Chronic obstructive pulmonary disease (COPD) is a common and disabling condition usually caused by cigarette smoke exposure and characterized by progressive symptoms of dyspnea. Acute exacerbations of COPD (AECOPD) are characterized by an increase in the ongoing symptoms of cough, sputum production, sputum purulence, and/or shortness of breath which usually require changes in baseline medical management [
Patients with COPD often experience exacerbations and some exacerbations are severe enough to require emergency department (ED) management. Treatment guidelines for AECOPD exist and prolonged treatment may occur in the ED, some exacerbations result in complications including pneumothorax, pneumonia, need for noninvasive ventilation, intubation and mechanical ventilation, and death. Due to the advanced age of patients, associated severe comorbidities, and severity of many presentations, hospital admission is a common outcome in severe exacerbations (33% in COPD [
Based on data health professional-diagnosed COPD from the 1994-95 National Health Survey, prevalence rates are estimated to be 4.7%, 5.4%, and 8.3% in Canadians aged 55–64, 65–74, and ≥75, respectively [
ED presentations for COPD are important events for patients and the health care system. This study describes the temporal trends in the presentations of individuals (age ≥55 years) to EDs for COPD during a 12-year period (April 1, 1999, to March 31, 2011) and provides time series models to quantify the temporal patterns.
This project is a secondary analysis of existing population-based, administrative databases.
Alberta is a Western Canadian province, which operates under a single health authority, providing government funded health care services to more than 4.1 million residents. All visits to Alberta’s more than 100 EDs are tracked in the provincial Ambulatory Care Classification System [
All COPD ED visits for patients 55 years or older between April 1, 1999, and March 31, 2011 (fiscal years 2000 to 2011), were extracted from ACCS, where COPD was defined by either the first or second diagnosis field containing any of the diagnostic codes 490.x, 491.x, 492.x, 494.x, and 496.x (ICD-9-CM) or J40.x, J41.x, J42.x, J43.x, J44.x, and J47.x (ICD-10-CA). The date and time of the ED visit were extracted as well as the mode of release from the ED (i.e., admitted to hospital or discharged).
Two age groups were formed with patients classified as either 55–64 or 65+ based on age at time of ED visit. Alberta is comprised of five geographical and administrative health zones (North, Edmonton, Central, Calgary, and South; Figure
Map of Alberta depicting administrative health zones.
For each month, COPD ED visit rates per 100,000 individuals were calculated by age group, sex, zone, and overall. Additionally, rates for age group within each sex were calculated. Since population counts were only available at the end of each fiscal year, linear interpolation and extrapolation were used to obtain monthly population estimates and the months were normalized to 30 days so that comparisons were made across equal time periods. Patient demographics were summarized using frequencies and percentages, and the overall rates during the period were summarized by the average monthly COPD ED visits per 100,000 individuals. Three ED visits with missing zone were excluded from the zone specific analyses.
In order to summarize and differentiate between seasonal and trend components of the time series, seasonal-trend decomposition based on LOESS (STL) [
To estimate parametric models for monthly rates of COPD ED visits over time, seasonal autoregressive integrated moving average (SARIMA) models were considered. SARIMA models can be denoted by ARIMA
SAS version 9.4 for Linux (SAS Institute Inc., Cary, NC) and R version 3.2.0 for Windows (including the “stats” and “forecast” packages) were used to carry out all statistical analyses.
The study was approved by the University of Alberta’s Health Research Ethics Board and patients were not contacted during this study. The data are reported in aggregate, and small cell sizes are suppressed to protect anonymity.
There were a total of 188,824 ED visits for COPD between April, 1999, and March, 2011 (197.7 and 232.6 visits per 100,000 individuals, resp.), of which most (76.5%) listed COPD as the primary diagnosis. Average monthly rates of COPD ED visits per 100,000 individuals were higher for males than females (218.6 versus 180.2), higher for ages 65 and over compared to ages 55–64 (278.1 versus 109.5), and varied by geographic zones, ranging from a low of 108.6 in the Calgary zone to a high of 463.9 in the North zone (Table
Patient demographics and time series characteristics.
