Estimating Cannabis Involvement in Fatal Crashes in Washington State Before and After the Legalization of Recreational Cannabis Consumption Using Multiple Imputation of Missing Values

Abstract

The government of Washington state legalized recreational cannabis consumption in December 2012. We used data on all drivers involved in fatal crashes in Washington in the years 2008–2019 (n = 8,282) to estimate prevalence in fatal crashes of drivers with ∆⁹-tetrahydrocannabinol (THC; the main psychoactive compound in cannabis) in their blood before and after legalization. However, nearly half of the drivers were not tested for drugs; we therefore used multiple imputation to estimate THC presence and concentration among them. We used logistic regression followed by marginal standardization to estimate the adjusted prevalence of THC-positive drivers after legalization relative to what would have been predicted without legalization. In the combined observed and imputed data, the proportion of drivers positive for THC was 9.3% before and 19.1% after legalization (adjusted prevalence ratio: 2.3, 95% confidence interval: 1.3, 4.1). The proportion of drivers with high THC concentrations increased substantially (adjusted prevalence ratio: 4.7, 95% confidence interval: 1.5, 15.1). Some of the increased prevalence of THC-positive drivers might have reflected cannabis use unassociated with driving; however, the increased prevalence of drivers with high THC concentrations suggests an increase in the prevalence of driving shortly after using cannabis. Other jurisdictions should compile quantitative data on drug test results of drivers to enable surveillance and evaluation.

Abbreviations

 
• PR

  prevalence ratio

 
• THC

  ∆⁹-tetrahydrocannabinol

In 2012, Colorado and Washington became the first 2 states in the United States to legalize recreational use of cannabis; 15 states and the District of Columbia had legalized it by November 2020 (1). Self-reported cannabis use has increased in recent years (2), as has the proportion of drivers on the road who tested positive for cannabis (3). After alcohol, cannabis is the drug most frequently detected in crash-involved drivers (4). Research suggests that driving after using cannabis may increase crash risk (5, 6).

As states relax cannabis regulations, it is important to quantify the associations of these legislative changes with the prevalence of recent cannabis use among drivers involved in crashes. However, most crash-involved drivers are not tested for drugs. Testing is more common among drivers in fatal crashes; however, drug information from national fatal crash data is known to be unreliable because of inconsistencies within and between states in testing and reporting (7).

To address these limitations, the Washington Traffic Safety Commission worked with the state’s centralized toxicology laboratory to append blood test results for cannabinoids, including ∆⁹-tetrahydrocannabinol (THC), the primary psychoactive component of cannabis, to their state’s fatal crash data (8). Detectable THC in blood suggests recent cannabis use (9). Washington’s data have been used to quantify the proportion of drivers in fatal crashes who tested positive for THC (8). However, this does not allow inference regarding the prevalence of THC among all drivers involved in fatal crashes because many are not tested for drugs. Analyzing only those who were tested would provide a valid estimate of prevalence only if the probability of being tested were uncorrelated with actual drug presence.

Multiple imputation (10) enables valid statistical inference in the presence of missing data by modeling the conditional distribution of the incomplete variable (e.g., presence of THC) in relation to variables associated with its observed values, variables associated with its probability of missingness, and other variables of analytic interest. Although more commonly used to model missing covariates (11), multiple imputation can also be used to model outcomes. The United States National Highway Traffic Safety Administration uses multiple imputation to estimate the distribution of blood alcohol concentration among drivers in fatal crashes who were not tested for alcohol (12). In another study, researchers used multiple imputation to estimate the prevalence of any cannabinoids, including but not limited to THC, among drivers in fatal crashes nationwide (13).

We previously used multiple imputation to estimate the crude proportion of drivers involved in fatal crashes in Washington State who were THC-positive and the change in the proportion after legalization of recreational cannabis use (14, 15). The present study extends our previous work by accounting for interactions between cannabis and other substances in our imputation model, examining THC concentrations in addition to binary THC presence, and estimating adjusted postlegalization changes in the prevalence of drivers positive for THC alone and in combination with other substances.

