Concussions and risk of a subsequent traffic crash: retrospective cohort analysis in Ontario, Canada

Abstract

Background

Concussion is an acute injury that may contribute to short-term limitations and potential long-term risks.

Objective

To test whether a past concussion is associated with the risk of a subsequent serious motor vehicle crash.

Design

Population-based longitudinal cohort analysis.

Setting

Ontario, Canada, from 1 April 2002 to 31 March 2022 (178 emergency departments).

Patients

Adults diagnosed with a concussion (cases) or an acute ankle sprain (controls), excluding individuals with a disqualifying illness (blindness, dementia, delirium), severe cases resulting in hospitalisation or those who died within 90 days.

Primary measure

Subsequent motor vehicle crash requiring emergency medical care.

Results

We identified 3 037 028 patients, including 425 158 with a concussion and 2 611 870 with an ankle sprain. A total of 200 603 patients were injured in a subsequent motor vehicle crash over a median follow-up of 10 years, equal to an absolute risk of 1 in 15 patients (6.64 per 1000 patient-years). Patients with a concussion had a 49% higher motor vehicle crash risk compared with those with ankle sprain (adjusted relative risk=1.49, 95% CI 1.47 to 1.50, p<0.001). The increased risk was particularly high in the early weeks after a concussion, remained independent of other observed risk factors, applied to diverse clinical groups and was further accentuated after repeated concussions. The risk extended across a spectrum of crash severity, was accentuated for single-vehicle events, replicated in analyses with artificial intelligence methods adjusting for confounding and remained distinct from the risks of other unrelated medical emergencies.

Conclusions

This study suggests a significant increased risk of a motor vehicle crash after a concussion that may justify a safety warning from clinicians.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Comprehensive population-based longitudinal cohort analysis of concussion patients that avoids small sample size, skewed recruitment, surrogate endpoints or brevity bias.

• Detailed statistics adjusting for observed characteristics, secondary analyses testing dose-response gradients and tracer analyses examining other unrelated emergencies.

• Additional analyses applying artificial intelligence techniques as a further test of robustness and adjustment for confounding.

• Study limitations include a lack of data on concussion severity, prior brain injuries during remote years, interval treatments and amounts of traffic exposure.

• Additional limitations include unmeasured factors that change over time and that can predispose patients both to concussions and to a motor vehicle traffic crash.

Equator reporting

Strengthening the Reporting of Observational Studies in Epidemiology checklist (STROBE) statement for cohort studies attached as separate file.

Introduction

Motor vehicle crashes are a common cause of mortality, morbidity and economic hardship. One factor contributing to motor vehicle crash risk can be a medical condition that causes impairment; for example, uncontrolled epilepsy, sleep apnoea or a substance use disorder.¹ Thoughtful counselling and care of medical conditions, therefore, may lead to fewer traffic crashes and could help save lives (eg, nighttime continuous positive airway pressure for patients with sleep apnoea).² In addition, some policies mandate physicians to warn patients who are diagnosed with epilepsy, or another condition that potentially impairs a patient’s ability to drive a motor vehicle.^{3 4} An accurate estimate of motor vehicle crash risk is essential for balancing safety with freedom in the context of a medical condition.^{5 6}

A concussion is a type of traumatic brain injury causing temporary deficits in functioning after which the individual recovers.⁷ Symptoms after a concussion can sometimes linger for weeks, including headaches, insomnia, dizziness, amnesia, fatigue, confusion, irritability, depression, delayed reaction times and difficulty concentrating.⁸,¹⁰ Whether recovery is complete with no ongoing impairment is often uncertain.¹¹,¹³ We hypothesised a concussion might lead to neurocognitive deficits that increase subsequent motor vehicle crash risk.^{14 15} Alternatively, patients might react with sustained caution that would decrease motor vehicle crash risk.¹⁶ A further nuance is that a concussion can predispose patients to another concussion with further neurological consequences.^{17 18}

