Did Starting Medical Cannabis Change Simulator Driving? A 44-Person Feasibility Study

What did the study actually find?

The most informative comparison was not a single mean reaction time at follow-up. It was whether the change from baseline to approximately one month differed between people starting medical marijuana and controls assessed at the same intervals. The investigators used linear mixed-effects models incorporating group, time, sex and a group-by-time interaction, with a participant-level random intercept. All six reported interaction confidence intervals included zero, and the corresponding reported P values ranged from .25 to .97 except for a .40 and .48 result; none met the authors’ .05 threshold.

This is a failure to demonstrate a difference in this small, incomplete sample—not a proof of equivalence, safety, unchanged ability in every driver or an absence of acute THC effects. The original article expressly characterized its driving findings as exploratory. Its main contribution was showing what is feasible and what can go wrong when researchers try to observe medical-marijuana initiation in a realistic simulator among older adults.

Who took part, and when were they tested?

The cohort comprised 44 adults, all at least 50 years old and reporting chronic nonmalignant pain. Twenty-three were recruited while newly registered to obtain medical marijuana through a Florida treatment clinic; 21 nonusing comparison participants were recruited separately and matched on age, reported race and sex within practical limits. The total mean age was 62.80 years (SD 6.62); 25 were recorded as male and 19 as female. Both groups reported regular driving. Recruitment and data collection took place in Gainesville, Florida, with the baseline test before medical-marijuana initiation and follow-up approximately one month later.

The clinical and comparison groups were not assigned treatment at random. They were observed after making different real-world treatment decisions. Matching on age, race and sex can help comparability on those attributes, but it cannot remove all differences in pain, medications, sleep, motivation, illness severity or cannabis-use history. The original Methods describe a broader parent protocol with a planned three-month visit; the published feasibility analysis reports baseline and the one-month assessment only.

Was this a cannabis intoxication experiment?

No. The investigators did not administer a standardized THC dose on the testing day, randomize participants to THC and placebo, or assign identical time intervals from consumption to simulator exposure. The marijuana group had started obtaining and using treatment in everyday conditions. Exposure interviews described forms of administration that included smokable products, edibles, oral and sublingual preparations, and concentrates. Study personnel categorized use based on products, amounts, routes and frequency during the preceding seven days.

At follow-up, the authors classified 42% of their marijuana group as heavy users, 16% as moderate users, 32% as light users and 10% as having reported no use in the preceding week. These percentages describe a heterogeneous treatment-pattern classification, not measured THC doses. One participant’s classification cannot be translated into a universal serum or whole-blood level, an acute impairment duration, a predictable reaction-time effect, or a fact about a different driver.

Because no uniform dosing time was imposed, the observed one-month result is not an experiment on the immediate effects of driving one, two or several hours after a known cannabis dose. A genuine acute-dose question would require exposure details and appropriately timed specimens or performance tests designed for that question.

What did the simulator measure?

Participants used an immersive, high-fidelity simulator. A five-minute acclimation scenario introduced basic lane keeping, lane changes, mirror use and stopping. A roughly 15-minute main drive represented rural, residential, suburban and city settings with varying traffic, turns, pedestrians and cyclists. The test included a scripted hazard in which another vehicle pulled out and a divided-attention exercise in which a visual triangle appeared and the driver had to press a button. Reaction times, missed responses and related measures were captured by the simulator; a trained reviewer could use playback to evaluate errors.

The testing program is more behaviorally relevant than a questionnaire asking whether somebody felt impaired. Still, it is not actual driving in uncontrolled traffic. Participants encountered artificial hazards and a computer-based environment, and the simulator induced sickness in many of them. Driving-simulator reaction time is a measured task endpoint, not a measurement of THC, blood alcohol concentration, neural injury, a clinical intoxication diagnosis or a retrospective reconstruction of any roadway incident.

The researchers varied the two drive versions and the positions or colors of some visual events. Those scenario differences can help counter practice effects, but they also make it important to follow the published longitudinal analysis rather than selecting whichever task or visit generates the desired conclusion.

Why did one follow-up result appear to favor cannabis users?

At the one-month assessment, the authors highlighted divided-attention situation 3 (DA3): the marijuana-group mean was 2.57 (SD 1.63; n=14) and the control mean was 5.79 (SD 4.32; n=7), with an unadjusted between-group t(19)=-2.50 and P=.02. The reported Hedges g was -1.11 (95% CI -2.04 to -0.16). That isolated comparison is real reported data, and it would be misleading to omit it.