METHODS

Design, setting, and participants

We estimated the proportion of drivers involved in fatal crashes in Washington State who were positive for THC before and after recreational cannabis legalization. Data were from the Washington State Fatality Analysis Reporting System, a database comprising records of all persons and vehicles involved in motor vehicle crashes that occur on public roads in Washington state and result in a death within 30 days. The Washington Traffic Safety Commission maintains these data and shared them with the authors’ institution under a data-sharing agreement. We examined data from all motor vehicle drivers involved in fatal crashes in Washington state from January 1, 2008, through December 31, 2019. The authors had no contact with human subjects nor access to any personally identifiable information.

Variables

Main outcome.

The main outcome measure was the proportion of all drivers in fatal crashes who had a concentration of THC in whole blood of 1.0 ng/mL or greater (hereafter referred to as THC-positive). THC presence was categorized as follows:

• Positive if a sample of the driver’s blood was tested for drugs and THC was detected at a concentration of ≥1.0 ng/mL.

• Negative if the driver’s blood was tested and THC was not detected or if the driver’s urine was screened and no cannabinoids were detected.

• Unknown if the driver was not tested for drugs, if test results were unknown, if it could not be determined whether the driver was tested, or if only a screening test was performed and it detected cannabinoids. The screening test detects both THC and 11-nor-9-carboxy-tetrahydrocannabinol but does not distinguish between them. (11-Nor-9-carboxy-tetrahydrocannabinol is a nonpsychoactive compound that remains detectable for a longer duration after cannabis consumption than THC (16, 17).) THC was also categorized as unknown under the following special circumstances:

  • The driver tested positive for 3 drugs ranked higher than cannabinoids in the National Highway Traffic Safety Administration’s hierarchy of reporting drug test results, because the database only allowed the reporting of a maximum of 3 drugs (8).

  • A screening test detected cannabinoids but a confirmatory test did not detect THC between December 3, 2012, and May 8, 2014, during which time the state toxicology laboratory used testing procedures that only detected THC in concentrations ≥2 ng/mL (Fiona Couper, Washington State Toxicologist, personal communication, 2016).

Procedures for collection and coding of data pertaining to THC have been described by Grondel (8). Procedures for collection and coding of all other variables in fatal crashes were established and described by the National Highway Traffic Safety Administration (18).

Secondary outcomes.

Secondary outcomes included the proportions of drivers positive for THC at specific concentrations and the proportions positive for other substances. THC concentrations examined were 0, 1.0–4.9 ng/mL, 5.0–9.9 ng/mL, and ≥10.0 ng/mL. Other substances examined were alcohol and/or other drugs (narcotics, depressants, stimulants, hallucinogens, and phencyclidine) alone or in combination with THC.

Main independent variable.

The main independent variable was whether the crash occurred before or after Washington state legalized recreational cannabis consumption. The prelegalization period was January 1, 2008–December 5, 2012; the postlegalization period was December 6, 2012–December 31, 2019.

Statistical methods

Multiple imputation of missing values.

Data on the presence of THC, other drugs, and alcohol were missing for 49%, 45%, and 44% of drivers, respectively. Multiple imputation (10) was used to create 50 copies of the data in which missing values were replaced with imputed values. The focus of the imputation procedure was imputing missing values of THC, alcohol, and other drugs; other variables were imputed as needed to enable the imputation of those variables.

We performed the imputation using chained equations (11). We imputed alcohol and THC data using 2-stage models in which we first imputed a binary indicator for presence of the substance and then imputed the continuous concentration if the substance was imputed as present to allow for the possibility that predictors of presence of these substances and predictors of their concentrations might differ. THC presence was imputed conditional upon confirmed or imputed cannabinoid presence to accommodate a small number of records with unconfirmed positive screening tests (n = 91). Imputation was performed separately for the prelegalization and postlegalization periods to allow the relationships between the outcome variables and all predictors to vary in relation to the legal status of cannabis. Imputation models included variables associated with the values of the variable being imputed and/or its missingness, as well as other variables of analytic interest. Web Appendix 1 (available at https://doi.org/10.1093/aje/kwab184) provides further details on the imputation models, including a list of all variables used (Web Table 1). All missing values were imputed successfully, enabling analyses of the combined observed and imputed data to include all 8,282 drivers.

Validation of imputation model.

The performance of the imputation model was assessed by deleting the observed values of the outcome variables from a random subset of the complete cases, imputing them, and comparing the distributions of the outcome variables when derived from the imputed values versus from the corresponding observed values. Details of the validation procedure are provided in Web Appendix 2. Comparisons of the overall prevalence of THC, alcohol, and other drugs in the imputed versus observed data are shown in Web Table 2, and comparisons of the ratios of the prevalence of each respective substance after versus before recreational cannabis legalization in the imputed versus observed data are shown in Web Table 3.

