Does Caffeine Reduce Night-Shift Errors, and When Should You Stop?

Ker et al. (2010) set out to establish whether caffeine prevents injuries and errors in shift workers. Thirteen randomized trials met the inclusion criteria and ten contributed data that could be pooled, but none of them collected outcome data on the occurrence of injuries. The review reports what caffeine did to measured performance and to error counts on tasks, and states that there are no trials from which an effect on injuries can be assessed.

That gap is not a technicality. The authors accept that higher performance levels are plausibly associated with fewer injuries, while refusing to put a number on how large any such reduction would be. A fatigue policy should copy that discipline: caffeine has trial evidence behind alertness, and no trial evidence behind accident or incident rates.

The exposure conditions matter too. Ten of the thirteen trials induced circadian disruption in a sleep laboratory rather than following people through real rosters, and only three ran in the field: an eastward flight crossing seven time zones, night-shift work, and night-time driving. Per-trial samples ran from 6 to 68 participants, the average age in many trials sat between 20 and 30, and no participant had a confirmed diagnosis of shift work disorder, so the authors call generalization to older workers and to real shift work unclear.

The trials delivered caffeine in different forms. Eight used a tablet or capsule, three of them a single 200 mg dose and three a single 300 mg dose, with one at 4 mg/kg and one dosing hourly at 0.3 mg/kg. Two used coffee at 200 mg and at 2 mg/kg, two used caffeinated food at 200 mg once or twice, and one used a caffeinated energy drink at 2 x 80 mg.

Against placebo the pooled effects were significant in two cognitive domains and not in three. Orientation and attention improved by a standardized mean difference of -0.55 (95% CI -0.83 to -0.27, P <= 0.0001) across six trials and 22 outcome variables, and memory improved by SMD -1.08 (95% CI -2.07 to -0.09, P = 0.03) across two trials, where a negative value favors caffeine. Perception (SMD -0.77, 95% CI -1.73 to 0.20, P = 0.12) and concept formation and reasoning (SMD -0.41, 95% CI -1.04 to 0.23, P = 0.21) had confidence intervals crossing zero, and the single verbal functioning trial pointed the other way at SMD 0.18 (95% CI -0.50 to 0.87, P = 0.60).

Two trials measured errors directly. In real night-time driving, caffeine lowered the odds of an inappropriate line crossing to OR 0.14 (95% CI 0.05 to 0.41, P = 0.0003), which is the odds of an error event rather than a count of crashes. In a flight simulator, deviation from the altitude envelope improved by SMD -1.30 (95% CI -1.93 to -0.68, P < 0.0001) and deviation from the velocity envelope by SMD -1.18 (95% CI -1.79 to -0.56, P = 0.0002).

Read all of it with the review's own caution. Only 4 of 13 trials were at low risk of bias for allocation concealment and only 2 of 13 for incomplete outcome data, no trial could be checked against a protocol for selective reporting, and selective reporting was judged to affect 5 trials, 3 of which clearly presented only statistically significant results. All 13 were at low risk for blinding. The authors warn that selective reporting possibly leads to over-estimated intervention effects, and the review is old: most databases were searched only to July 2008, with PubMed to April 2010, and no updated version has appeared since.

The head-to-head comparisons are more mixed than the usual summary allows. On cognitive outcomes caffeine did not differ significantly from a nap in any domain, including orientation and attention (SMD -0.14, 95% CI -0.72 to 0.45, P = 0.65) and memory (SMD -0.80, 95% CI -2.00 to 0.40, P = 0.19). On the error outcome it did differ: night driving errors favored caffeine over a nap at OR 0.71 (95% CI 0.56 to 0.90, P = 0.005).

Against modafinil, caffeine was no different on altitude deviation (SMD -0.25, 95% CI -0.81 to 0.32, P = 0.40) and significantly worse on velocity deviation (SMD 0.60, 95% CI 0.02 to 1.18, P = 0.04). Caffeine combined with a nap beat placebo on verbal functioning (SMD -0.77, 95% CI -1.48 to -0.06, P = 0.03) but not on orientation and attention (P = 0.11), and caffeine added to dim light showed nothing significant on memory (P = 0.35) or on orientation and attention (P = 0.75). Each of these rests on one or two trials, so an absent difference is weak evidence of equivalence rather than proof that the options are interchangeable.

The other half of the question is when to stop. Drake et al. (2013) gave 12 healthy adults 400 mg of caffeine in pill form at either 6, 3, or 0 hours before bedtime, with matching placebos at the other two times and one all-placebo night, in a double-blind, double-dummy crossover. Monitored total sleep time fell from 7.68 hours on placebo to 6.50 hours when caffeine was taken six hours before bed, a loss of 1.18 hours or roughly 71 minutes, with F(3,33) = 3.43, P = 0.03 across conditions.

The result that should change practice is the awareness gap. At bedtime and at three hours before bed, the sleep diary registered the disruption alongside the monitor. At six hours only the objective measure did, which the authors flag as a likely reason people ignore advice about afternoon caffeine. Slow wave sleep was also lower at the six-hour timing (48.91 minutes against 71.45 on placebo, F = 4.26, P = 0.01), while REM was unaffected at every timing (F = 0.30, P = 0.83).

Take the limits seriously before turning this into a rule. The sample was 12 healthy day workers with normal sleep and moderate habitual intake of 115 mg per day, each condition ran for a single night, sleep was recorded by a single-channel EEG headband rather than polysomnography, and one 400 mg dose is about four home-brewed cups by the paper's own conversion. Participants slept at night, so applying the finding to a night worker means counting back from a morning sleep period, which the trial did not test.

The funding deserves a note. The study was supported by an investigator-initiated grant from Zeo Inc., the maker of the headband that produced every objective number, and one co-author was a Zeo vice president until 2010 and reports no current financial interest. The Cochrane review, by contrast, declares no known conflicts and was supported by UK public research funding. Drake et al. (2013) recommend refraining from substantial caffeine for at least six hours before bedtime and keeping afternoon caffeine to before 17:00, while noting that published cutoff advice ranges widely from 4 to 11 hours.