The objective of this study was to test whether a concussion was associated with the risk of a subsequent serious motor vehicle crash. That is, a concussion might be both a consequence of a past motor vehicle crash and a contributor to a future motor vehicle crash.^{19 20} A greater awareness of motor vehicle crash risk might lead to better care of patients immediately after a concussion, more efforts for prevention and informed traffic safety policy.^{21 22} In addition, other medical contributors to motor vehicle crashes might be identified by statistical analyses that combine electronic medical records with artificial intelligence algorithms to provide further perspective for understanding the potential traffic risk for patients after an acute concussion.²³,²⁵

Methods

Study setting

We conducted a population-based retrospective cohort analysis of adults diagnosed with a concussion in Ontario, Canada, from 1 April 2002 to 31 March 2022 (20 years), reflecting all data available and providing 1 year of minimum follow-up.²⁶,³⁰ Ontario was Canada’s largest province with a population of 13 410 100 in 2012 (study midpoint), a total of 178 emergency departments, a motor vehicle crash risk of 4.6 per 1000 annually for the general population (fatal and injury events) and an average cost of $CDN60 000 per collision (total societal).³¹,³⁴ During the study, universal health insurance covered outpatient and emergency medical care for all patients.^{35 36} About 1-in-500 adults seek medical care for a concussion each year in this setting.³⁷,³⁹

Concussion patients

We identified patients diagnosed with a concussion by assessing physician billing data using unique encoded patient identifiers by searching for a concussion diagnosis (International Classification of Disease version nine code: 850).⁴⁰ This diagnostic code has been validated with high specificity (99%), moderate sensitivity (46–76%), yet lacks a grade for concussion severity.^{41 42} Patients younger than 18 years were excluded to ensure each individual was eligible for a full driving licence. Similarly, patients with a past diagnosis of blindness, dementia or delirium were excluded to avoid those unlikely to drive.⁴³,⁴⁵ Patients admitted to hospital or who died within 1 month of injury were excluded to reduce confounding from severe brain injury or comorbidity.⁴⁶

Control patients

To test the association between concussions and crash risk, we selected active controls as patients diagnosed with an ankle sprain (code: 845).⁴⁷ The purpose was to identify controls who lived in the same community, shared healthcare access, sustained an injury, contacted a physician and were neurologically unaffected. This approach yielded estimates of crash risks after a concussion and after a minor injury (yet can underestimate relative risks).^{48 49} Patients younger than 18 years, diagnosed with blindness, dementia or delirium, or admitted to hospital or dying within 1 month were excluded to match the concussion patients. Patients with both a concussion and a sprain were included in the concussion group (overlap cases also examined in secondary analyses).

Baseline characteristics

Patient characteristics were obtained by deterministic linkages of healthcare records.⁵⁰ The demographic registry was used to identify a patient’s age (years), sex (binary), home (urban, rural) and socioeconomic status (quintile).^{51 52} The physician services database provided data on clinic visits, emergency contacts and hospitalisations in the year prior to injury.⁵³ Additional clinical details included diagnoses associated with motor vehicle risk including alcohol misuse, epilepsy, sleep apnoea, diabetes, glaucoma and depression. Diagnoses that were not documented were assumed to be not present. The databases contained no information on other factors that might influence risk, including concussion severity, mechanism of injury, extent of amnesia, past infractions, licence status, trip distances or prior brain injuries before the study.

Motor vehicle crashes

We identified subsequent motor vehicle crashes based on patients receiving emergency medical care in the region (178 hospitals).⁵⁴ Crashes were defined as incidents requiring acute treatment (codes: V00–V69) during the study (end-date: 31 March 2023).^{50 55} The methods were validated in past research and no cases were excluded.²¹⁵⁶,⁵⁸ Further details included position (driver, passenger, pedestrian), vehicle (regular car, light truck, motorcycle, heavy transport) and configuration (single, multiple vehicles).²⁴ Added features included season (four categories), day (weekend, weekday), time (six sectors of 4 hours), ambulance care (yes, no), triage severity (higher, lower), hospital admission (yes, no), length of stay (≤2 days, ≥3 days) and outcome (dead, alive).⁵⁴⁵⁹,⁶³

Statistical analysis

The primary analysis evaluated subsequent emergency visits for motor vehicle crashes and compared the two groups (concussion, ankle sprain). Adjusted comparisons used multivariable proportional hazards regression models to test the strength of association after accounting for available demographic and medical predictors. Hazard Ratios (HRs) were used for estimating relative risks to account for different durations of follow-up and censoring of interval deaths. Individuals with multiple crashes were analysed according to first incident to avoid duplications (analyses based on multiple crashes also tested in secondary analyses). Patients diagnosed with both a concussion and an ankle sprain were examined through sensitivity analysis.