Statistical analysis.

We estimated the association of recreational cannabis legalization with changes in the proportion of drivers involved in fatal crashes who were THC positive using prevalence ratios. Crude prevalence ratios were calculated by dividing the proportion of drivers positive for THC after legalization by the corresponding proportion before legalization. Adjusted prevalence ratios were estimated using logistic regression followed by marginal standardization (19, 20). Logistic regression models included a linear trend (number of months since January 2008), a binary indicator for the legal status of cannabis to model a step increase or decrease associated with legalization, and interaction of the trend with the legal status of cannabis to model a change in trend. The model was adjusted for the driver’s age (using age in years and its square to allow nonlinearity), sex, driver’s license status (valid, invalid, or unlicensed), vehicle type (light vehicle or large truck/bus), crash location (urban or rural), and seasonal variation (first 2 sine and cosine terms of the Fourier series expansion of week of year divided by 52). After estimating the model, we used marginal standardization to estimate adjusted prevalence ratios that compared the proportion of drivers who were THC positive after cannabis legalization versus the estimated proportion who would have been THC positive had legalization not occurred and prelegalization trends continued. Associations of cannabis legalization with other outcome measures (proportions of THC-positive drivers at various THC concentrations; positivity for alcohol and other drugs, alone and in combination with THC) were estimated similarly.

We performed all analyses in each of the 50 imputed data sets individually and then averaged the results obtained from each set. We computed variances of means, proportions, and prevalence ratios using Rubin’s Rules (10) to account for both the variability in the observed data and the uncertainty in the imputed data. Complete case analyses were also performed for comparison. The imputation and statistical analyses were performed using Stata, version 15 (StataCorp LP, College Station, Texas).

RESULTS

Characteristics of the sample

There were 8,282 drivers involved in fatal crashes over the period examined, of whom 735 (8.9%) tested positive for THC, 3,528 (42.6%) tested negative, and 4,019 (48.5%) were not tested or had indeterminate test results (Table 1). Survival status differed markedly between drivers with known THC status (positive or negative) versus those for whom THC status was unknown: 73% of those with known THC status died compared with only 17% of those with unknown THC status. Almost all drivers with known THC status also had known test results for alcohol and all had known test results for other drugs, whereas the majority of those with unknown THC status were not tested for alcohol or other drugs either. Among drivers with known THC status, those who were THC positive were younger, a greater proportion were male, and greater proportions were positive for alcohol and/or other drugs compared with those who were THC negative. Differences in driving history, vehicle characteristics, and crash characteristics were also observed.

Prevalence of THC-positive drivers

Using the imputed data in conjunction with the observed data, we estimated that the proportion of drivers involved in fatal crashes in Washington state who were THC positive approximately doubled, from 9.3% before legalization of recreational cannabis to 19.1% after legalization (Table 2). After adjustment for driver age, sex, license status, vehicle type, urban versus rural location, seasonal variation, and preexisting trend, we estimated that the postlegalization proportion of drivers involved in fatal crashes who were THC positive was 2.3 (95% confidence interval: 1.3, 4.1) times as high as it would have been had legalization not occurred (Figure 1).

Figure 2 shows the monthly proportions of drivers with various concentrations of THC and modeled trends. The proportion of drivers with THC concentrations of 1.0–4.9 ng/mL (Figure 2B) nearly tripled after legalization; however, the adjusted prevalence ratio (PR) was not statistically significant (adjusted PR = 2.4, 95% confidence interval: 0.8, 7.0). The proportion of drivers with THC concentrations of 5.0–9.9 ng/mL was higher after legalization than before; however, the adjusted PR was exactly 1.0, and the modeled postlegalization trend was indistinguishable from that predicted from continuation of the prelegalization trend (Figure 2C). The proportion of drivers with THC concentrations ≥10 ng/mL increased from 3.0% before legalization to 5.5% after legalization (Figure 2D). The adjusted PR for drivers with THC concentrations ≥10 ng/mL was much larger than the crude PR because the decreasing trend prelegalization would have predicted a lower prevalence of drivers with THC concentrations ≥10 ng/mL after legalization than before (Figure 2D).

