Setting and Data Sources
Our study took place in Alberta, a Canadian province with approximately 4.3 million residents, the vast majority of whom (> 99%) are registered with the province’s universal publicly funded health insurance plan. Each member of the provincial health insurance system is captured in the Population Registry, which includes general demographic information (e.g., date of birth, sex) and information on the date and reason for insurance cancellations (e.g., emigrated from Alberta or died). Each individual in our databases had a unique lifetime identifier, which allowed us to deterministically link between various health administrative datasets.
In Alberta, all interactions with the health system are captured in one or more health administrative datasets, including hospitalization and ambulatory care/emergency department (ED) visits (Morbidity and Ambulatory Care Abstract Reporting System [MACARS]) as well as physician visits (Physician Claim Database). Healthcare costs are also captured in these databases. Individual-level data on laboratory-confirmed RSV cases in Alberta were obtained from the Alberta Precision Laboratory databases, which capture the results of most respiratory tests performed in Alberta. Specifically, Albertans with a respiratory illness who are seen by a health professional can be tested with a respiratory laboratory panel that includes an RSV test. In recent years, RSV testing in Alberta has been primarily conducted on inpatients and in EDs, with limited testing in outpatient settings. Moreover, we obtained data on the national positivity rate for RSV (i.e., of all the tests for RSV, the percent that are positive) per week from the Public Health Agency of Canada .
In MACARS, healthcare costs associated with inpatient visits are estimated using Resource Intensity Weight (RIW) methodology, as described by the Canadian Institute of Health Information , which multiplies the individual case RIW based on age and case mix group by the average provincial cost per weighted case. Similarly, the costs associated with ambulatory care and ED visits are calculated using RIW based on age and the Comprehensive Ambulatory Classification System , which classifies individuals who received ambulatory care (i.e., ED visits, day surgeries) by a number of data elements, including diagnosis, interventions, mode of visit, and visit disposition, which determine their RIW [15, 16]. To estimate physician costs, we used the Physician Claims Database, which captures all physician billing claims in the province, including those from outpatient and inpatient visits. We did not estimate laboratory or prescription and over-the-counter drug costs, as these costs are not captured in the health administrative datasets.
Study Population and Design
Objective 1: Respiratory Syncytial Virus (RSV) Case Cost
We used a retrospective case–control design to estimate the attributable healthcare costs per RSV case in Alberta. We measured costs from the healthcare perspective and did not consider societal or personal costs associated with RSV infection. We assumed healthcare costs attributable to RSV equaled the mean differences in all-cause healthcare costs between cases and controls. We chose all-cause healthcare because very little is known about the long-term effects of RSV on health costs and outcomes, and an RSV case’s contacts with the healthcare system are often not captured under an RSV diagnostic code. The index date for cases was the date of their first interaction with the health system for an RSV-associated event or the order date for their laboratory test; matched controls were given the same index date.
We used two different case definitions to identify RSV in the health administrative data. Case definition one (i.e., laboratory-confirmed RSV) was defined as any individual who had a laboratory-positive RSV test. The date the RSV laboratory test was ordered was considered the index date for these cases. We enrolled laboratory-confirmed cases from 1 September 2014 to 31 August 2018, with costing follow-up occurring until 31 August 2019. The purpose of this case definition was to capture RSV cases with a high level of specificity. However, children and hospitalized cases are tested at a much higher rate than other populations in Alberta and generally have more serious outcomes, so this case definition likely overestimates the cost per case.
Therefore, our second case definition, “ARI attributed to RSV,” sought to capture possible RSV cases (i.e., not laboratory confirmed) by defining a case as a medically attended visit for ARI where the most responsible diagnosis could be attributed to an RSV infection. “Most responsible diagnosis” was defined as the first diagnosis classification by the Canadian International Classification of Diseases, Ninth (ICD-9) or Tenth Revision (ICD-10-CA) codes. Relevant ICD codes were based on a review of the literature [17, 18] and the expert opinion of two infectious disease physicians (see Appendix A in the electronic supplementary material [ESM]). Moreover, for case definition two, the event had to occur during RSV season (1 November–30 April) [19, 20]. Although a wide variety of infections can cause ARI, the methods to treat them, and therefore the costs, are generally independent of the causative agent, particularly for nonhospitalized cases. Since case definition two identified many cases (>3,000,000), we used the simple random sampling function in SAS© to select 10% of the population for the final costing analysis. The index date for these cases was the date of their first medically attended ARI event. For ARI attributed to RSV, we captured costs from 1 November 2010 through 30 April 2018, with costing follow-up until 30 April 2019.