Secondary analyses explored additional nuances to further check the robustness of a potential association between a past concussion and subsequent motor vehicle crash risk (online supplemental appendix S1 and S2). Prespecified subgroup analyses were used to check for replication according to younger age, patients specifically diagnosed with an alcohol use disorder and those with differing past healthcare utilisation. Similarly, subtype analyses were used to examine whether the association extended across the spectrum of differing crash times, collision configurations, individual positions, injury severities and hospital outcome (alive, dead). Results from primary and secondary analyses appear as relative risk estimates along with 95% CIs.

Additional analyses applied machine learning techniques to further address confounding by adjusting for all baseline characteristics (eg, exact age) and interactions (eg, alcohol misuse combined with depression). The approach used the XGBoost algorithm (V.1.4.0.1) running on the R platform (V.3.6.1) and trained on patients who had an ankle sprain to create an ensemble of multidimensional decision trees (incorporating all continuous and categorical baseline patient data). The resulting model was then applied to the cohort of concussion patients to estimate the expected motor vehicle crashes (online supplemental appendix S3). The ultimate multidimensional model then identified absolute and relative risks based on contrasting the expected to observed crashes at a 5-year landmark.

Further analyses tested other potentially relevant distinctions around motor vehicle crash risk (online supplemental appendix S4). Stringency tests were applied to re-examine subsequent motor vehicle risks after excluding all patients who had a prior motor vehicle crash in the decade before the index injury. Dose-response gradients were examined using time-varying transition models to distinguish intervals of patient-time for patients diagnosed with more than one concussion during follow-up.^{64 65} Control analyses were designed to explore alternative common outcomes to identify whether the observed risk of a motor vehicle crash after a concussion was distinct from the risk of conjunctivitis, pneumonia, appendicitis, kidney stones, meningitis or other unrelated medical emergencies.

Ethics statement

Study reporting followed the STROBE. Funding organisations had no role in the design and conduct of the study; collection, management, analysis and interpretation of the data; and preparation, review or approval of the submitted manuscript.

Patient and public involvement

Patients were not directly involved in setting the research agenda, designing the study approach or conducting the analysis.

Results

Overview

A total of 3 037 028 patients were identified, of whom 425 158 were diagnosed with a concussion and 2 611 870 were diagnosed with an ankle sprain (online supplemental appendix S5). The two groups spanned a diverse range of age, sex, home locations and socioeconomic status. The largest relative differences were that concussion patients were marginally younger; more likely to have a past diagnosis of syncope, anxiety, depression, substance misuse or attention-deficit hyperactivity disorder (ADHD); and more likely to have a prior emergency visit (table 1). Other differences were small or less frequent among concussion patients, including diabetes and hypertension. Health services utilisation in the year prior to injury was similar for the two groups when measured by average number of clinical contacts or hospital admissions.

Table 1. Baseline characteristics.

Variable		Acute Injury
		Concussion	Ankle
		(n=425 158)	(n=2 611 870)
Demographic
Age (years)	18–39	220 139 (51.8%)	1 110 621 (42.5%)
	40–64	147 604 (34.7%)	1 130 753 (43.3%)
	≥65	57 415 (13.5%)	370 496 (14.2%)
Sex	Male	240 614 (56.6%)	1 442 527 (55.2%)
	Female	184 544 (43.4%)	1 169 343 (44.8%)
Home	Urban	373 813 (87.9%)	2 346 064 (89.8%)
	Rural	51 345 (12.1%)	265 806 (10.2%)
Socioeconomic status*	Highest	92 574 (21.8%)	527 175 (20.2%)
	Next highest	86 340 (20.3%)	535 038 (20.5%)
	Middle	82 879 (19.5%)	527 277 (20.2%)
	Next lowest	81 015 (19.1%)	514 146 (19.7%)
	Lowest	82 350 (19.4%)	508 234 (19.5%)
Diagnosis†
Alcohol misuse	Yes	5302 (1.2%)	14 569 (0.6%)
Diabetes	Yes	28 185 (6.6%)	201 741 (7.7%)
Hypertension	Yes	40 385 (9.5%)	336 725 (12.9%)
Heart disease	Yes	29 127 (6.9%)	177 122 (6.8%)
Syncope	Yes	60 104 (14.1%)	191 735 (7.3%)
Sleep apnoea	Yes	29 262 (6.9%)	157 946 (6.0%)
Osteoarthritis	Yes	25 467 (6.0%)	192 547 (7.4%)
Fibromyalgia	Yes	5967 (1.4%)	31 807 (1.2%)
Stroke	Yes	4545 (1.1%)	13 784 (0.5%)
Epilepsy	Yes	5343 (1.3%)	16 649 (0.6%)
Depression	Yes	24 020 (5.6%)	87 498 (3.4%)
Anxiety	Yes	93 858 (22.1%)	407 674 (15.6%)
ADHD	Yes	4649 (1.1%)	10 873 (0.4%)
Learning disability	Yes	366 (0.1%)	1638 (0.1%)
Glaucoma	Yes	12 335 (2.9%)	77 553 (3.0%)
HIV or AIDS	Yes	312 (0.1%)	1934 (0.1%)
Cancer	Yes	20 464 (4.8%)	127 940 (4.9%)
General†
Clinical contacts	≥3	340 350 (80.1%)	1 993 664 (76.3%)
Emergency visit	Yes	224 984 (52.9%)	836 377 (32.0%)
Hospital admission	Yes	33 155 (7.8%)	157 535 (6.0%)

^*Based on home neighbourhood, missing data coded as lower.

^†Based on previous year.

ADHD, attention-deficit hyperactivity disorder.

Motor vehicle crashes

A total of 200 603 patients were injured in a motor vehicle crash during follow-up, equivalent to an average risk of 1-in-15 per decade. Patients after a concussion accounted for 33 508 crashes over 3 166 761 patient-years, equal to an absolute risk of 10.58 per 1000 annually. Patients after an ankle sprain accounted for 167 095 crashes over 27 022 446 patient-years, equal to an absolute risk of 6.18 per 1000 annually (marginally above population norm). The difference corresponded to a 61% relative increased risk after a concussion (relative risk: 1.61; 95% CI 1.59 to 1.63, p<0.001). The risk was especially high in the first month that showed a sixfold increased relative risk and accounted for over 5000 crashes itself (figure 1).

Figure 1. Cumulative incidence plots of absolute risk of a traffic crash. Panel A for short-term risks (days) and Panel B for long-term risks (years). X-axis shows time and Y-axis shows cumulative incidence of crashes (per 1000 patients). Red line for patients after a concussion, blue line for patients after an ankle sprain and square brackets for total count in each group with a crash by corresponding time. Results show substantial increased traffic risks for patients after a concussion, especially during the initial weeks.

Additional predictors

The risk of a motor vehicle crash was related to other characteristics. Specifically, younger age, male sex, a rural home location and lower socioeconomic status were each associated with increased motor vehicle crash risk (table 2). As expected, a diagnosis of alcohol misuse was a substantial risk factor. A diagnosis of diabetes, sleep apnoea, epilepsy, syncope, depression, fibromyalgia, ADHD and HIV were risk factors to lesser degrees. Conversely, diagnoses of hypertension or past cancer were not significant risks. The model fit was mediocre (c-statistic=0.603 at 5-year landmark). Adjusting for all measured characteristics showed a 49% relative increased risk for patients after a concussion (relative risk: 1.49; 95% CI 1.47 to 1.50, p<0.001).

Table 2. Predictors of traffic risks.

	Basic analysis*		Adjusted analysis†
	Relative risk‡	CI§	Relative risk‡	CI§
Predictor
Concussion	1.61	1.59 to 1.63	1.49	1.47 to 1.50
Demographic
Younger age (<40 years)	1.44	1.42 to 1.45	1.43	1.41 to 1.44
Elder age (≥65 years)	0.70	0.68 to 0.71	0.68	0.67 to 0.69
Male sex	1.17	1.16 to 1.18	1.16	1.15 to 1.17
Rural home	1.16	1.14 to 1.17	1.18	1.16 to 1.19
Socioeconomic status¶
Highest	0.84	0.83 to 0.85	0.85	0.84 to 0.87
Next to highest	0.93	0.91 to 0.94	0.93	0.92 to 0.95
Next to lowest	1.07	1.05 to 1.08	1.06	1.05 to 1.08
Lowest	1.22	1.20 to 1.23	1.18	1.16 to 1.19
Diagnosis
Alcohol misuse	2.13	2.05 to 2.22	1.62	1.56 to 1.69
Diabetes	0.96	0.94 to 0.98	1.15	1.12 to 1.17
Hypertension	0.76	0.75 to 0.78	0.95	0.93 to 0.96
Heart disease	0.89	0.87 to 0.91	1.05	1.03 to 1.07
Syncope	1.27	1.26 to 1.29	1.21	1.19 to 1.22
Sleep apnoea	1.13	1.11 to 1.15	1.14	1.12 to 1.16
Osteoarthritis	0.93	0.91 to 0.95	1.15	1.13 to 1.17
Fibromyalgia	1.28	1.24 to 1.33	1.26	1.22 to 1.31
Stroke	0.96	0.90 to 1.03	1.01	0.94 to 1.07
Epilepsy	1.57	1.50 to 1.64	1.28	1.22 to 1.33
Depression	1.45	1.42 to 1.48	1.21	1.19 to 1.24
Anxiety	1.38	1.36 to 1.39	1.30	1.29 to 1.32
ADHD	1.95	1.86 to 2.05	1.40	1.33 to 1.47
Learning disability	1.36	1.17 to 1.58	1.02	0.88 to 1.19
Glaucoma	0.77	0.74 to 0.79	1.00	0.97 to 1.03
HIV or AIDS	1.48	1.29 to 1.70	1.29	1.12 to 1.48
Cancer	0.79	0.77 to 0.81	0.96	0.94 to 0.98

^*No adjustments for baseline differences.

^†Adjusted for other differences through regression model.

^‡Estimates based on Hazard Ratio from Cox survival model.

^§CIs based on 95% distribution.

^¶Referent is middle socioeconomic status.

ADHD, attention-deficit hyperactivity disorder.

Specific patients

The increased traffic risk after a concussion replicated in diverse subgroups. For example, motor vehicle risk was higher for men than women, yet an increased relative risk was apparent in each subgroup (table 3). Moreover, the absolute risk was higher for a woman after a concussion than a man with no concussion (9.90 vs 6.70 per 1000, p<0.001). Similarly, average traffic risk was increased for low-income patients yet the absolute risk was higher for an affluent patient after a concussion than a disadvantaged patient with no concussion. Analyses based on age, home location, season, specific diagnoses and past healthcare utilisation replicated the increased risks after a concussion, as did analyses excluding patients with a crash in the prior decade (online supplemental appendix S7– S13)

Table 3. Additional analyses.