The overall proportion of drivers positive for other drugs increased significantly after recreational cannabis legalization (Figure 1). The proportion with blood alcohol concentration ≥ 0.08 g/dL was lower after legalization than before, but the difference was no longer present after adjustment for covariates and prelegalization trend (adjusted PR = 1.0). The proportions of drivers positive for alcohol and/or other drugs in conjunction with THC were higher after recreational cannabis legalization than before (Figure 3B–3D), but they were not statistically significant after adjustment (Figure 1).

Analyses of complete cases, discussed in Web Appendix 3 and shown in Web Table 4, produced slightly different numeric estimates but generally showed the same major patterns of results.

DISCUSSION

As states legalize the recreational use of cannabis, it is importation to consider the potential impact of these laws on rates of motor vehicle crashes, injuries, and deaths. This is difficult, however, because of the limitations of most available data. We used data from the state of Washington, which has a centralized toxicology lab and detailed data on THC among drivers in fatal crashes that are unique among US states, to estimate the proportion of drivers in fatal crashes who had detectable THC in their blood before and after recreational cannabis legalization. Because not all drivers were tested for drugs and tests were not missing completely at random, we used multiple imputation to estimate the prevalence of THC among drivers who were not tested. We estimate that in the 7 years after Washington state’s legalization of recreational cannabis, the proportion of drivers involved in fatal crashes who had detectable THC in their blood was more than double what would have been expected if trends observed before recreational cannabis legalization had continued.

Detectable THC in blood generally suggests recent use of cannabis; thus, our results suggest that the proportion of drivers involved in fatal crashes who had recently used cannabis increased after recreational cannabis legalization, which in turn suggests that the frequency of cannabis use shortly before driving increased. However, depending on frequency of use, route of administration, dosage, and other individual factors, THC may sometimes be detectable in blood for days or even weeks after a person last consumed cannabis (16, 21), whereas acute impairment from cannabis generally subsides within a few hours (21). Thus, some THC-positive drivers may not have consumed cannabis recently nor been experiencing impairment when they crashed. In contrast, higher THC concentrations (e.g., ≥10 ng/mL) are unlikely to be observed more than a short time after cannabis consumption (9, 16). The increased prevalence of drivers with THC concentrations ≥10 ng/mL provides stronger evidence that the frequency of cannabis use shortly before driving increased after legalization. Also importantly, among infrequent cannabis users, THC present at the time of a crash may dissipate to below detectable levels before blood can even be collected (22); thus, it is possible that some surviving drivers who were not THC positive had actually consumed cannabis shortly before driving.

Relationship to other research

In an annual national survey, the proportion of adults who reported cannabis use in the past month increased from 6.8% in 2011 to 11.6% in 2019 (23). Washington state surveys indicate that rates of adult cannabis use in the past month increased from 7.8% in 2011 (before legalization) to 9.3% in 2013 (after legalization) and 15.6% by 2017 (24); surveys in Colorado found prevalences of 11.0% in 2011, 12.9% in 2013, and 17.2% in 2017 (25). A study of recreational cannabis legalization in Alaska, Colorado, Oregon, and Washington found that after adjustment for national trends and other covariates, the prevalences of past-month cannabis use and frequent cannabis use increased significantly after legalization among adults ≥26 years of age but not among younger adults (26).

Among studies of crashes before and after recreational cannabis legalization, our research is unique in having examined confirmed THC presence and quantitative THC concentrations of drivers; however, others have examined the impact of legalization on rates of total crashes, fatalities, or insurance claims irrespective of cannabis involvement. One study of legalization in Washington, Colorado, and Oregon identified a 7.8% increase in total traffic fatalities per capita after legalization compared with controls (27); another detected a temporary increase in fatalities per capita after retail cannabis sales began in these states, as well as in neighboring states without legalized recreational cannabis (28). Two studies reported increases of 1.8–2.1 fatal crashes per billion miles driven (a ratio increase of roughly 20%) in states with legal recreational cannabis after cannabis stores opened versus before legalization relative to various comparison states (29, 30); another found an increase of 1.46 fatalities per billion miles driven (roughly 15%) in Colorado but no significant increase in Washington relative to controls (31). Rates of collision insurance claims per insured vehicle increased by 12.5% in Colorado and by 9.7% in Washington after retail cannabis stores opened relative to control states but did not increase significantly in Oregon (32). Rates of police-reported crashes per registered vehicle increased by 5.2% in the same states relative to controls (33). Another study found that Washington and Colorado experienced increases in fatalities per mile driven similar to those in control states after opening retail stores; however, the study included more than a year after legalization in the “before” period (34). Two studies also examined changes in rates of crashes reported in federal data as having involved cannabis (34, 35); however, the reporting of drug involvement in those data are known to be inconsistent (7).

