This is a retrospective longitudinal cohort study of adult patients who have been authorized for medical cannabis (the exposure) matched to patients selected from the general population of Ontario who did not receive cannabis authorization. Each authorized cannabis patient was matched to up to three controls.
To proceed to control matching, first, an index date is assigned to each patient who is eligible to be selected as control (from the general population) so that the distribution of the eligible controls’ index dates is similar to that of the cannabis patients. Next, baseline characteristics were assessed before or at the index date. Finally, each cannabis patient was matched to up to three controls based on age (±1 years), sex, Local Health Integration Network location, income quartile, and history of health conditions including diabetes, heart disease, chronic obstructive pulmonary disease, asthma, cancer, musculoskeletal issues, neurological issues, pain, behavioral issues, fatigue, malnutrition, and other metabolic diseases. Matching was completed with replacement and thus an unauthorized patient could have been utilized for one or more authorized patients.
Cannabis users and their matched controls were followed from the index date (first date of cannabis authorization for the cannabis cohort and pseudo date for the controls) until the occurrence of the event of interest, censoring (death or moved out of province), or the end of the study data (March 31st, 2017) which ever occurred first. Data collected during follow-up visits in the clinics served to assess cannabis exposure time. Cannabis could be consumed by smoking, vaporising or oral ingestion.
The study population was Ontario adult patients who received an authorization to access cannabis for medical purposes in a chain of cannabis clinics between April 2014 and March 2017. These clinics offer consultation for cannabis use and follow-up to all patients based on self-referral or physician referral (15). To be included in the initial matched cohort, patients had to be aged 18 years or over and have been registered as eligible for the Ontario Health Insurance Plan (i.e., residents of Ontario). Patients were excluded if they had invalid or duplicate identifiers. Controls who had any diagnostic codes related to cannabis use during the study period (ICD-10 codes T407 and F12) were excluded.
This study mainly used Ontario administrative health data that served to select the controls and assess the study outcomes and co-variates. The cannabis cohort was selected using data collected in a group of Ontario cannabis clinics. These data were described in a previous paper [Eurich, 2019]. Briefly, in the study period (2014-2017), cannabis access for medical use in Canada was conditional on obtaining a medical prescription and administrative authorization (from Health Canada). Thus, all patients in our cannabis cohort (group) were formally authorized to use cannabis. Patients could be referred in the cannabis clinics by other physicians or self-referred. A comprehensive assessment was made during the initial visit and follow-up visits and data were captured electronically with patients’ consent. As these rich clinical data are only available for the cannabis cohort, both the controls and cannabis cohort administrative health data were used to assess the study variables. The Ontario Institute for Clinical Evaluative Sciences (ICES) provided the administrative data. These data include individual data files for each beneficiary, inpatient files, physician billings (inpatient and outpatient physician services) and prescription drug claims (16). The Ontario Health Insurance Plan (OHIP) (17) contains information on physician services, including diagnostic codes. The Discharge Abstract Database (DAD) and the National Ambulatory Care Reporting System (NACRS) contain all data on hospitalizations and emergency department visits, respectively. For each emergency visit or hospitalization, up to 25 possible diagnoses were registered according to the International Classification of Diseases system- tenth Revision (ICD‐10). Of these entries, only one indicated the most responsible diagnosis for the visit. The administrative databases were linked using the unique and encrypted patient health insurance number and covered the period of April 24, 2012 to March 31, 2017. We have previously assessed the healthcare utilizations of the cannabis cohort compared to controls using these data [Eurich, 2020].
For primary CV endpoint, we considered emergency department (ED) visits or hospitalizations with a main diagnostic code for acute coronary syndrome (ACS) or stroke. The following ICD-10 codes were used to assess this outcome in the databases: I20, I21, I24, I60-I64 (see appendix 1 for more details).
A secondary outcome was defined as ED visit or hospitalization with a main diagnostic code for any CV event. The ICD-10 codes I00 to I99 excluding codes I05 to I09, (i.e., chronic rheumatic heart disease) were used to assess this secondary outcome (see appendix 1).
Demographic variables included age, sex, nearest census-based neighbourhood income quintile and area of residence. We also assessed the following existing morbidities in the period going from 2012 to the index date: asthma, diabetes, metabolic disease, CHF, COPD, cancer, musculoskeletal issues, fatigue, pain, behavioural issues and neurological disorders (see appendix 2 for ICD-9 and ICD-10 codes used to assess these variables). Finally, as only congestive heart failure was considered in the initial matching, we also assessed the presence of any cardiovascular event as well as the presence of ACS or stroke in the period before the index date (see appendix 2 for details on the definitions and ICD-9 and ICD-10 codes used to define these variables) to characterize CV event history.
Descriptive statistics were used to assess the characteristics of the study sample (mean and standard deviation or median for continuous variables; numbers and proportions for categorical variables). Incidence rates of CV events per 1000 person-years and 95% confidence intervals were calculated for each group. For both the primary and secondary outcomes, conditional Cox proportional hazards regressions, that account for the matching, were used to assess the association between cannabis use and the study outcomes. The models were further sequentially adjusted for history of ACS/stroke and for history of any CV event, respectively. Schoenfeld residuals were used to assess the proportional hazards assumption while the Martingale residuals were used to assess nonlinearity (for continuous covariates). Hazard ratios (HR) and 95% confidence intervals (95%CI) were derived for each model.
In sensitivity analyses, we stratified each outcome-specific analysis by sex to assess possible sex-difference. We finally tested for interaction between sex and cannabis exposure. For all analyses, a two‐side P<0.05 was considered as statistically significant. The analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA).