EVIDENCE LIBRARY
Can a ‘Very Low Crash Risk’ Screen Miss Speeding After Drinking? A 16-Driver Study
A 16-driver pilot found more simulated speeding and mistakes at falling alcohol levels although UFOV still rated crash risk ‘very low.’ Read exact limitations.
Short answer: Yes. In a small 2015 controlled-laboratory pilot, eight young adults tested after a social drinking episode spent more time speeding and made more simulated driving errors at the study’s moderate, declining-alcohol assessment, even though their Useful Field of View (UFOV) software continued to assign the same “very low” predicted crash-risk category. The result is an important warning against substituting a categorical visual-attention score for directly observed driving behavior. It is not a validated UFOV false-negative rate, proof of a particular driver’s impairment, or a measured real-world crash risk.
The primary research is Tremblay and colleagues’ 2015 PLOS ONE pilot, DOI: 10.1371/journal.pone.0118348. Its most defensible finding concerns disagreement between two different laboratory endpoints: changing simulator behavior and an unchanged broad predicted-risk category. The distinction matters when experts consider whether a normal screening result excludes an adverse functional finding.
What did the young drivers actually do after drinking?
The eight participants in the alcohol-exposure group completed four assessments across roughly 24 hours. At the noon assessment, when their protocol called for an alcohol level in the 0.05–0.07 g/dL BAC-style range, they spent an average of 22.3% of the simulated drive above the speed limit, compared with 14.0% during their sober baseline drive. That is an increase of 8.3 percentage points, not an 8.3% relative increase. The within-group comparison was reported as p = .011.
The same group averaged 9.8 simulator mistakes at the moderate-level assessment versus 6.8 sober, a difference of three errors (p = .023). The simulator’s composite “mistakes” category encompassed multiple driving events and rule-handling errors; it did not report that three more real collisions had occurred. The researchers’ direct observations were of a programmed urban simulation, not vehicles driven on public roads.
Neither endpoint warrants an unqualified statement that the alcohol caused every additional error. The alcohol group experienced a social drinking setting, and study sessions always progressed in the same time order. The investigators did not conduct a randomized placebo-controlled alcohol comparison against another otherwise identical social session.
Who was studied, and what does the denominator mean?
Twenty university-associated drivers initially volunteered: ten in an alcohol group and ten in a no-alcohol comparison group. Four could not complete the protocol because of simulator sickness—two from each group. The analyzed sample was therefore 16 people, eight per group, not 20 completed four-session datasets. The alcohol group contained six men and two women and had a mean age of 21.6 years; the comparison group contained four men and four women and had a mean age of 20.9 years.
The authors explicitly stated that the study was insufficiently powered for a reliable statistical comparison between groups. They instead made repeated-measures comparisons within each group over time. Accordingly, the main alcohol-group contrasts do not amount to a direct randomized treatment-effect estimate. The small sample also makes the group means vulnerable to individual variability and prevents credible estimates of screening sensitivity, specificity, or prevalence of impairment.
Eligibility required drivers to be at least 19 years old, hold a valid licence, and meet the specified drinking-history requirements. This is not a sample of inexperienced teenage learners, professional commercial drivers, older adults, or all drivers apprehended for suspected DUI. The study population must remain visible whenever its findings are used to discuss a litigated case.
How were alcohol exposure and the four testing sessions established?
Participants first undertook a sober morning baseline at approximately 8 a.m. The alcohol group then consumed a prescribed amount of vodka during a supervised, hour-long, socially framed episode; their mean intake was 5.3 ± 1.4 servings, each defined in the protocol as 30 mL of 40% vodka. Follow-up assessments occurred around noon, 3 p.m., and 8 a.m. the next day. The no-alcohol group was tested at corresponding clock times without the drinking episode or equivalent social context.
The authors described the alcohol stages using BAC terminology: sober at baseline; approximately 0.05–0.07 g/dL at the first post-drinking assessment; approximately 0.01–0.04 g/dL at the later post-drinking assessment; and sober the following morning. A Dräger Alcotest 7410 GLC breath instrument was used to check alcohol status before test periods. A Widmark-based calculation informed the expected waiting interval.