Unlike the previously referenced studies, we did not seek to quantify the association of recreational cannabis legalization with changes in total fatal crashes or total fatalities. However, our results can provide some insights. Assuming recreational cannabis legalization would only increase rates of crashes involving THC-positive drivers and not other crashes, the increase in total fatal crashes associated with legalization can be approximated by the ratio of the proportions of THC-negative drivers before versus after legalization. We found that the proportions of those who were THC positive were 9.3% before and 19.1% after legalization; thus, legalization could have been associated with up to approximately a 12% increase ((1–0.093)/(1–0.191) = 1.12) in total fatal crashes. Provided the relative risk of crashing for a THC-positive versus THC-negative driver is finite: Some THC-positive drivers still would have crashed had they not used cannabis, and thus the actual increase in total fatal crashes associated with legalization would be somewhat smaller. These results are consistent with and lend support to previous findings that total crashes, fatalities, or insurance claims increased by 5%–12% after cannabis legalization (27, 32, 33). However, estimates that fatal crashes increased by substantially larger amounts (29, 30) seem unlikely in light of our results, unless legalization also increased rates of crashes not involving cannabis.

Limitations

Several limitations should be noted. Nearly half of all drivers in the present study were not tested for drugs, so we used multiple imputation to attempt to estimate the proportion of those drivers who were THC positive and their THC concentrations. Our imputation model performed well when we simulated missing data by deleting a subset of observations at random and imputing them; however, results could be biased if either the presence or concentration of THC were correlated with other factors not accounted for in the model.

Regarding the association of recreational cannabis legalization with increased prevalence of THC-positive drivers in fatal crashes, it is possible that other factors caused or contributed to the increase. The prevalence of drivers positive for other drugs also increased significantly after cannabis legalization; it seems unlikely that was causal. Ideally, we would have compared Washington to other states that did not legalize recreational cannabis at the same time; however, we are not aware of any other state with comparable data on confirmed THC presence and concentration among drivers involved in fatal crashes.

Regarding practical significance, our results are not necessarily generalizable to other states. While other studies have observed increases in crash rates after recreational cannabis legalization among the first few states to legalize recreational cannabis, it is possible that associations might differ in states that enact similar legislation later. For instance, if early legalization in some states has influenced attitudes about cannabis and rates of use in other states, states that legalize cannabis later might experience smaller changes in THC involvement in traffic crashes than did states that legalized it earlier. Finally, the fact that crash-involved drivers were THC positive does not confirm that they had used cannabis recently nor that they were impaired (9). Research generally suggests that drivers with detectable THC have a somewhat elevated crash risk (5, 6, 36, 37); however, a recent well-controlled study found no relationship after adjustment for confounding factors (38).

CONCLUSION

In the 7 years since Washington state legalized recreational cannabis use, the proportion of drivers involved in fatal crashes who were THC positive was more than double what we estimate it would have been had legalization not occurred. Although the presence of THC does not guarantee either recent cannabis use or impairment, the proportion of drivers with high THC concentrations (≥10 ng/mL) also increased significantly, strongly suggesting that the prevalence of cannabis use shortly before driving increased after the legalization of recreational cannabis use. Comparison to other states was not possible because detailed quantitative data on THC detected in blood of crash-involved drivers are not available from other states. Other jurisdictions should compile quantitative data on drug test results of crash-involved drivers to enable surveillance and evaluation.

ACKNOWLEDGMENTS

Author affiliations: AAA Foundation for Traffic Safety, Washington, District of Columbia, United States (Brian C. Tefft, Lindsay S. Arnold).

We thank Dr. Staci Hoff and Dr. Fiona Couper for assistance in interpreting the data and for helpful comments on an earlier draft of this article, and Dr. Hoff for providing the data.

The data examined in this study are available upon request from the Washington Traffic Safety Commission (https://wtsc.wa.gov/request-fatal-crash-data/).

Conflict of interest: none declared.

REFERENCES