How Many Days Off Does It Take to Recover From a Block of Night Shifts?

The design is unusually strong for a field study of rostering. Garde, Nabe-Nielsen, et al. (2020) had 73 non-smoking male Danish police officers each work three schedules in turn: 2 night shifts followed by 2 recovery days, then 4 and 4, then 7 and 7, with a seven-day night-free washout before each one and 26 data-collection days in total. Sleep was measured every day by wrist actigraphy alongside the Karolinska Sleep Diary, and 64 officers completed all three schedules. Because every officer served as his own comparison, stable differences between people are removed, but a within-person design removes neither order effects nor self-selection, so the schedule contrasts stay associations rather than experimental effects. The order of schedules was neither fixed nor randomized. Duty planners mixed it around staffing needs and officer preference, leaving 28% starting on 2+2, 27% on 4+4, and 45% on 7+7, so allocation shaped by who tolerates long night runs cannot be ruled out.

The design limit that matters most for this question is easy to state and easy to miss. Recovery days always equaled nights worked, so the study never tested two days off after seven nights, which is exactly the arrangement most rosters reach for. Anyone quoting these numbers to defend a short break after a long block is citing evidence that was never collected. It is also worth knowing what a recovery day was: day shifts counted as recovery days, though fewer than 9% of officers worked a day shift on the first one, and excluding that first day 38% of recovery days were daytime work days against 62% days off. Average awakening was 06:48 on recovery days with work and 07:25 on recovery days without.

The analysis set its significance threshold at P<0.001 to control for multiple testing, which is stricter than the usual convention. Several results that would pass a conventional test do not pass this one, including difficulty of awakening between night and recovery days (P=0.008), sleep quality (P=0.026), and sleep efficiency (P=0.040). Those should be read as nothing found, not as small effects.

Pooled across all three schedules, the gap between a night-shift day and a recovery day was large and consistent. Primary sleep averaged 5:07 after night shifts against 6:39 on recovery days, a difference of 1:32 (SD 0:04, P<0.001). Counting naps as well, total sleep time was 5:43 against 6:47, a difference of 1:04 (SD 0:04, P<0.001). Naps carried a real share of that load: 43% to 65% of participants napped on night-work days against 9% to 27% on recovery days, with a first nap averaging 55.2 minutes (SD 43.1). Only one participant never napped at all.

The finding that answers the title question is what happened across a run of days off. Sleep did not keep getting longer. Primary sleep duration rose by 0:02 per additional recovery day (SD 0:02, P=0.232), total sleep time by 0:00 (SD 0:02, P=0.712), and sleep efficiency by 0.01 (SD 0.14, P=0.950). The seventh day off recorded roughly the same amount of sleep as the first. The work side mirrors that shape. Primary sleep after a night shift lengthened by 0:03 per additional consecutive night (SD 0:01, P=0.038), which registers at a conventional threshold but not at the paper's stricter P<0.001 cut-off, and three minutes is nothing a roster would ever notice. That is why the authors conclude that sleep duration did not change with an increasing number of consecutive night shifts, and why the night-versus-recovery difference of 1:32 (SD 0:04, P<0.001) carries the real signal rather than any drift within a block.

How sleep felt was a different story. On the diary items, where the 1 to 5 scale runs from better to poorer, ratings improved with each additional recovery day. Disturbed sleep fell by 0.086 per day (SD 0.02, P<0.001), difficulty falling asleep by 0.083 (P<0.001), and non-refreshing sleep by 0.069 (P<0.001). Number of awakenings (-0.077, P=0.001), difficulty of awakening (-0.063, P=0.001), premature awakening (-0.060, P=0.003), and sleep quality (-0.057, P=0.003) moved the same way, sitting at or narrowly outside the paper's strict cut-off. Duration and quality came apart: the ratings tracked the number of days off while the minutes slept did not.

That leaves a deficit the study measured but never saw repaid. Garde, Nabe-Nielsen, et al. (2020) offered their own estimate, multiplying the number of consecutive nights by the 1:04 total-sleep-time gap and adding the extra 0:53 lost on the last night, which gives roughly 3:01 hours of cumulative sleep loss after two consecutive nights, 5:09 after four, and 8:21 after seven. Read that as the authors' arithmetic rather than an observed quantity, and note that recovery-day sleep duration never rose to pay it back inside the measured window.