		Total Events	Risk after Concussion*	Risk after Ankle sprain*	Relative Risk†	CI	P value
Primary analysis
Traffic crash		200 603	10.58	6.18	1.61	1.59 to 1.63	<0.001
Patient group ‡
Age (years)	18–39	111 693	12.43	7.56	1.52	1.50 to 1.54	<0.001
	40–64	74 922	9.43	5.34	1.68	1.64 to 1.71	<0.001
	≥65	13 988	5.39	4.05	1.29	1.23 to 1.36	<0.001
Sex	Male	97 393	11.38	6.70	1.61	1.59 to 1.64	<0.001
	Female	103 210	9.90	5.77	1.60	1.58 to 1.63	<0.001
Home	Rural	24 315	11.21	6.98	1.49	1.44 to 1.54	<0.001
	Urban	176 288	10.49	6.09	1.62	1.60 to 1.64	<0.001
Income quintile	Highest	34 436	8.82	5.12	1.61	1.57 to 1.66	<0.001
	Next to highest	38 199	9.94	5.66	1.65	1.60 to 1.69	<0.001
	Middle	40 070	10.75	6.14	1.65	1.61 to 1.69	<0.001
	Next to lowest	41 583	10.96	6.62	1.56	1.52 to 1.60	<0.001
	Lowest	46 315	12.71	7.51	1.60	1.56 to 1.64	<0.001
Injury severity§
Ambulance	No	112 709	6.08	3.37	1.68	1.66 to 1.71	<0.001
	Yes	87 894	4.16	2.67	1.48	1.45 to 1.50	<0.001
Triage severity¶	Lower	153 556	7.77	4.71	1.54	1.52 to 1.56	<0.001
	Higher	47 047	2.55	1.37	1.79	1.74 to 1.83	<0.001
Hospital admission	No	191 296	10.13	5.88	1.62	1.60 to 1.64	<0.001
	Yes	9307	0.39	0.28	1.36	1.28 to 1.44	<0.001
Length of stay	≤2 days	194 198	10.28	5.97	1.62	1.60 to 1.64	<0.001
	≥3 days	6405	0.26	0.19	1.32	1.23 to 1.42	<0.001
Outcome**	Alive	200 004	10.56	6.16	1.61	1.59 to 1.63	<0.001
	Death	599	0.02	0.02	1.32	1.04 to 1.66	0.022
Crash details
Time of day††	Dawn	8841	0.43	0.26	1.52	1.44 to 1.61	<0.001
	Morning	39 271	1.93	1.17	1.56	1.51 to 1.60	<0.001
	Afternoon	52 628	2.61	1.57	1.58	1.54 to 1.61	<0.001
	Evening	56 101	2.80	1.67	1.57	1.54 to 1.61	<0.001
	Night	33 493	1.71	0.99	1.60	1.55 to 1.65	<0.001
	Late night	10 269	0.56	0.30	1.72	1.63 to 1.80	<0.001
Day of week	Weekday	146 054	7.62	4.43	1.62	1.59 to 1.64	<0.001
	Weekend	54 549	2.69	1.63	1.54	1.50 to 1.57	<0.001
Configuration‡‡	Single vehicle	54 352	2.88	1.60	1.69	1.65 to 1.73	<0.001
	Multivehicle	105 219	5.21	3.19	1.53	1.51 to 1.56	<0.001
	Non-motorised	30 323	1.45	0.90	1.52	1.47 to 1.57	<0.001
	Unlisted	10 709	0.59	0.31	1.76	1.68 to 1.85	<0.001
Role	Driver	103 556	5.21	3.13	1.57	1.54 to 1.59	<0.001
	Passenger	45 422	2.31	1.34	1.59	1.56 to 1.63	<0.001
	Pedestrian§§	51 625	2.63	1.53	1.62	1.58 to 1.65	<0.001
Sensitivity analysis
No past crashes¶¶		176 274	9.32	5.85	1.51	1.49 to 1.53	<0.001
No dual diagnoses***		186 860	9.42	6.18	1.43	1.41 to 1.45	<0.001
Negative controls†††
Conjunctivitis		75 181	2.56	2.40	1.02	1.00 to 1.05	<0.001
Sinusitis		37 915	1.40	1.18	1.12	1.09 to 1.16	<0.001
Pneumonia		195 236	7.51	6.24	1.20	1.18 to 1.21	<0.001
Appendicitis		30 729	1.23	0.94	1.29	1.25 to 1.33	<0.001
Cellulitis		219 699	8.06	7.16	1.09	1.08 to 1.11	<0.001
Gall stone		51 731	1.69	1.64	1.03	1.00 to 1.06	0.048
Kidney stone		79 844	2.63	2.55	1.00	0.98 to 1.02	0.909
Pancreatitis		23 455	0.83	0.73	1.15	1.11 to 1.20	<0.001
Meningitis		156	0.01	0.00	1.29	0.81 to 2.04	0.285

^*Risk is rate per 1000 person-years.

^†Calculated without adjustments for covariates.

^‡Subgroup with corresponding characteristics.

^§Restricted to crashes of corresponding severity.

