Did Cannabis Legalization Accelerate THC-Positive Crash Trauma? What 11,491 Screened Patients Actually Show

Short answer: In a five-center San Diego County retrospective study, the share of screened motor-vehicle and motorcycle injury patients classified as THC-only positive increased from 7.3% in 2010 to 14.8% in 2018. Yet the investigators’ model did not detect a statistically significant change in that trend associated with their 2016 policy breakpoint (interaction estimate B = 0.683, p = 0.923). THC-only positivity also was not an independent predictor of more severe injuries after multivariable adjustment. These observations describe a selected trauma population; they do not determine whether legalization caused more crashes or whether any THC-positive individual was impaired when a collision occurred.

The underlying original study is Borst and colleagues, Driving under the influence: a multi-center evaluation of vehicular crashes in the era of cannabis legalization, Trauma Surgery & Acute Care Open (2021), DOI 10.1136/tsaco-2021-000736. The distinctions among a urine-screen result, a blood-alcohol grouping, a trauma outcome, and actual driving conduct are essential to interpreting it.

What question did the investigators actually test?

The research addressed two related but different questions. First, did the proportion of toxicology-screened crash-injury patients who tested positive for cannabis change from 2010 through mid-2018, including around a policy breakpoint the authors associated with California’s 2016 legalization? Second, was a positive toxicology category associated with worse injuries after accounting for other recorded variables?

Neither question is equivalent to estimating the risk that a cannabis-using driver will cause a crash. To obtain that latter quantity, researchers would need exposure information on relevant drivers who did and did not crash, a definition of acute impairment, and defensible control of other determinants of collisions. An analysis beginning with patients who already sustained crash injuries and reached participating trauma facilities cannot supply those missing denominators.

This is why the first number to inspect is not the percentage testing positive. It is the population from which the percentage was calculated.

Who was in the five-center cohort?

The investigators reviewed registries at five San Diego County trauma centers for January 2010 through June 2018. The 11,491 included patients were at least 16 years old, had motor-vehicle or motorcycle crash injuries, and had completed both the specified blood and urine toxicology testing. Of the included patients, 8,443 were motor-vehicle occupants and 3,048 were involved in motorcycle crashes. These were injured patients in a healthcare system, not a random roadside survey and not a roster of adjudicated at-fault drivers.

The investigators reported a mean toxicology-screening rate of 59.4% across the study period (reported SD 3.3 percentage points), with changes in testing coverage over time. Therefore, the 11,491 were not all injured road users treated during those years. People tested and not tested could differ in clinically and legally important ways. The separate medical-examiner dataset of 777 people who died at the scene involved different ascertainment and a shorter 2011–2018 window; it must not be merged into the hospital denominator.

What did “THC-positive” mean?

For the primary trauma-center classification, THC-positive meant a positive urine toxicology screen. The study does not supply a directly quantified, time-of-driving concentration of active THC in forensic venous whole blood for every THC-positive patient. The toxicology-positive label is thus evidence about the particular screening result, not itself a validated instantaneous measure of brain exposure, intoxication, impaired psychomotor performance, or collision causation.

The authors separately treated a reported blood-alcohol result above 0.08% as an alcohol-positive grouping criterion. That criterion must not be silently converted into a claim that the investigators performed a specified forensic headspace-GC whole-blood assay at the time of driving. Their report does not provide a case-by-case alcohol collection interval, complete specimen-handling record, or independently reviewed evidentiary chain of custody. Urine THC and a blood-alcohol classification are measurements from different matrices answering different questions.

For more context on why physical specimens are not interchangeable, see the published biological matrices guide and specimen collection guide.

How were the 11,491 patients classified?

The investigators used five mutually exclusive report categories. Their categories separate THC-only from alcohol-only and mixed-drug patterns so that “any cannabis-positive” is not confused with “THC-positive with no alcohol or other studied drug category.” The fifth category included methamphetamine or cocaine toxicology patterns and should not automatically be read as additional THC-only cases.

The counts in Table 1 sum to the 11,491 screened patients. Because the categories are mutually exclusive for this presentation, their percentages should not be summed indiscriminately with separately reported “any THC” percentages or fatal-scene figures. In particular, a positive urine screen does not establish route, quantity, exact time of consumption, or who was driving.

Did THC-only positivity rise over the observation years?

Yes, descriptively. The original paper reports the THC-only subgroup at 7.3% in 2010 and 14.8% in 2018, with a reported high near 14.9% in 2017. Those percentages reflect the study’s included, screened trauma cases. They do not measure the fraction of all drivers using cannabis, the county’s per-mile collision rate, the probability that a particular crash was cannabis-caused, or the prevalence of acute THC intoxication among every injured patient.

The study also noted a declining share of all-screen-negative patients and an increase in the methamphetamine/cocaine category. Several drug-use trends can change together as a population, its treatment practices, and testing procedures evolve. A higher share of THC-positive screens is therefore not an isolated controlled experiment in which only one legal rule changed.