Measurement boundary: The study does not document independent headspace-gas-chromatography testing of paired venous whole-blood samples for these comparisons. A breath analyzer’s result expressed on a BAC-like scale and a Widmark projection are not a measured forensic whole-blood ethanol concentration. The original authors’ “BAC” labels describe the experimental stages; they cannot be imported as case-specific analytical blood results or retrograde calculations. For the distinct analytical concepts, see breath alcohol measurement and the site’s overview of forensic pharmacokinetics.
What did the two principal research instruments measure?
The Virage VS500M motion-feedback simulator presented a roughly 13.6-minute urban drive on multiple screens. The vehicle-like cabin included steering, pedals, dashboard, mirrors, and programmed traffic. Its outputs included percentage of time speeding, a count of assigned driving mistakes, a weighted composite performance score, and individual steering, safety, legality, mobility, sharing, and ecodriving subscores. These are observed simulator behaviors and software scores, not actual traffic violations or verified collision events.
The separate UFOV test used touchscreen visual-processing tasks: processing speed, divided attention, and selective attention. Its software combined subtests into one of five predicted relative crash-risk categories, from “very low” to “very high.” That label was an algorithmic or normative classification, not a contemporaneous measure of how long the participant actually exceeded the speed limit. The study was designed in part to ask whether these outcome families moved together during alcohol decline.
A mean heart-rate measurement and other electrocardiographic heart-rate-variability measures were also collected. Such physiological measures address acute autonomic response under the protocol. They are distinct again from chemical ethanol analysis, visual-attention performance, simulated driving, and any individualized forensic impairment assessment.
What are the original four-session simulator results?
| Alcohol-group measure, n = 8 | Sober baseline | Moderate stage, ~0.05–0.07 g/dL | Lower stage, ~0.01–0.04 g/dL | Next morning |
|---|---|---|---|---|
| Time above posted speed limit, mean % (SD) | 14.0 (7.1) | 22.3 (8.4) | 13.1 (6.5) | 12.6 (7.8) |
| Simulator mistakes, mean count (SD) | 6.8 (3.5) | 9.8 (3.9) | 5.6 (2.6) | 5.6 (1.8) |
| Weighted simulator performance score, mean (SD), 0–100 | 33.9 (27.3) | 27.8 (23.5) | 52.6 (18.0) | 54.4 (19.5) |
| UFOV predicted crash-risk category | Very low | Very low | Very low | Very low |
Source: Tremblay et al., original Table 2 and accompanying post-hoc results, DOI 10.1371/journal.pone.0118348. Values are descriptive group means and standard deviations. The study’s 0.05–0.07 and 0.01–0.04 g/dL conditions reflect its breath-checked, BAC-styled protocol rather than a new direct venous-blood assay. Repeated observations on eight people should not be treated as 32 independent drivers.
For speeding and mistakes, the reported pairwise differences were strongest at the moderate stage relative to sober baseline. The weighted simulator score did not show that same straightforward decline: the authors reported higher overall scores at the later lower-alcohol and next-morning assessments compared with baseline. An expert cannot collapse all these distinct indices into one claim that every driving ability became progressively worse or progressively better as the displayed alcohol stage declined.
Why did “very low crash risk” remain unchanged?
The UFOV software assigned the alcohol group the same relative risk category at all four sessions, even while speed-limit compliance and errors changed in the driving simulation. The investigators also did not find a significant relationship between that categorical score and the simulated driving variables in their small data set. The notable discrepancy is between a coarse visual-processing-based category and specific driving behaviors recorded under a different task.
This result does not tell us how many intoxicated drivers in a population would be missed by the UFOV classification. To call a result a clinical or forensic “false negative,” a study would first need a defined target condition, an independently justified reference standard, a classification decision threshold, and enough observations to estimate diagnostic performance. Tremblay and colleagues did not establish such a validation framework.
There is also a ceiling-or-floor issue within some UFOV subtests. Processing speed was recorded at the shortest observed display of 17 milliseconds across sessions. An instrument that gives many high-performing young adults its best available score can have little room to signal modest change. But restricted range in a small pilot is not a measured universal sensitivity limit; a separate, adequately powered validation study would be needed to determine how the classification behaves in relevant populations.
Did the underlying attention scores change even when the risk category did not?
