Study design
This retrospective cohort study extracted data from population-based databases of patients residing in Alberta, Canada. No funding organization had any role in the conduct and reporting of this study.
Study setting and population
All presentations acute asthma at the 18 highest-volume EDs during April 1, 2014, to March 31, 2019 were extracted for patients aged 18 to 55 years. The highest-volume EDs in Alberta were the focus for this project and were categorized into three groups: regional, urban, and academic/teaching.
Study protocol
There were 6 administrative patient databases used for the study. The National Ambulatory Care Reporting System (NACRS) database collects information on ED presentations. Linkages to other large health administrative databases were made to provide other information for analysis. The Annual Cumulative Registry File (CRF) provided demographic and geographic information; the Discharge Abstract Database (DAD) provided inpatient hospitalization data; the Physician Claim File (PCF) provided physician follow-up visits; and the Alberta Vital Statistics (VS) provided death records. Community size and neighbourhood income level were provided by Statistics Canada based on postal code.
The NACRS database provided the dates and times of ED presentations, triage level, physician initial assessment (PIA) diagnostic and intervention data, and disposition status. We defined the start of the ED presentation as the minimum date-time between the patient registration and triage. Fiscal year, month and year, weekday/weekend, and shift time were defined by the ED presentation date-time. The International Classification of Diseases (ICD-10-CA) [13] was used to define the asthma cases and the study population. Since asthma diagnoses for ages greater than 55 may be inaccurate and ED research has found that patients above 55 years of age frequently have COPD [14], this project only includes patients diagnosed with asthma between 18 and 55 years of age. Triage was determined based on the Canadian Triage and Acuity Scale (CTAS) used in most Canadian EDs, which is used to indicate the level of acuity and urgency for each ED presentation [15, 16]. There are five levels of triage with associated PIA targets: CTAS 1 (Resuscitation): immediate (= 0 min); CTAS 2 (Emergent): ≤15 min; CTAS 3 (Urgent): ≤ 30 min; CTAS 4 (Semi-urgent/less urgent): ≤ 60 min; CTAS 4 (Non-urgent): ≤ 120 min. Ten disposition codes represent the patient’s status at the end of each ED presentation, and were grouped into six disposition categories: discharged (i.e., discharged home, discharged from program or clinic), admitted to hospital (i.e., admitted to critical care or operating room, admitted to regular ward), transferred (i.e., to another acute care facility, to another non-acute care facility), death (i.e., on arrival or in the ED), and left without completion of care (left without being seen [LWBS] or left against medical advice [LAMA]).
The CRF database provided demographic and geographic information on individual patients. Age was defined as the age at the date of the ED presentation and calculated based on the patients’ birthdate. Patient sex was coded as male or female. Using the postal code of the patients’ residence at fiscal year-end, patients were geo-coded into one of the province’s five health zones (North, Edmonton, Central, Calgary, South). Community size and neighbourhood income quintile were identified by linking postal codes and the 2006 census data, the Postal Code Conversion File Plus [17].
The PCF database provided data on follow-up visits, including the physician’s specialty, if any. Physician specialists of interest were coded as respiratory medicine specialists or other specialists. DAD database provided the start and end date-time of each patient’s hospitalization. The ICD diagnostic codes were used to determine the Charlson Comorbidity Score [18] and comorbidity indicator variables.
Crowding metrics
Length of stay (LOS) and PIA time were calculated as ED crowding metrics [1] using data extracted from NACRS for patients presenting at any age for any condition (i.e., not restricted to adults or patients with asthma). For each ED presentation not ending in “left without completion of care”, the PIA time was defined as the difference between the presentation start time and the first physician assessment time. The first interaction for patients in Canadian EDs is with a nurse who performs triage. In the event that a patient is severe (CTAS 1 and some CTAS 2 cases), registration occurs after the PIA. Because, the PIA may occur before the reported ED start time in critical situations, negative values of PIA time were set as zero [19]. Each metric was calculated for each of the high volume ED facilities each hour within the same date during the study period. Hourly facility-specific ED crowding metrics (mean or median for skewed data) were calculated using all ED presentations within the same hour (e.g., 08:00–08:59) at the same facility. For each ED presentation, LOS was defined as the time from the start of ED presentation to the time of disposition; LOS for hospitalized patients and discharged patients were also calculated. Because the crowding metrics are continuous measures, they provide a measure of crowding on a continuum rather than classifying EDs into crowded or not crowded categories.
Key outcome measures
Outcome variables were derived or calculated based on the original variables of the data sources and were broadly classified as duration time variables and categorical outcome variables. Duration time variables measure delays that potentially affect patient outcomes and categorical outcome variables measure patient severity and status at the end of ED presentation. Some outcomes were only applicable to subsets of the ED presentation data (e.g., outcomes measure for admitted or discharged patients only). Emergency department LOS was defined as the difference between the presentation start time and end time. For admitted patients, the end time was defined as the time of departure from the ED to an in-patient unit. For discharged patients, the end time was defined as the time of disposition decision [19]. The length of hospital admission was calculated as the time from the start of the hospitalization to the time of discharge.
For discharged patients, the time to next follow-up visit with a physician was defined as the number of days between the ED presentation and the next physician visit. The time-to-event variable was censored if: (1) the patient died within 183 days (i.e., 6 months) after the ED presentation and there was no follow-up physician visit after the visit (data censored on the date of death); (2) follow-up physician visit did not occur within 183 days. The time to next follow-up visit with a respiratory medicine specialist, and time to next follow-up visit with other specialists, were defined similarly for discharged patients. The time to next ED presentation was censored at March 31, 2019 or death date during the study period. Corresponding censoring indicators were developed for all time-to-event outcomes.
Data analysis
Numerical summaries (e.g., means and standard deviations [SDs]; medians and interquartile ranges [IQR]) and frequency distributions (percentages) described ED presentations. For medians, IQRs represent the 25th percentile and 75th percentile. Three CTAS groups were collapsed from the triage levels: high acuity for CTAS 1/ 2, moderate acuity for CTAS 3, and low acuity for CTAS 4/5. Linear and Cox proportional hazards (PH) models were applied for duration time variables, including time-to-event variables with censoring indicators, and generalized linear models used for categorical outcome variables. Models were applied to assess the association between the ED crowding metric and the outcome variables by CTAS groups. Random effects accommodated the multiple correlated data from the same patient and variation in each ED facility.
Each crowding metric variable was included in both the model for unadjusted estimates and the full models for adjusted estimates to assess its effects on outcomes. Other predictors included in the full models were: age, sex, zone, weekday/weekend, month of year, shift, fiscal year, ED category, income level (lowest and second lowest neighbourhood income quintile vs. others), community size (population ≥ 100,000 vs. others), and selected comorbidities (i.e., mild/moderate/severe liver disease, diabetes mellitus, cancer or metastatic solid tumor, myocardial infarction, congestive heart failure, renal disease, peripheral vascular disease, stroke).
In addition, estimates, odds ratios (ORs), hazard ratios (HRs), and associated 95% confidence intervals (CIs) were provided. A p-value (p) less than 0.05 was considered to be statistically significant. Statistical analyses were conducted in R [20] and SAS [21].