COPD ED visits, | Mean monthly COPD ED visits/100,000 | Change in trend component, % | Seasonal high month | Seasonal low month | |
---|---|---|---|---|---|
Overall | 188,824 | 198.4 | −0.4 | March | August |
Sex | |||||
Female | 90317 (47.8) | 180.2 | 10.8 | March | August |
Age 55–64 | 25483 (28.2) | 113.2 | 8.6 | March | August |
Age 65+ | 64834 (71.8) | 234.5 | 20.2 | March | August |
Male | 98507 (52.2) | 218.6 | −9.7 | March | August |
Age 55–64 | 24132 (24.5) | 105.8 | −1.8 | January | August |
Age 65+ | 74375 (75.5) | 331.7 | −3.5 | March | August |
Zone | |||||
North | 45411 (24.0) | 463.9 | 40.6 | March | August |
Edmonton | 48314 (25.6) | 155.6 | 4.1 | December | August |
Central | 44111 (23.4) | 328.7 | −17.6 | March | August |
Calgary | 35160 (18.6) | 108.6 | −6.5 | March | August |
South | 15825 (8.4) | 184.3 | −6.9 | March | August |
Each time series is shown in Figure
Monthly COPD ED visits per 100,000 individuals, April, 1999, to March, 2011, with overlay of trend component.
Overall
Female, age 55–64
Female, age 65+
Male, age 55–64
Male, age 65+
North
Edmonton
Central
Calgary
South
Comparison of seasonal components by month.
Despite similar seasonality across each series, there existed different patterns of long term trend. The trend component for the overall rate of COPD ED visits was relatively flat, with a 0.4% decrease from April, 1999, to March, 2011. The differences in sex were small but opposite in direction (10.8% increase for females; 9.7% decrease for males). For females, the age 65+ group increased by 20.2% compared to only 8.6% for age 55–64, while for males both age groups saw small decreases (3.5% for age 65+ and 1.8% for age 55–64). Three of the zones saw only small changes over the time period, including the urban zones of Edmonton (4.1% increase) and Calgary (6.5% decrease), as well as the South (6.9% decrease). The Central zone, however, had a decrease of 17.6% and the North zone had an increase of 40.6% over the time period (Table
To parametrically model the ED visit rates for COPD over time, SARIMA models were fit for each series. The OCSB test suggested seasonal differencing was not required for any of the series, and the KPSS test suggested nonseasonal differencing was required only for the North, Central, and South zones to achieve stationarity. For the overall series, the model that resulted in the smallest AIC was ARIMA
Overall optimal model and parameter estimates for COPD ED visit rates.
Parameter | Overall |
---|---|
Optimal model | ( |
Intercept | 197.31 (10.31) |
AR(1) | 0.44 (0.08) |
MA(1) | — |
MA(2) | — |
Seasonal AR(1) | 0.89 (0.07) |
Seasonal AR(2) | |
Seasonal MA(1) | −0.52 (0.16) |
Seasonal MA(2) | — |
AIC | 1337.9 |
Parameter estimates are displayed with standard error in parentheses.
Subseries optimal models and parameter estimates for COPD ED visit rates.
Parameter | Female, age 55–64 | Female, age 65+ | Male, age 55–64 | Male, age 65+ | North | Edmonton | Central | Calgary | South |
---|---|---|---|---|---|---|---|---|---|
Optimal model | | | | | | | | | |
Intercept | 113.49 (4.48) | — | 106.99 (5.97) | 328.80 (16.10) | — | 152.99 (6.16) | — | 108.63 (5.65) | — |
AR(1) | — | — | 0.39 (0.08) | 0.46 (0.08) | — | — | 0.32 (0.09) | 0.46 (0.08) | 0.22 (0.09) |
MA(1) | 0.33 (0.07) | −0.61 (0.08) | — | — | −0.60 (0.08) | 0.42 (0.08) | −0.97 (0.02) | — | −0.97 (0.02) |
MA(2) | — | −0.37 (0.08) | — | — | −0.34 (0.07) | — | — | — | — |
Seasonal AR(1) | 0.32 (0.09) | 0.38 (0.09) | 0.32 (0.09) | 0.88 (0.07) | 0.92 (0.07) | 0.83 (0.08) | 0.92 (0.06) | 0.95 (0.06) | 0.93 (0.06) |
Seasonal AR(2) | 0.19 (0.10) | 0.31 (0.09) | 0.33 (0.10) | — | — | — | — | — | — |
Seasonal MA(1) | — | — | — | −0.56 (0.16) | −0.67 (0.17) | −0.46 (0.14) | −0.68 (0.13) | −0.54 (0.13) | −0.71 (0.14) |
Seasonal MA(2) | — | — | — | — | — | — | — | −0.23 (0.12) | — |
AIC | 1309.7 | 1404.7 | 1269.8 | 1482.2 | 1635.5 | 1303.8 | 1590.1 | 1185.8 | 1467.7 |
Parameter estimates are displayed with standard error in parentheses.