UFOV selective-attention processing time in the alcohol group numerically changed across the four assessments. The paper reported an overall time effect for that subtest (p = .033), but its comparisons against sober baseline at the three later assessment points were not statistically significant (reported p values .732, .068, and .064). No meaningful change was established by the first two UFOV subtests across those sessions. Thus a significant overall repeated-measures test must not be misrepresented as a significant moderate-alcohol-versus-sober selective-attention result.
The no-alcohol group improved on the selective-attention subtest at the later sessions. The authors reported comparisons versus baseline at seven and 24 hours (p = .033 and .014). That pattern suggests test-retest practice can affect some cognitive measurements. It is particularly important where the same person completes four sessions using a similar task and the instrument’s composite category remains stable.
For a forensic comparison, ask which raw subtest, scoring range, and time point is being cited. “Very low risk” alone discards information about variability, possible task floors, and behavior measured by a different instrument. Conversely, an individual raw-score difference without an adequate normative distribution or validated outcome threshold cannot supply a person-specific crash probability.
What about heart rate: could a physiological change prove impairment?
The study recorded increased mean heart rate during the moderate and lower-alcohol assessment stages compared with baseline, with corresponding changes in several heart-rate-variability measures. The alcohol-group mean heart rate was approximately 79.1 beats per minute at sober baseline, 93.4 at the moderate stage, 91.7 at the lower stage, and 79.0 the next morning. Reported post-hoc comparisons versus sober for the moderate and lower stages had p = .039 and .046.
These are physiologic group patterns that may be consistent with response to an administered alcohol episode. Heart rate is not a quantitative ethanol assay; it also varies with activity, hydration, stress, temperature, posture, medication, and individual physiology. An autonomic group response does not validate a blood-to-breath partition factor, establish an individual’s driving competency, or prove a collision cause. The laboratory’s separate physiological and simulator endpoints should remain separate.
What did the no-alcohol comparison group actually control?
Eight participants who did not consume alcohol completed the four scheduled simulator and UFOV sessions. Their driving-simulator variables showed no statistically significant repeated-session changes, while some UFOV selective-attention scores improved. That comparison helps characterize possible repeated-exposure effects. It is not the equivalent of randomized, double-blind placebo control.
The groups differed in whether they consumed alcohol, experienced the social drinking context, and potentially anticipated changes in performance. The analysis was restricted by design to within-group tests rather than direct inferential comparisons of each group’s changes. The next-morning improvements seen in some alcohol-group simulator subscores remain compatible with practice, adaptation, social expectancy, time-of-day differences, or other explanations as well as changing alcohol exposure. The observed study contrasts do not independently partition those mechanisms.
When an expert uses these results, ask for the analysis establishing that a change is attributable to administered ethanol rather than merely to different sessions. The experiment offers valuable hypothesis-generating data but does not deliver that level of individualized causal certainty.
What is the precise distinction between “BAC,” impairment, and crash risk here?
| Concept | What this study supplied | What it did not establish |
|---|---|---|
| Alcohol exposure | Observed vodka dose, elapsed time, Dräger breath checks, and Widmark-guided target stages | Contemporaneous direct forensic whole-blood GC concentration or a validated individualized prior BAC |
| Driving behavior | Percent simulated time speeding, composite mistakes, task-specific simulator scores | Actual road operation, verified real-world crash events, or individual legal impairment classification |
| Visual processing | UFOV processing, divided/selective-attention tests, and a software-generated relative category | Universal sensitivity/specificity, personal crash probability, or a blood ethanol measurement |
| Physiologic response | Electrocardiographic heart rate and heart-rate-variability indices | A causal biomarker uniquely attributable to ethanol or a functional driving impairment threshold |
| Case-specific inference | Selected pilot-group observations under controlled task conditions | Proof a named person was driving unsafely, violated a specific law, or caused an actual collision |
The terms are not interchangeable. A breath reading is an exhaled-air observation; a simulator error is a task event; a software risk label is a predicted category. None is a direct measurement of the other. This is why the sample’s “very low” predicted category cannot be interpreted as either confirmed safe driving or a verified contradiction in measured blood alcohol.
How much weight can a court or expert place on this pilot?
Its major strengths are repeated observations of the same drivers, a detailed task protocol, a time course after an observed drinking episode, and direct collection of simulator outcomes rather than relying solely on self-report. It asks a useful measurement question: did a general visual-attention prediction move in the same direction as specific driving behavior? The observed answer, within this selected sample, was no.