But the isolated follow-up comparison asks whether two small groups differed at that visit; it does not ask whether the difference in their changes since baseline was statistically established. The model addressing the longitudinal question found a DA3 group-by-time coefficient of -2.47 (95% CI -6.76 to 1.82), P=.25. Its interval crosses zero, so the data do not demonstrate a differential one-month change for that task. The article also examined several outcomes and visit contrasts, increasing the importance of distinguishing exploratory unadjusted results from a replicated, prespecified effect.

The original article’s reaction-time numerals are also not consistently labeled with an unambiguous time unit across the Methods and result displays. For this reason, the above 2.57 and 5.79 values are presented exactly as the publication’s reported scores, not converted here into milliseconds or asserted to represent a legally meaningful stopping distance. A forensic reconstruction would require validated time units, task geometry and an individual measurement record.

What were the six longitudinal driving outcomes?

The authors’ mixed models included participant-level repeated observations and controlled for sex. The key results below are the original source Table 5 group-by-time terms, not separate intervention-versus-control means at follow-up. Their direction reflects the model coding and must not be treated as a stand-alone dose-response coefficient.

These null comparisons do not turn all possible acute cannabis effects into null effects. They show that this study’s six longitudinal estimates were too uncertain to establish a change relative to its control group. The confidence intervals permit changes in either direction. More participants with complete measurements and better exposure timing could narrow those intervals.

How did simulator sickness affect the evidence?

Simulator sickness was not a minor side note. Only 18 of 44 people supplied completed driving sessions at both visits. The original account classified 22 of 26 missing cases at baseline and 19 of 26 at follow-up as attributable to simulator sickness, while its detailed table also distinguished participants who declined a later drive because of sickness at the earlier assessment. Noncompletion was therefore a major study characteristic, not an incidental lost measurement.

The authors ran missingness diagnostics and reported that Little’s test did not reject missing completely at random for the focal variables (χ²(67)=64.23, P=.57). That nonrejection does not prove that missingness was harmless, and it does not establish that people who stopped because of nausea would have performed like those who continued. Their attrition comparison additionally found that missing-data participants were more often female, prompting inclusion of sex in the models.

Mixed-effects models can use partial repeated data without forcing an analysis only of complete cases. This is an advantage, but it does not manufacture an unobserved simulator result, ensure no selection bias or substitute for a larger adequately powered sample. An expert relying on the model should disclose actual contributors per task and visit, acknowledge missingness mechanisms and avoid describing the analysis as a 44-person fully observed driving experiment.

Did the urine drug tests prove anyone was driving impaired?

No. At one month, 17 of 19 tested marijuana-group participants screened positive on the study’s urine drug test; the paper reported no marijuana-positive control participants. Urine screening was part of exposure verification, separate from the simulator’s kinematic and reaction-time outcomes. It was not a paired quantitative venous whole-blood THC measurement, a timed plasma pharmacokinetic curve, an observed sobriety examination or a forensic confirmation of impairment at the simulated driving moment.

Even a drug identification in a particular specimen supports the proposition that the test detected its target under its method and cutoff; the meaning depends on analyte, matrix and detection window. It does not identify the moment of cannabis use or show that a specific cognitive response was pharmacologically caused by THC at the visit. The study does not provide an individualized urine-screen-to-reaction-time conversion and does not validate a numerical impairment threshold from a positive urine result.

The published urine table’s heading and control percentage denominators are not entirely explicit, so this article preserves the clearly stated 17/19 figure without inventing an exact number of control urine tests. Any case-specific conclusion would require the actual toxicology laboratory report, collection chronology, assay method, quantitative or confirmatory information if obtained, and clinical or behavioral evidence.

For the broader laboratory interpretation framework, see the site’s independently published resources on confirmatory drug testing, biological matrices, and specimen collection.

What other differences could explain the observations?

At baseline, the marijuana group reported higher pain over the preceding 24 hours (mean 5.12 versus 2.73 in controls). The paper also describes a baseline difference in depression scores, although some numerical test and P-value entries in its descriptive table and prose are inconsistent. The direction and exact size of any psychological association should therefore be checked against the original measurements rather than resolved by guesswork. Pain, sleep, psychological status, medications, driving experience and willingness to complete a difficult simulator may all influence task performance or the completeness of the resulting record.

The study captured some driving habits, pain, cognitive screening and drug-use indicators but did not establish that the cohorts were exchangeable on every relevant factor. Participants chose treatment rather than being randomized to use or not use medical marijuana. Moreover, the article did not establish a controlled same-day interval since each participant’s last THC product, and cannabis preparations and routes varied substantially. Association with enrollment group cannot automatically be translated into the causal effect of THC itself.