Controls for both case definitions were drawn from Alberta’s publicly funded health insurance plan over the year of analysis (i.e., did not emigrate or die within 1 year of the index date). For laboratory-confirmed RSV, each case was matched 1:5 to a non-RSV control; we selected a high ratio to account for variability in control costs. For ARI attributed to RSV, the cases were matched 1:1 to a non-ARI control. In both analyses, controls were defined as no RSV or ARI code within the year of analysis. Controls were randomly selected from all possible matches. For both case definitions, matching of cases to controls was based on five criteria: (1) age in months for cases aged < 1 year and age in years for cases aged ≥ 1 year, (2) urban/rural status (living in urban [≥ 25,000 population] vs. rural [< 25,000 population] areas according to postal code and census data), (3) sex, (4) prematurity (gestational age < 36 weeks) for cases aged < 1 year, and (5) Charlson Comorbidity Index (CCI) score [21, 22] for cases aged ≥ 1 year. Matching was done using a macro program coded in SAS©. The CCI is a measure of the number and intensity of an individual’s comorbidities; a higher number signifies more comorbidities. Previous research has demonstrated that CCI score is a reliable predictor of costs . Moreover, by using the same index date for cases and their controls, we accounted for time of year and RSV season [19, 20]. Individuals could be an RSV case and/or control more than once (since they could acquire RSV multiple times); however, to allow for full cost follow-up, they had to have at least 365 days following their initial identification as a case or control prior to being considered a new case/control. Each laboratory-confirmed RSV case had at least one exact match, but 127 did not have five exact matches. They were retained in the analysis to capture the costs associated with these populations, as they were generally part of smaller demographic groups (e.g., children who were premature).
Objective 2: RSV Burden
To estimate the burden of RSV, we calculated the incidence of medically attended RSV from 1 September 2010 to 31 August 2019 (nine complete RSV seasons) in Alberta. We multiplied the weekly age-specific incidence of medically attended ARI by the national RSV laboratory test positivity rate. ARI was defined as a physician visit with a relevant ICD-9 code or a hospitalization or ED visit with a relevant ICD-10 code (Appendix A in the ESM). However, for the RSV burden analysis, the case did not have to occur during the RSV season, and a case was considered incident if the individual did not have an ARI-relevant ICD code in the preceding 30 days. We did not include any date limits so we could capture RSV cases throughout the calendar year and observe the seasonality of RSV infections. To determine the percentage of total RSV tests that were positive for RSV infection each week from 2010 to 2019, we used aggregated national data from the Public Health Agency of Canada .
Objective 1: RSV Case Cost
We tracked both case and control costs using the phase-of-care method , where costs were calculated over two different phases: (1) acute infection (30 days’ follow-up) and (2) continuing care (365 days’ follow-up). We selected 30 days to capture the majority of costs associated with an acute RSV infection, and our hospitalization data showed length of stays ranging from 0 to 22 days for 99% of cases. A continuing care phase of 365 days is consistent with other studies, allowing us to account for the long-term morbidity associated with RSV [24, 25] without becoming too far removed from the initial infection. We calculated the mean and standard deviation for the attributable RSV case costs stratified by RSV season, age, prematurity, CCI score, and sex and tested for significant differences in case and control costs using the Wilcoxon signed rank test. We selected a matched samples nonparametric test to account for non-normal distribution in costing data. We applied the Bonferroni adjustment as a multiple testing correction to account for the multiple subgroup analyses. Moreover, we calculated the mean and standard deviation of RSV-attributable case costs associated with hospitalization, ED visits and physician billing. All costs were inflated to Canadian dollars ($CAD), year 2020 values, using the healthcare consumer price index as reported by Statistics Canada .
Objective 2: RSV Burden
The age-specific weekly rate of RSV was calculated as the number of incident ARIs per week and age group, divided by the total age-specific population in that week in Alberta, multiplied by the percent of total RSV laboratory tests positive for RSV by week (Fig. 1).
RSV incidence rates are presented as RSV cases per 100,000 people per week and per year by age group. As our ARI incidence rates were derived from the total Alberta population, we did not weigh the outcomes by age and sex distributions.