^¶Higher is CTAS triage 1 or 2, lower is all remaining.

^**Denotes hospital case fatality.

^††Boundaries are 04:00 hours, 08:00 hours, 12:00 hours, 16:00 hours, 20:00 hours, 00:00 hours.

^‡‡ICD-10 codes single and multiple vehicle events.

^§§Includes bicycles, scooters and other non-protected road users.

^¶¶Excludes those with any crash in prior 10 years.

^***Excludes individuals with both a concussion and an ankle sprain.

^†††Evaluates medical emergencies unrelated to traffic risks.

CTAS, Canadian Emergency Department Triage and Acuity Scale; ICD-10, International Classification of Diseases, 10th Revision.

Potential confounding

Further artificial intelligence techniques were applied to check robustness by adjusting for interactions, non-linear relationships and potential unmeasured latent associations. Gradient boosting was tuned with a 0.04 learning rate, weight of 3 for minimum leaf count, maximum tree depth of 3 levels, total of 500 iterations and typical run times of an hour. The model fit was marginally improved (c-statistic=0.614 at 5-year landmark) and the calibration was excellent (online supplemental appendix S6). The model applied to patients who had a concussion estimated 16 298 expected crashes (in contrast to 22 896 observed crashes at 5-year landmark). This contrast suggested a 40% relative increased risk for patients after a concussion (relative risk: 1.40; 95% CI 1.38 to 1.43, p<0.001).

Multiple concussions

During follow-up, the concussion group accumulated a total of 489 440 total concussions (figure 2). This included 47 016 patients who had two or more concussions, accounted for 3928 crashes over 263 039 patient-years and had an absolute risk of 14.9 crashes per 1000 annually. A significant dose-response gradient was apparent (figure 2), equal to a 73% increase in traffic risk after a single concussion (95% CI 71% to 74%, p<0.001), a 114% increase after two concussions (95% CI 106% to 121%, p<0.001) and a 124% increase after three concussions (95% CI 107% to 143%, p<0.001). In contrast, a more modest dose-response gradient in crash risk was apparent for patients after multiple ankle sprains (figure 2).

Figure 2. Dose response gradients of traffic crash risks. Panel A is for patients after concussions and Panel B is for patients after ankle sprains. The X-axis shows count of incidents and Y-axis shows rate of traffic crashes (per 1000 patient-years). Red bars for concussion patients, blue bars for ankle sprain patients and vertical lines for 95% CI. Square brackets indicate total crashes and person-years at each interval. Results show substantial risks for patients after multiple concussions (absolute and relative to ankle sprains).

Crash features

Motor vehicle crashes varied in features allowing further exploration of motor vehicle crash risk after a concussion (table 3). Crashes assigned the highest triage acuity were common after a concussion and showed a distinct accentuation of risk after a concussion. In addition, single-vehicle crashes were the second most common crash configuration and also showed a further accentuation of risk after a concussion. Conversely, only minor relative risks were observed for conjunctivitis, sinusitis or other unrelated medical emergencies after a concussion. If overall motor vehicle crash risk had been the same between the two groups, the net estimate was equal to saving 1633 total ambulance transports, 59 978 total days in hospital and $CDN835 million in societal costs.

Discussion

Overview

We studied thousands of patients after an acute injury and found concussions were associated with an increased risk of a subsequent motor vehicle crash. The increased relative risk of a motor vehicle crash was large in magnitude, sustained in time, applicable to diverse patients, relevant for different locations and further accentuated late at night (table 3). The increased risk of a motor vehicle crash was especially evident for the first month after a concussion, distinct from other risk factors such as ADHD, unrelated to the risks of other medical emergencies and increased further after a second concussion. The magnitude of increase was greater than associated with sleep apnoea or epilepsy. These data are more powerful than past research on concussions and stronger than traditional road traffic safety science.⁶⁶

Limitations

One limitation of the study is that correlation does not prove causality. In particular, factors that predispose patients to concussions may also predispose patients to a traffic crash.^{67 68} This idea implies a concussion is not necessarily the cause of a subsequent motor vehicle crash and prioritising concussion treatment might not necessarily improve road traffic safety.⁶⁹ The prevailing risk after a concussion, therefore, underscores that some failures to stay out of harm’s way may reflect a general risk-seeking attitude, a tendency to attribute fault to others or some other characteristic that leads to a crash that could have been avoided.^{70 71} Such debates around the interpretation of risk are difficult to resolve since a randomised blinded trial of concussions would be unthinkable.