Forensic interpretation should preserve the distinction between a descriptive time series and a causal counterfactual. Saying “THC-only positivity was higher at the end of the series” is supported by the reported data. Saying “a specific law caused that entire rise” requires analyses this dataset alone cannot provide.

Did the investigators demonstrate an accelerated trend after the policy breakpoint?

No statistically significant acceleration was detected in the authors’ specified comparison. They reported an interaction estimate of B = 0.683, with p = 0.923, when comparing the temporal trend before and after a breakpoint they associated with 2016 legislation. The report’s own conclusion was that positivity continued increasing over the longer observation window but the rate of increase was not shown to accelerate at that breakpoint.

A nonsignificant interaction is not a proof that the law had zero effect. It is a failure, within this particular model and dataset, to demonstrate the specified change. The historical definition and timing of the study’s policy breakpoint also matter: the reported model refers to a January 2016 breakpoint, and the original paper does not provide the multiple-policy causal identification necessary to treat that coding choice as a definitive estimate of a legal intervention.

The authors additionally reported changing screening coverage, a geographically confined population, and no matched comparison region with a differently timed policy history. These limitations prevent a clean separation of legislation, pre-existing cannabis-use trends, age composition, crash exposure, clinical testing decisions, and other time-varying influences.

Was THC-only positivity an independent predictor of more severe injury?

The paper distinguishes unadjusted subgroup comparisons from adjusted injury-severity models. Although THC-only patients differed from screen-negative patients on several demographic and descriptive variables, THC-only positivity was not an independent predictor of a higher injury-severity category in the reported multivariable analysis. That analysis adjusted for sex, age, race, admission year, payor, mechanism, and institution, and compared Injury Severity Score (ISS) categories with ISS 1–8 as the reference range.

In contrast, the study reported higher adjusted odds in several injury-severity categories for alcohol-only, THC-plus-alcohol, and methamphetamine/cocaine-category patients relative to the screen-negative comparison group. This is a group-level statistical association within a selected clinical cohort. It does not prove which drug caused a crash, whether an individual was impaired, or that every mixed-positive patient had a pharmacologic interaction.

Individual injury severity is affected by many independent factors: speed, crash geometry, impact energy, restraint use, protective equipment, vehicle type, medical vulnerability, and the circumstances of medical referral. Statistical adjustment for available variables does not make those unmeasured influences disappear.

What did the original patient comparisons show?

Table 2 keeps the comparisons anchored to their proper denominators and avoids treating different outcomes as interchangeable. “No seatbelt” is calculated among motor-vehicle occupants in each category, not among all 11,491 people or motorcyclists. The authors’ published tabular values are the appropriate source for that comparison.

The authors found age and sex distributions differed among the groups. Consequently, a raw difference cannot simply be labeled an acute cannabis effect. Also, an observation such as lower restraint use does not reveal whether a positive urine result preceded the event, reflected remote exposure, or co-occurred with other unmeasured risk-taking factors.

What about the 777 people who died at the crash scene?

The county medical-examiner records were analyzed separately, covering 2011–2018 rather than the full 2010–June 2018 hospital inclusion window. They recorded 207 THC-positive people among 777 scene deaths (26.6%). This number is not a mortality rate for the trauma-center patients: individuals who died before hospital admission are not exchangeable with those who survived to undergo the hospital’s completed screening protocol.

These decedents were not a verified set of at-fault cannabis-impaired drivers. The data concern THC positivity in people killed at crash scenes, with the same fundamental missing facts about timing, active exposure, driving role, impairment, and mechanism-specific causation. Combining 207/777 with the hospital cohort as if both shared one sampling frame would create a false denominator.

Can urine THC detection identify an impaired driver at the crash time?

No. A urine result speaks to the tested specimen and analytical target. It cannot automatically establish the concentration of active delta-9-THC in venous whole blood at a prior time, and even a measured contemporaneous blood THC concentration is not identical to a directly observed driving error or a scientifically validated individual impairment decision. The present study did not perform paired laboratory THC pharmacokinetics with crash-time driving measurements.

Another analytical distinction matters: a screening assay is a method-bound classification, not necessarily a mass-spectrometric quantified confirmation. Without a fully documented laboratory method and case-specific record, the reader should not invent a confirmatory cutoff, analyte identity, quantification limit, or admissibility conclusion. The confirmatory drug testing guide explains why a later confirmatory result would answer a narrower analytical question than impairment or fault.

For litigation, the relevant source documents include the exact specimen, collection time, chain of custody, analyte and assay, potential metabolites, measured units, laboratory quality controls, the driver’s identity, observations of behavior, and physical collision evidence. None of these can be replaced by a group percentage taken from a retrospective injury registry.

Does a statistically nonsignificant result exonerate or implicate an individual?

