How Dangerous Is the Drive Home After a Night Shift?

Sixteen people who work regular night shifts, a mean of 3.1 nights per week, each drove an instrumented vehicle for two daytime sessions of two hours on a 0.8-km closed test track. One session followed a normal night of sleep, a mean of 7.6 hours (SD 2.4), and the other followed a real night shift, after a mean of 0.4 hours (SD 1.1) of sleep (Lee et al., 2016). Both drives started between 9:30 AM and 2:30 PM and were time-matched, so the same person, at the same time of day, in the same vehicle, provided both data points.

Eleven near-crashes occurred across 6 of the 16 postnight-shift drives, and none occurred in any of the 16 postsleep drives (Fisher's exact P = 0.0088). Every one of those near-crashes came with an emergency braking maneuver, so the 11 braking events and the 11 near-crashes are the same incidents counted once rather than two separate tallies. Seven of the 16 postnight-shift drives were also ended early for safety because the driver could no longer hold adequate control of the vehicle, six of those called by the investigator and one by the participant, against zero early terminations after sleep (Fisher's exact P = 0.0034).

The continuous measures moved in the same direction. Lane excursions ran at 3.09 per minute after a night shift against 1.49 per minute after sleep, a rate ratio of 2.08 (95% CI 1.98, 2.19; P < 0.0001), and on straight sections, where lane keeping asks least of the driver, at 0.59 against 0.19 per minute, a rate ratio of 3.11 (95% CI 2.72, 3.55; P < 0.0001). EEG microsleep episodes ran at 1.00 per hour against 0.47, a rate ratio of 2.09 (95% CI 1.05, 4.18; P = 0.0368), slow eye movements at 20.1 per hour against 10.6, a rate ratio of 1.89 (95% CI 1.63, 2.19; P < 0.0001), and scores on the Johns Drowsiness Scale at 1.71 against 0.97 (P = 0.0209).

Nothing critical happened at the start of any drive. Every emergency braking maneuver, near-crash, and early termination in the postnight-shift condition occurred 45 minutes or more after the drive began (Lee et al., 2016). A worker who feels alert pulling out of the lot is therefore not producing evidence that the trip is safe, because the measured failures showed up later in the trip than most people would think to check themselves.

Time behind the wheel mattered across all the drives, not only the postnight-shift ones. Microsleep risk, defined as more than three seconds of any sleep stage, rose after 30 minutes of driving, going from 0.13 to 1.31 microsleep episodes per hour (P = 0.04). Slow eye movements and lane-crossing events also rose as a function of driving duration (z = 8.41, P < 0.001), and that rise was steeper after a night shift than after sleep, shown by a condition-by-driving-block interaction (z = 4.07, P < 0.001).

For an operations leader, this puts commute length in view, but it does not mark out a safe distance. No critical event was recorded in the first 45 minutes of any drive, and 16 drives are far too few to establish that a short commute is safe: the continuous drowsiness measures were already running higher after a night shift well before the 45-minute mark, and every driver was stopped every 15 minutes for a sleepiness assessment, an alerting break that no real commute provides. What the data do support is that a long drive home runs deep into the period when drowsiness measures climbed fastest and when every serious incident in the experiment occurred, which makes commute distance a legitimate input to how a night roster is built rather than a threshold that clears anyone to drive.

The published paper contradicts itself on one measure, and anyone quoting it should know which figures hold. The abstract groups blink duration with measures reported at P < 0.05, while Table 2 of the same paper reports blink time duration as 0.18 seconds (SD 0.10) after a night shift against 0.13 seconds (SD 0.07) after sleep at P = 0.18, which is not significant. Time with eyes closed was also not significant, 0.36 percent (SD 0.13) against 0.32 percent (SD 0.12) at P = 0.21. The eye measure that did reach significance was interevent duration, 0.12 seconds (SD 0.04) against 0.10 seconds (SD 0.02) at P = 0.0192, which is a different variable. The unambiguous findings are the lane excursions, microsleep episodes, slow eye movements, drowsiness scores, and the near-crash and termination counts.

The track conditions cut in both directions, and the authors say as much. The loop had simple geometry, no navigation requirement, no rumble strip, and no other vehicles or pedestrians, which the authors argue means a real road would demand more and the study likely understates real impairment. Working the other way, drivers stopped every 15 minutes for sleepiness assessments, an alerting break that no real commute provides, and safety staff terminated 7 drives that on a public road would have continued to whatever came next.

The sample is 16 people with a mean age of 48.7 years (SD 14.8, range 19 to 65), a mean of 27.4 years of driving experience, and 5 of the 16 screening at risk of sleep apnea. The headline percentages of 37.5 and 43.8 rest on 6 and 7 individuals respectively, so they describe what happened in a small controlled study rather than a rate in any working population. Drive order was not randomized either, because each participant's own work schedule set it, and 12 of the 16 drove the postnight-shift session first. The authors tested for this and report no significant order effect in a post hoc analysis, but a test across 16 participants has little power to detect one, so an order or practice effect cannot be fully excluded. The work was supported in part by the Liberty Mutual Research Institute for Safety, whose test track was the study site.

A second line of evidence comes from a prospective nationwide survey of 2,737 US first-year medical residents who filed 17,003 monthly reports of their work hours, crashes, near-misses, and involuntary sleep episodes (Barger et al., 2005). Extended shifts of 24 hours or more were associated with 2.3 times the odds of a motor vehicle crash (95% CI 1.6 to 3.3) and 5.9 times the odds of a near-miss incident (95% CI 5.4 to 6.3), compared with non-extended shifts. Those are odds, not counts of crashes and not absolute risks, and the study is observational, so it reports an association rather than an effect.

The authors also described a dose pattern. Each extended shift scheduled in a month was associated with a 9.1 percent increase in the monthly risk of a motor vehicle crash (95% CI 3.4 to 14.7 percent) and a 16.2 percent increase in the monthly risk of a crash during the commute from work (95% CI 7.8 to 24.7 percent). In months with five or more extended shifts, the odds of falling asleep while driving were 2.39 times higher (95% CI 2.31 to 2.46) and the odds of falling asleep while stopped in traffic were 3.69 times higher (95% CI 3.60 to 3.77).

Three limits stop this from becoming a general night-shift number. The population is first-year residents surveyed in 2002 and 2003, before the duty-hour revisions of 2003, 2011, and 2017. An extended shift there means 24 hours or more, far longer than a standard 8 or 12-hour night, so the 2.3 odds ratio is not the risk of driving home from an ordinary night shift. Outcomes were self-reported, which carries recall and reporting bias, and the near-miss estimate has a notably tight confidence interval because near-misses were far more numerous than crashes.

The two studies should not be merged into one causal claim. One is a within-subject controlled experiment that permits causal language about degraded driving performance; the other is an observational cohort survey that permits only associational language. They also share a senior author and the same Brigham and Women's Hospital sleep medicine program, so their agreement is not the agreement of separate research groups working from different methods and samples.