Our study was set in British Columbia (BC), Canada, a province with a current population of 5.5 million residents, most of whom live in urban areas. Almost all BC residents have publicly funded universal health insurance that provides access to medical care (including ICD implantation) that is free at the point of delivery. The Insurance Corporation of British Columbia (ICBC) maintains licensing data for all 3.2 million drivers licensed in BC and provides mandatory basic automobile insurance for all vehicles registered in BC.

We obtained granular health data from multiple sources, including information on the indication for and date of ICD implantation from Cardiac Services BC’s cardiac device registries (Item S2). ^,, We linked this to population-based administrative health data (e.g., hospitalization records, physician fee-for-service claims, outpatient prescription drug fills, death records) and province-wide driving history and crash data (e.g., details on licensure, crashes and traffic contraventions [“violations” or “tickets”]; Item S2). ^, Comprehensive details on data sources and linkage are available elsewhere. ^,,,,

We identified all individuals with a first ICD implantation occurring in BC between January 1, 2003 and October 31, 2018, then matched each ICD recipient to 3 controls based on age and sex (Item S3). We excluded individuals without an active driver’s license on the index date (defined as the ICD implantation date for ICD recipients and their matched controls). We excluded the matched controls of any excluded ICD recipient.

ICD indication (primary vs secondary prevention of sudden cardiac death) is not always captured in administrative data and was missing for 19% of the cohort (Item S1). Our team previously developed and validated an algorithm to impute ICD indication with 90% sensitivity and 89% specificity. We used a similar algorithm to impute ICD indication in 40 imputation sets. We split each imputation set into a primary prevention ICD cohort and a secondary prevention ICD cohort, then separately estimated road exposure in each cohort. The descriptive analyses and final results reflect values pooled across all 40 imputation sets. We analyzed primary prevention and secondary prevention ICD recipients separately because these groups exhibit substantial differences in baseline comorbidities, baseline medication use, prior health services use, risk of subsequent ventricular arrhythmia, and typical driving restrictions.

We estimated monthly “road exposure relative to baseline” (RERB) for each individual in our cohort to reflect the person-time or driver-kilometers during which they were at risk of crashing. We expressed RERB as a unitless proportion ranging from 0 (complete cessation of driving during the month) to 1 (no reduction in driving during the month). We generated RERB best estimates, lower-bound estimates and upper-bound estimates by following the steps below (described in greater detail in Item S4).

First, we used the 2003 Canadian Cardiovascular Society guidelines and empirical individual-level health data to establish the likely duration of the postimplantation driving restriction for each individual in our cohort. We set the duration of the driving restriction to: (1) 1 month after ICD implantation date for individuals in the primary prevention cohort; and (2) between 1 week and 6 months after ICD implantation date for individuals in the secondary prevention cohort, with a minimum restriction of 1 week after ICD implantation, 1 month after ventricular tachycardia (VT) if LVEF ≥35%, 3 months after VT if LVEF <35%, and 6 months after cardiac arrest or syncope. The date of VT, cardiac arrest or syncope was defined as the date of the hospital admission or outpatient clinic visit with diagnostic codes for the condition of interest. If there were multiple such medical contacts, we used the most recent date that fell prior to ICD implantation.

Adherence to medical driving restrictions is imperfect and decreases over time. We drew on published studies of primary and secondary prevention ICD recipients to generate a best estimate, a lower-bound estimate and an upper-bound estimate of the cumulative proportion of drivers with early-return-to-driving (i.e., nonadherence to guideline-directed driving restrictions) by month after implantation (Item S4). ^, We used these values to estimate monthly RERB for each individual in our cohort for the duration of their postimplantation driving restriction.

After their driving restriction ends, some patients do not return to their baseline driving routines because of illness, frailty, employment changes, or anxiety about sudden cardiac incapacitation while driving. We drew on prior studies to estimate that 10% of primary prevention ICD recipients and 20% of secondary prevention ICD recipients voluntarily never return to driving. ^, For individuals in our primary and secondary prevention ICD cohorts, we set the RERB to 0.90 or 0.80, respectively, for all months after their postimplantation driving restriction ended. For individuals with NYHA IV heart failure, we set RERB to 0 for the entire 1-year follow-up interval because these individuals typically have severe symptoms and functional limitations and are very unlikely to return to driving.

We set individual RERBs to 0 for ICD recipients and controls for all months after any of the following events, whether they occurred during or after the recipient’s driving restriction: (1) start of a license suspension or expiry that lasted for >30 days; (2) admission date for a subsequent hospitalization that lasted for >30 days; (3) subsequent syncope or cardiac arrest; or, (4) death. For controls, we assumed these were the only reasons that individual RERBs deviated from 1.

We set individual RERBs to 0 for 1 week after any subsequent cardiac device procedures (e.g., implantation, explanation, battery change, lead extraction) to account for temporary driving restrictions, with subsequent return to preprocedure individual RERB established as described above.

For our main analysis, we focused on a 1-month follow-up interval for the primary prevention ICD cohort and a 6-month follow-up interval for the secondary prevention ICD cohort, intervals intended to reflect the typical driving restrictions in force during the study interval (Item S1). We first compared crash incidence rates among ICD recipients relative to those among controls using quasi-Poisson regression (outcome = number of crashes as a driver for that individual in the follow-up interval; exposure of interest = status as ICD recipient vs control) without adjustment for estimated RERB but with adjustment for other potential confounders: driver age, sex and residential neighborhood household income quintile; Charlson comorbidity index ≥2 based on a 1-year lookback; number of overnight hospitalizations in the past year; number of physician clinic visits in the past year; presence of hospitalizations or clinic visits for substance misuse in the past year; number of active medications at baseline; presence of “other heart condition” (congestive heart failure, atrial fibrillation or flutter, other arrythmias, or chronic ischemic heart disease in 1-year lookback); full driver license (vs learner or novice license) at index date; number of crashes in the past year; number of impairment-related traffic contraventions in the past year; number of nonimpairment-related traffic contraventions in the past year; active vehicle insurance at any point in the past year; and index year (Item S5). This analysis is similar to the analysis performed in the original cohort study in that it examines crash risk per unit time, with the implicit assumption that road exposure in ICD recipients is the same as in controls. We then performed an analysis that adjusted for each individual’s road exposure by summing the monthly estimated RERBs across the follow-up interval and adding this as an offset variable to the quasi-Poisson regression described above. All models used quasi-Poisson regression because non-RERB-adjusted models and RERB-adjusted models that used standard Poisson regression demonstrated mild overdispersion (dispersion ratios of ∼1.2). We performed sensitivity analyses that included repeating the RERB-adjusted analysis using our prespecified lower-bound and upper-bound estimates for RERB. Finally, we examined crash rates by month in the first year after ICD implantation in each cohort, both with and without adjustment for RERB.

The University of British Columbia Clinical Research Ethics Board approved the study and waived requirements for individual consent (H16-02043). Data were deidentified before release to investigators. All inferences, opinions, and conclusions drawn are those of the authors and do not reflect the opinions or policies of the Data Stewards.