This was a matched pair population-based cohort study that used electronic medical record (EMR) data from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN). CPCSSN is a federated network of fourteen practice-based research and learning networks (PBRLN) across Canada. Primary care providers, known as sentinels, share clinical data on their patients with CPCSSN to form a national data repository of patient clinical records. The CPCSSN repository has data on over 1.8 million patients, contributed by over 1,500 primary care providers from British Columbia, Alberta, Manitoba, Ontario, Quebec, Nova Scotia and Newfoundland. While patients represented within this pan-Canadian data repository are older and have more females than the overall Canadian population, the data repository is representative of Canadians who visit primary care [19].

This study excluded clinical data from Manitoba and Quebec as we were unable to classify COVID-19 cases in the data from these two PBLRNs. In these two provinces the lab results confirming a COVID-19 infection were recorded almost exclusively as PDF documents, which are not available (at this time) within the CPCSSN database.

Participants included all patients that visited their primary care provider and met the inclusion criteria for COVID-19, RTI, or negative (see Additional file 1 for case definitions). To be included a patient had to have a documented birthyear and sex.

A patient’s visit was included as a COVID-19 primary care visit if there was a diagnosis and/or lab-confirmation of the COVID-19 virus during a primary care visit between April 2020 and December 2021. A patient could only be included as a COVID-19 visit once (incident case = index event).

A patient’s visit was included as an RTI primary care visit if they had a diagnosis of RTI during a primary care visit between April 2020 and December 2021. Encounters were only included if there was no mention of COVID-19 in any record associated with the encounter date.

A patient’s visit was included as a ‘negative’ primary care visit if they had a visit with no diagnostic or lab data corresponding to a condition with respiratory symptoms (see Additional file 1 for list of exclusion criteria). Encounters were only included if there was no mention of COVID-19 in any record associated with the encounter date.

Each group was mutually exclusive. Other than the COVID-19 visit group, a patient visit was eligible to be included once every six months. A patient visit (for RTI or negative) within six months of a previous visit (for RTI or negative, respectively) were not eligible for inclusion as the outcome was evaluated up to 180 days after the index event.

We also calculated the proportion of visits that had COVID-19 documentation within the EMR out of all visits during the study time period.

Four outcomes were evaluated at four follow up intervals (30 days, 60 days, 90 days, and 180 days): (a) receipt of an antibiotic prescription; (b) receipt of a non-antibiotic prescription; (c) a subsequent primary care visit (for any reason); and (d) a subsequent primary care visit with a bacterial infection diagnosis (see Additional file 2 for a full description of how each outcome was defined).

In order to compare the COVID-19 visits with each of the comparison groups the COVID-19 subjects were matched (1:1) to an RTI visit and a negative visit on the following covariates: index month (month of diagnosis), age group (0–18, 19–39, 40–64, 65 +), sex (male, female), and province (British Columbia, Alberta, Ontario, Nova Scotia, and Newfoundland).

Patient age was determined at time of the index visit. Rural or urban patient locations were determined based on the second digit of the postal code, which indicates whether the patient lives in an urban (1–9) or rural area (0), as defined by Canada Post delivery areas. If a patient was missing a postal code their rural urban status was classified using the postal code of the clinic.

A material and social deprivation index (Pampolon), derived from a linkage between postal code and census data, was used as a proxy for socioeconomic status [20]. Measures of material and social deprivation were derived from the postal code using the Statistics Canada Postal Code Conversion File, along with the material and social deprivation index [20]. This index uses socioeconomic indicators from the census, including education, employment, and income (the material component), as well as marital status and family structure (the social component), to assign scores to dissemination areas (DA) in the form of population quintiles. Mean imputation was used to compute a material or social deprivation score for patients that were missing this covariate.

CPCSSN validated case definitions were used to classify a patient’s comorbidity status, specifically for the following conditions: chronic kidney disease (CKD), chronic obstructive pulmonary disease (COPD), dementia, depression, diabetes, epilepsy, hypertension, osteoarthritis, Parkinson’s disease, dyslipidemia, and asthma [21,22,23,24]. A patient was classified as obese if there was a recorded BMI observation ≥ 30 kg/m2.

A patient was classified as a smoker, non-smoker, past smoker or unknown based on smoking risk factor records. Similarly, alcohol use was classified as yes, no, or unknown based on alcohol risk factor records.

A patient was classified as pregnant, HIV positive, or has cancer if they had an associated ICD-9 code during the study period (March 2020 to December 2022). The ICD-9 codes used to classify patients for each of these covariates are described in Additional file 2.

Conditional logistic regression, a specialized type of regression appropriate for individually matched case–control data, was used to evaluate the association between COVID-19 and each of the four outcomes. Each model was adjusted for location (rural or urban), material and social deprivation, smoking status, alcohol use, obesity, pregnancy, HIV, cancer and number of chronic conditions. As this was an exploratory analysis and there were no a priori hypotheses, all covariates were kept in the fully adjusted model.

The data analyses were completed using SAS© 9.4.