A related difficulty is in identifying mechanisms underlying an association between past concussions and future risks. One possibility is confounding, since many conditions affect driving risk beyond alcohol misuse, diabetes and sleep apnoea, as identified and adjusted in our study. A different explanation could be the specifics of each case since each concussion is unique (anatomy, energy, angularity, severity), average risks may not apply to distinct patients and diagnostic codes are inexact. Other possible mechanisms include decreased attention, delayed reactions, transient confusion, mood disorders, cardiac arrhythmias, memory deficits, sleep difficulties or other subclinical difficulties that may linger after a concussion.⁷²,⁷⁴ Further science on neurocognitive deficits after a concussion is needed.

Many other limitations merit emphasis. The study lacks an at-fault analysis, yet some estimation of responsibility can be based on single vehicle events.^{75 76} The study provides no adjustment for distances driven; instead, indirect inferences are grounded in the observed consistency across different subgroups such as women who tend to drive less.⁷⁷ The study offers no direct test of minor incidents, prior brain injuries or traffic in other regions; however, results were consistent for those with differing characteristics.⁷⁸ The study does not directly evaluate methods to mitigate risk, yet clinician counselling can be effective for patients with epilepsy or other medical conditions that require driver warnings.²¹⁷⁹,⁸¹ Together, these limitations are topics for future research.

Past research

Past research suggests concussions can sometimes lead to neurological disabilities that could contribute to traffic risks. A driving simulator study found decreased vehicle control and persistent driving impairments after a concussion (n=28).⁸² Another simulator study showed altered driving skills based on simulated driving performance for young adults post-concussion (n=51).⁸³ A survey of adults with a mild traumatic brain injury showed more aberrant driving behaviours as assessed by self-report (n=389).⁸⁴ Together, these studies suggest a potential association between concussions and traffic risks yet are limited due to small sample sizes, unrepresentative volunteers, artificial circumstances and surrogate markers of traffic risks (no real crashes).

Future research on the association between a past concussion and future crash risks is also limited by weaknesses shared by our study. One barrier is that intuitions are sufficiently powerful to seemingly negate the need for observational evidence. A different reason is that many people recognise an acute concussion, implement behavioural compensations and presume the adaptations are sufficient.⁸⁵ Some further contributors are the variable severity of a concussion, lack of a valid animal model for studying traffic risks and difficulties with long-term follow-up of adults in a dynamic free society.^{86 87} Another nuance might be misleading cultural norms such as a movie hero who is knocked out in one scene and seems perfectly recovered in the next.⁸⁸

Conclusions

The risk of a motor vehicle crash after a concussion suggests current mitigating efforts are insufficient⁸⁹; however, driving cessation may be unreasonable since the risk also extends to patients as pedestrians.⁹⁰ Instead, clinicians might warn concussion patients to be cautious about prevailing motor vehicle crash risks along with standard anticoncussion campaigns.^{80 91 92} In addition, effective care of symptoms might reduce risks after a concussion, such as treating headaches, insomnia, depression and substance misuse.⁹³,⁹⁵ Further strategies might include temporarily reducing traffic exposure by avoiding high-speed trips during adverse late-night hours.^{96 97} Ultimately, an awareness of traffic risk after a concussion may help protect the patient and the community.

Data availability statement

The study data set is held securely in coded form at ICES (formerly known as the Institute for Clinical Evaluative Sciences). While legal data sharing agreements between ICES and data providers (eg, healthcare organisations and government) prohibit ICES from making the data set publicly available, access may be granted to investigators who meet criteria for confidential access, available at www.ices.on.ca/DAS (email das@ices.on.ca). The full data set creation plan and analytical code are available from the authors on request, understanding that the computer programmes might rely on coding templates or macros unique to ICES. Parts of this material are based on data and information compiled and provided by Ontario Ministry of Health (MOH).