When studying older adults, investigator and litigant attention should also extend to baseline vision, comorbid illness, medication combinations and simulator adaptation. The authors themselves highlighted gaps involving sleep quality, psychoactive medication exposure and broader comorbidity data. This is not a criticism of the investigators’ stated feasibility purpose; it is a boundary on using feasibility findings to establish individual fault or safe driving fitness.

What can a toxicologist or attorney responsibly cite from this paper?

The defensible description is that a nonrandomized feasibility cohort obtained some prospective simulator and exposure data from older medical-marijuana initiators, but the longitudinal analysis did not show a significant group-by-time change in the six reported driving outcomes. One unadjusted follow-up comparison was favorable to initiators; it did not survive as a demonstrated longitudinal interaction. The two statements are not logically inconsistent because they address different estimands.

For expert review, preserve the distinction among four layers: exposure history (reported product use and group membership), laboratory indication (urine screening at follow-up), measured performance (simulator reaction-time and response data where available), and legal inference (whether an individual was impaired at a specified time or caused an event). This paper observes portions of the first three; it does not independently establish the fourth.

The article should not be offered as proof that medical marijuana restores driving capacity. Neither should it be selectively quoted as though the absence of statistical significance settles every claim about harm. A non-significant interaction is a quantitative result with an interval of uncertainty. In litigation, counsel and experts should request the full testing chronology, exact measured outcome and actual toxicological specimen evidence rather than applying group averages to a named driver.

What would stronger evidence require?

A future study aimed at acute impairment would benefit from clearly identified cannabinoid products and doses, documented timing from consumption to specimen collection and driving task, measured and fully specified biological matrices, suitably sensitive behavioral endpoints, balanced samples and tolerable simulator procedures. If a treatment-initiation question is the goal, exposure variability needs careful measurement and the analysis should explicitly distinguish treatment effects from symptom changes in chronic pain or from between-group selection.

Repeated observations in adequately sized cohorts are particularly important because a person may have a different baseline ability and adaptation trajectory than another driver. Mitigating simulator sickness and investigating why missing outcomes are missing are equally substantive scientific tasks. On-road work may offer complementary functional validity, subject to appropriate controls and safety conditions. None of these improvements would create an automatic universal THC-to-impairment conversion.

Bottom line for forensic interpretation

The Ennis and colleagues 2026 study is not an affirmative validation of impaired or unimpaired driving from a urine cannabis result. It is a preliminary longitudinal simulator study with 44 enrollees, only 18 complete paired simulator records, considerable heterogeneity of cannabis use, a nominally significant single follow-up contrast and six nonsignificant adjusted longitudinal group-by-time findings. Both the observed task data and their limitations belong in any balanced scientific discussion. The full primary original may be reviewed at DOI 10.2196/79735.

Frequently asked questions

Does this study prove cannabis makes older drivers safer?

No. It did not randomize THC exposure, document standardized acute doses, or demonstrate a statistically significant change in driving outcomes relative to matched controls over time. The isolated follow-up group difference is not proof of benefit.

Did the researchers directly measure THC in venous whole blood?

The reported exposure verification was a urine drug screen. The paper does not present a contemporaneous forensic venous whole-blood THC series linked to each participant’s simulator performance. A urine result must not be mislabeled as a quantitative blood concentration.

Why does an unadjusted P=.02 not settle the issue?

It describes one difference between small groups at follow-up. The corresponding group-by-time model asks a different question—whether their changes from baseline differ—and reported P=.25 with an interval including zero. Multiple exploratory endpoints warrant restraint.

How many people completed both simulator assessments?

Eighteen of 44, or 41%, completed both visits: 12 in the marijuana group and six controls. The mixed models used additional partial observations where available, so different outcomes have different effective denominators.

Can 17 positive urine results be converted into an impairment rate?

No. Seventeen of 19 screened marijuana-group participants were reported positive at follow-up. Those are drug-screen observations, not 17 demonstrated episodes of unsafe driving, nor a validated individual time-since-use or THC effect estimate.

Does a nonsignificant group-by-time interaction prove there was no effect?

No. Each estimated interaction has uncertainty; all six reported intervals included zero. In this small study with substantial missingness, a failure to detect a difference does not establish equivalence, absence of acute effects, or safety for each person.

Can a courtroom expert apply the study’s reaction times to a collision reconstruction?

Not directly. The publication uses scripted simulator tasks with unit-reporting limitations, no matching real crash, nonrandomized exposure and heterogeneous participants. A case opinion needs independently established timing, validated measurements, actual scene and driving evidence, and an appropriately limited inference.