC Emerg Med. 2012;12(15) 

In This Article

Methods

Study Design and Population

We conducted a retrospective data analysis to investigate the duration of ED visits using the Healthcare Cost and Utilization Project (HCUP)b State Emergency Department Databases (SEDD) for 2008. HCUP is maintained by the Agency for Healthcare Research and Quality (AHRQ). HCUP databases are publicly available for all researchers and can be purchased through the HCUP Central Distributor.c The SEDD employed in this study include data on 4.9 million T&R ED visits in three states: Arizona, Massachusetts, and Utah. In general, the SEDD provide detailed diagnoses and procedures, total charges, and patient demographics. Demographics include gender, age, race, and expected payment source (e.g., Medicare, Medicaid, private insurance, other insurance, and self-pay). However, the SEDD from these three states also provide admission and discharge time for 99.1 percent of all visits, from which durationd may be calculated.

We obtained information about hospital characteristics (e.g., urban versus rural, ownership type, teaching status, bed size, and system member) from the 2008 American Hospital Association (AHA) Annual Survey Database and linked them to SEDD files using hospital identifiers. In addition, we obtained information about the trauma level of the hospital using the Trauma Information Exchange Program database (TIEP), collected by the American Trauma Society and the Johns Hopkins Center for Injury Research and Policy. Finally, we used the 2008 Area Resource File (ARF)e to obtain county-level income information.

The proper measures of ED LOS and ED crowding are not straightforward.[15] Few investigators have attempted to develop models characterizing the completion times of different phases of emergency care. Multivariate linear regression techniques used to estimate ED waiting room time, treatment time, and boarding time for patients who were admitted or discharged from a hospital's main ED or urgent care area.[7] Similarly, discrete-time survival analysis is applied to evaluate the effect of crowding on the different phases of ED care.[4] Both studies estimated the influence of various patient, temporal, and system factors on the mean or median completion times for different phases of emergency care. Few researchers[16] contributed to this literature by demonstrating that the degree of crowding in a hospital can be accurately measured.

Because the proper measures of ED LOS were not readily available in our data, we computed the duration for each visit by taking the difference between admission and discharge times, which is the total of the time patients waited in ED rooms plus their treatment time. Ideally, one would separate the times into components identified as important in the literature. Unfortunately, HCUP data lacks sufficient detail to do this.

Statistical Analyses

We initially performed extensive secondary data analyses to explore ED LOS by admission hour, day of the week, patient volume, patient characteristics, hospital characteristics and area characteristics. The frequencies, means, medians, and 95% confidence intervals for several of these variables were based on data for all T&R ED visits (excluding encounters where there was evidence that the patient also received observation services) in Arizona, Massachusetts, and Utah during 2008. Duration was expressed in minutes measured as the difference between admission time and discharge time.

The mean (median) duration for a specific admission hour was measured as the mean (median) value of the durations of all visits admitted to EDs at that specific hour during 2008. The total volume of visits for a specific admission hour was measured as the total number of T&R visits to the EDs observed at that specific hour during 2008. (Note that it was not possible to include ED visits that resulted in subsequent admission to the hospital in the analysis.) We applied a similar approach when reporting the mean duration of ED visits across patient demographics and hospital characteristics. For example, the mean duration of ED visits for Medicare patients was measured as the total duration of T&R ED visits by all Medicare patients divided by the total number of T&R visits by Medicare patients during 2008. Data were analyzed with SAS 9.02 and Stata 12.

Severity of illness is an important factor that can affect the mean duration of ED visits. To further explore the potential relationship between the mean duration of visits and various disease groups, we grouped ED visits into major disease categories based on Clinical Classification Software—a diagnosis and procedure categorization scheme based on the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM). While the HCUP SEDD provide all diagnosis codes for every visit, they may not clearly differentiate between the primary diagnosis codes and other diagnosis codes. Therefore, we used all diagnosis codes reported for each visit when developing our major disease categories.

While this study is mostly observational, we also investigated the factors affecting the duration of T&R ED visits using several multivariable regression models. We attempted to explain the variability in the duration of T&R ED visits using admission day of the week, admission hour of the day, and patient and hospital characteristics. More specifically, we estimated several regression models to examine factors associated with the duration of patients' T&R ED visits. We initially estimated a linear regression model that controls for 1) admission day of the week; 2) patient characteristics including age, sex, race, primary payers, and major disease categories; and 3) hospital characteristics including hospital teaching status, hospital ownership status, trauma hospitals, hospital location, and hospital bed size. Next, we estimated the same model by further controlling for patients' admission hour of the day. Then, we developed a third model based on the second model by incorporating hospital-specific dummy variables to increase the robustness of our results.

Several previous studies[17–20] showed that linear regression models that contain a response variable at the individual level and predictors at both individual and higher levels of analysis disregard correlation structures in the data emanating from common influences operating within groups. For example, hospital attributes such as teaching status, bed size, or location may impose distinct effects on the duration of patients' visits to the EDs. Such "intra-class correlation" violates classical linear regression assumptions concerning random error, independence, and common variance. Mainly, in nested data containing two levels, the random error is composed of both individual and group-level components, and thus, independence of observations fails as a portion of the random error is attributable to the group error which is constant within each group. Ignoring this correlation may lead to underestimates of standard errors of coefficients and therefore overestimates of the significance levels of parameters in linear regression models. By nesting patients within hospitals, we estimated our fourth model, which is a random intercept two level model with level-1 predictors. This model allows intercepts to vary, and hence, duration of ED visits for each patient are predicted by the intercept that varies across hospitals. This model also provides information about intra-class correlations, which enable us to determine what fraction of variance in duration of patients' visits to the EDs are due to patient characteristics and which are due to hospital characteristics. Following the approach in the previous studies,[17,18] we used hospital means to centralize all variables pertinent to patient demographics. We also aimed to partition the variation in duration of patients' visits to the EDs between patient and hospital level, which in turn provides us an intra-class correlation.

bFurther details about HCUP databases are available at http://www.hcup-us.ahrq.gov/.
cFurther details are available at http://www.hcup-us.ahrq.gov/tech_assist/centdist.jsp.
dAs part of the HCUP Project, AHRQ negotiates with data organizations that maintain statewide data systems to acquire hospital-based data, process those data into research databases, and subsequently release a subset of those data to the public with a signed data use agreement. Some data elements are considered too sensitive by these data organizations for general release to the public. However, under the terms of their agreements with AHRQ, some AHRQ staff may use these more sensitive data for analysis. In this study, the Arizona Department of Health Services, the Massachusetts Division of Health Care Finance and Policy, and the Utah Department of Health granted permission for the data elements, admission hour and discharge hour, to be used internally by AHRQ.
eARF provide county level data. Further details are available at http://arf.hrsa.gov/.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....