The final night before days off stands apart from the nights around it, and the within-person design shows it by comparing the same ordinal night across schedules. Against the second night of the 7+7 schedule as reference, the last night of 2+2 was 1:01 shorter in primary sleep (SD 0:13, P<0.001) and 0:53 shorter in total sleep time (SD 0:14, P<0.001), with difficulty of awakening 0.58 higher (SD 0.13) and non-refreshing sleep 0.51 higher (SD 0.11), both P<0.001. The same comparison against the fourth night of 7+7 left the last night of 4+4 short by 0:51 in primary sleep (SD 0:10) and 0:53 in total sleep time (SD 0:11), with difficulty of awakening up 0.51 and non-refreshing sleep up 0.42. Every other diary item was unchanged.

The raw daily means show the same shape. In Jensen et al. (2022), total sleep time on the 2+2 nights ran 5:46 (SD 1:37) then 5:04 (SD 1:30); the 4+4 block ended at 5:08 (SD 1:18) after three nights near 5:55; and the 7+7 block held between 5:47 and 6:04 for six nights before dropping to 4:44 (SD 1:39) on the seventh, the shortest night-shift mean recorded in either paper. Within each condition sleep onset held steady, near 08:00 on night-shift days and near 23:45 on recovery days, so the shorter sleep on a block's last night came from waking earlier rather than from starting later, which is what you would expect from people cutting day sleep short to climb back onto a daytime rhythm.

This has a direct scheduling consequence that the papers describe without quantifying as a health outcome. The authors' own deficit formula charges every block one fixed extra loss of 0:53 on its final night, on top of 1:04 per night worked. A roster built from many short blocks therefore pays that fixed transition charge more often than one built from fewer, longer blocks. That is a genuine argument in the other direction from the usual advice to keep night runs short, and it should be weighed against the injury and health evidence rather than used to overturn it.

Two measures from the same project point in opposite directions, and honest use of this evidence means holding both. On the sleepiness side, Jensen et al. (2022) logged the Karolinska Sleepiness Scale, a 9-point scale on which higher means sleepier, every four hours on the last night shift and last recovery day. At the daily peak of 23:00 on the last recovery day, scores were 4.5 (SD 2.4) after 2+2, 4.9 (SD 2.1) after 4+4, and 5.1 (SD 2.3) after 7+7, with no significant difference between schedules (p=0.333). Nothing separated the schedules on the last night shift either (p=0.298), nor in the odds of a score of 7 or more (p=0.364 on night shifts, p=0.282 on recovery days). The numeric climb from 4.5 to 5.1 is not statistically distinguishable from no change and must not be described as a trend. The authors' reading is that recovery looked the same even after seven consecutive work days.

The counter-reading comes from the same data. Sleep duration on recovery days never increased across a block (P=0.232), so the estimated deficit of 3:01, 5:09, or 8:21 hours was not visibly repaid within the days measured. Sleepiness was recorded only on the last recovery day of each schedule, and the authors flag that limitation themselves as a possible reason no difference appeared, so the day-by-day shape of the sleepiness curve is unknown. Jensen et al. (2022) add that police work is high-arousal, and that "a high level of arousal masks the effects of sleep loss", which is a reason to expect a null sleepiness result here that would not transfer to quieter jobs. Sleepiness was in any case worse at work than at home, with scores of 7 or more on 22.8% of night-shift ratings against 14.2% of recovery-day ratings.

Two further limits travel with every number above. The sample was 73 experienced male non-smoking Danish police officers aged 25 to 62 (mean 38, SD 10), studied in 2013, and both papers explicitly warn against generalizing to women, with Jensen et al. (2022) noting that men and women rate sleepiness differently and that women report higher levels. Neither paper covers permanent night workers. Each schedule was also performed once, so only acute effects were captured and neither paper supports any inference about long-run scheduling. Jensen et al. (2022) close by noting that working many consecutive night shifts has been recognized as a long-run health hazard, which is the reason a null sleepiness result is not a safety endorsement of seven-night blocks.

The last caveat is about how much evidence this really is. Garde, Nabe-Nielsen, et al. (2020) and Jensen et al. (2022) are two outputs of a single project using the same 73 officers, the same 2013 data collection, and overlapping author teams, and the sleepiness paper takes its sleep-debt figures from the sleep paper. Their agreement is one study reported twice rather than corroboration from separate groups. A confident answer to how many days off a night block requires would need a second cohort, in a lower-arousal occupation, with recovery measured day by day and with the work-to-recovery ratio deliberately varied. No such study was found in this review.