Its limits are equally material. With eight alcohol-exposed completers and eight comparison participants, an effect can be unstable and individual driving histories difficult to account for. The paper’s authors expressly stated that the study was underpowered for between-group inference. There was no randomized matched placebo condition, no real traffic exposure, and no independent gold-standard whole-blood assay. Social drinking cues, expectation, session sequence, fatigue, and familiarity with the simulator were not fully isolated. Multiple endpoints and comparisons increase interpretive complexity.
Observed moderate-stage differences should be described with their original denominators and reported within-group p values, not recast as a validated impairment cutoff. A non-significant UFOV contrast cannot be translated into “alcohol has no visual effects,” and a stable UFOV category does not prove a specific person’s simulator errors were alcohol-caused. The scientific proposition is narrower: a broad screening category may not track a particular behavioral task outcome in a small experimental sample.
What documents would support or challenge application to a DUI case?
A case-specific review would ordinarily distinguish the timestamp, actual biological matrix and device, sampling sequence, calibration and quality controls, instrument output rather than reconstructed blood-equivalent labels, evidence of absorption or elimination phase, and time between driving and testing. This experiment supplies no single-person BAC adjustment and no universal dose-to-impairment conversion. The methodology of pharmacokinetic reconstruction is therefore a separate issue requiring its own evidence.
For the performance side, identify whether the asserted deficit concerns speeding, simulated handling, visual detection, divided attention, lane position, or verified dangerous driving. Obtain the underlying task definitions, individual records where available, scoring floors, post-hoc contrasts and comparator population. Do not let a categorical risk score substitute for raw performance measures or vice versa. When researching analogous experiments, the site’s evidence library is an index, not a replacement for the original methods and data.
Finally, causal interpretation requires incident-specific facts about road conditions, driving choices, vehicle dynamics, other substances, medical conditions, timing, and alternative causes. Neither a pilot simulator result nor an isolated UFOV score resolves that evidentiary burden. The original article is useful chiefly for showing why the types of measurements should be kept distinct.
Frequently asked questions
Did a “very low” UFOV crash-risk category mean the drivers were safe?
No. It was a model-generated relative category from a visual-processing screen. At the moderate descending-alcohol stage, the group’s observed simulated speeding and mistake counts increased compared with baseline despite that unchanged label. The study did not observe real crashes.
Were there 16 alcohol-exposed drivers?
No. There were 16 completed participants in total: eight in the alcohol group and eight in the no-alcohol comparison group. Twenty originally volunteered; four were excluded after simulator sickness prevented completion.
Was the moderate alcohol stage directly measured in venous whole blood?
No such direct forensic whole-blood assay is documented as the study’s exposure measurement. The authors used a Dräger breath instrument and Widmark-guided stage timing while describing results in BAC-style ranges.
Did the study prove the UFOV screen has a specific false-negative rate?
No. There were no real crash outcomes, no predefined individual impairment gold standard, and insufficient data to estimate validated classification sensitivity, specificity, or individual false-negative probability.
Did every simulator outcome become worse as alcohol levels fell?
No. The marked increases concerned moderate-stage speeding and mistakes; some composite and handling subscores improved at later stages. Different endpoints and possible repeated-practice or expectancy effects require separate interpretation.
Can a BAC-style value below a general adult legal limit prove a novice driver was lawful?
No. Legal rules depend on jurisdiction, licence class, age, and date; graduated licensing may impose a zero-alcohol rule. This study was not a legal determination and its breath-based stage label was not a case-specific forensic blood result.
Can this pilot establish that alcohol caused an individual crash?
No. There were only simulator drives, no observed on-road collisions, no randomized matched placebo alcohol trial, and no individualized causal model. Any legal application requires independent case facts and appropriately validated evidence.
Original primary scientific source
Tremblay M, Gallant F, Lavallière M, Chiasson M, Silvey D, Behm D, Albert WJ, Johnson MJ. Driving Performance on the Descending Limb of Blood Alcohol Concentration (BAC) in Undergraduate Students: A Pilot Study. PLOS ONE. 2015;10(2):e0118348. Published February 27, 2015. https://doi.org/10.1371/journal.pone.0118348. This analysis reports only this original primary experiment’s stated methods, measured outcomes and limitations; it is not a diagnostic protocol, individualized expert opinion, or substitute for a full review of a case record.