To evaluate the predictive ability of the optimal model for the overall series, the model parameters were estimated using the first 11 years of monthly data (April, 1999, to March, 2010; 132 monthly values) and then used to predict the last year of our period (April, 2010, to March, 2011; 12 monthly values), including 95% prediction intervals. The estimated model performed well, with the 95% prediction interval containing the actual values for all of the 12 time points (Figure
Comparison of observed and predicted rates based on the first 11 years of data.
We described the monthly rates of presentations to the ED for COPD during 12 years in a Canadian province and used time series modeling techniques to quantify the patterns. With over 188,000 ED visits for COPD during the study period, we demonstrated stable presentation rates, variations across age, sex, and location, and a consistent pattern reflecting the influences of environmental conditions. These findings were similar to our earlier investigation of ED visits for COPD in Alberta during 1999 to 2005 [
The stable presentation rates observed in this study are similar to those seen in other jurisdictions, where stable temporal trends were observed during 1999 to 2010 (adults aged ≥25 years) [
Relatively few studies have used time series analysis methods to investigate patterns of COPD presentations. In Australia, COPD mortality rates during 1922–2005 were used to create functional time series models and forecast age- and sex-specific mortality rates for 2006 to 2025 [
The study has several strengths and limitations. The large sample size, 12-year time frame, the use of population-based databases, and time series methods are all strengths of the study. In addition, the use of diagnostic codes assigned by trained medical records nosologists strengthens the validity of the work.
Study limitations include the lack of linked data on treatment or contact with other health care services before the ED treatment; our focus on ED visits did not identify patients with COPD who used alternative sources for the delivery of acute care. In addition, the diagnosis of COPD is notoriously difficult to validate without supplemental records such as pulmonary function testing, simple imaging (e.g., chest radiography), or advanced imaging (e.g., CT scans of chest). While the diagnosis was not confirmed and this may both over- and underestimate COPD cases, ED-based research suggests that the diagnosis of COPD is valid and underestimates the problem [
In summary, rates of presentations to the ED for COPD have been remarkably stable and demonstrate a consistent pattern reflecting the influences of a cold-weather environment. During the cold winter months, indoor exposure, seasonal viral infections, and use of enclosed spaces predispose patients, and especially those with COPD, to upper respiratory tract infections, influenza, and pneumonia complications. SARIMA models fit the overall and subgroup data well and subgroups have similar parameter estimates. Such models can be helpful for forecasting and future health care planning.
We provide a brief overview of the Box et al. [
Suppose that
To enable the time series analysis approach, the series must not exhibit trends in the mean or variance. If a series does exhibit trends, the series can be differenced. Let the difference operator be denoted by
In general, the analyst would perform an iterative procedure to determine a suitable model. The autocorrelation function (ACF) and partial autocorrelation function (PACF) would be plotted to identify the preliminary values for
This study is based in part on data provided by Alberta Health. The interpretation and conclusions contained herein are those of the researchers and do not necessarily represent the views of the Government of Alberta. Neither the government nor Alberta Health expresses any opinion in relation to this study. This study has not received funding. Sponsors have not had any role in the study. Dr. Rosychuk had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Dr. Rosychuk is salary-supported by Alberta Innovates-Health Solutions (AI-HS, Edmonton, Canada; formerly the Alberta Heritage Foundation for Medical Research) as a Health Scholar. Dr. Rowe’s research is supported by a Tier 1 Canada Research Chair in Evidence-Based Emergency Medicine (Canadian Institutes of Health Research, Ottawa, Canada). In the past 3 years, Dr. Rowe has been involved in primary research funded by Merck, CEMPRA, and GlaxoSmithKline (GSK). He is not a member of any speaker’s bureau or a paid consultant for any of these partners. The other authors declare no competing interests.
Rhonda J. Rosychuk, Erik Youngson, and Brian H. Rowe contributed substantially to the study design, interpretation of results, and the writing of the paper. Erik Youngson performed the data analysis.