Neither. The absence of a significant THC-only effect in a multivariable injury-severity model concerns the investigators’ sampled population, covariates, outcome grouping, and statistical precision. It cannot certify that every THC-positive driver was unimpaired. Equally, the overall rise in positive screens cannot certify that an individual drove impaired or that THC caused a specific crash.

Legal causation, clinical diagnosis, and statistical association use different evidentiary standards and different facts. A case-specific analysis should first establish what specimens were tested and how, then whether the measured findings can be connected to exposure at the relevant time, then whether reliable behavior and event evidence support a particular inference. If those links are missing, the uncertainty is substantive rather than a wording issue.

What are the most important design limitations?

Conditioning on crash injury and completed screening: the sample includes screened people who reached five trauma centers after a crash. It excludes uninjured drivers, untested injured people, many minor incidents, and those who did not enter these hospitals. That design cannot estimate population crash risk directly. Screening rates also varied over time and by institution.

Exposure timing: the study’s THC-positive criterion was a urine screen, with no established individual interval between cannabis exposure and crash. Active blood THC concentration at the driving time was not the study’s defining measurement. The alcohol grouping likewise is not a retrograde calculation of crash-time BAC.

Outcome and confounding: motor-vehicle occupants and motorcycle riders were combined despite differing protection and injury mechanisms. Demographics, other substances, restraint behavior, crash physics, and institution influence case mix. Regression addresses only measured covariates. Neither admission injury severity nor scene death alone identifies who caused a collision.

Policy inference: the series predates and spans legal transitions, but without an appropriate counterfactual group and validated exposure history, neither the descriptive rise nor the nonsignificant interaction proves what cannabis legislation caused or prevented. A non-significant policy interaction is a narrow statistical observation, not proof of no possible policy consequences.

How should an attorney or expert frame the finding?

A careful affirmative statement is: among 11,491 screened vehicular-injury patients at five San Diego County trauma centers, the proportion classified THC-only positive increased over 2010–2018, while the authors’ specified 2016 trend interaction was nonsignificant and THC-only positivity was not independently predictive of their adjusted injury-severity outcomes. This is a fair summary of the study’s scope.

A statement the study cannot support is: California legalization caused a specified percentage increase in THC-impaired crashes, or one person’s urine THC screen proves intoxication and responsibility for a collision. The same caution applies to the separate medical-examiner data; THC positivity among people killed at the scene is not an attributable-risk fraction.

In expert testimony, the most useful role of this paper is to explain the difference between toxicology prevalence, clinical case selection, policy trend modeling, individual drug impairment, and proof of accident causation. Its value lies partly in demonstrating where each inference must stop.

Frequently asked questions

1. How many crash-injury patients were in this study?

The hospital analysis included 11,491 people aged 16 or older from five San Diego County trauma centers who had both relevant blood and urine toxicology results. They were screened injured patients, not all drivers or all crashes.

2. Did the proportion classified THC-only positive increase?

Yes. The published study reports 7.3% in 2010 and 14.8% in 2018 for its THC-only category. Those are proportions of the included screened patients, not proof of an equal change in all drivers or actual intoxication.

3. Did the study prove legalization caused that increase?

No. Its reported pre/post-2016 trend interaction was not statistically significant (B = 0.683; p = 0.923), and the design lacked the controls required to assign a causal policy effect.

4. Was THC-only status an independent predictor of more severe injury?

Not in the reported multivariable ISS analysis. Other positive-drug categories showed adjusted injury-severity associations, but none of these group results determines an individual’s impairment or collision cause.

5. Was THC measured in blood at the time of driving?

No. The hospital grouping used urine THC screening. It did not supply a measured crash-time forensic whole-blood THC concentration for each patient, nor a validated individual pharmacokinetic back-calculation.

6. Does 26.6% THC-positive among 777 scene deaths mean cannabis caused 26.6% of fatal crashes?

No. That percentage describes a separate medical-examiner sample. It is not a causal attribution, a population fatal-crash rate, or proof every positive person was an impaired, culpable driver.

7. What additional evidence matters in a particular impaired-driving case?

Specimen identity and handling, validated method and target analytes, timed quantitative results where available, medication and exposure history, contemporaneous psychomotor observations, driver identity, collision reconstruction, and plausible non-drug explanations must all be assessed. Registry prevalence cannot substitute for that record.

Original primary study

Borst JM, Costantini TW, Reilly L, Smith AM, Stabley R, Steele J, Wintz D, Bansal V, Biffl WL, Godat LN. Driving under the influence: a multi-center evaluation of vehicular crashes in the era of cannabis legalization. Trauma Surgery & Acute Care Open. 2021;6:e000736. doi:10.1136/tsaco-2021-000736. The measurements and comparisons described here come from its original Methods, Results, Tables 1–3, figures, and discussion. Findings from its cited background literature are not presented as results of the Borst study.