Does Being On Call Cost You Sleep Even When Nobody Calls?

Hall et al. (2017) is the standard reference for on-call sleep. It is a systematic review of eight studies, all of them concerning on-call work performed from home, and all eight examined sleep. Only one looked at stress physiology, and the authors report that those data were not sufficient to draw meaningful conclusions. Their sleep verdict is deliberately hedged: working on-call from home appears to adversely affect sleep quantity, and in most cases, sleep quality.

The limitation matters more than the headline. The review states that the included studies did not differentiate between on-call nights from home with and without calls, so it cannot isolate the nights when nobody rang. It is also a narrative synthesis rather than a meta-analysis, which means there is no pooled estimate, no confidence interval, and no minutes-of-sleep-lost figure anywhere in it. Any specific number attached to this review in a slide deck did not come from the paper.

The oldest study on this page is a field recording, and the one that sits outside the Australian research program behind the rest of these citations. Torsvall and Akerstedt (1988) instrumented five male ships' engineers with EEG and ECG alongside subjective ratings. Sleep during on-call nights, which included two alarms, was shortened and contained less slow wave sleep and less rapid eye movement sleep, with lower spectral power density and a higher heart rate. Rated sleep quality was lower and sleepiness was higher during the following day. The line that carries the whole topic is the authors' observation that many of the effects were observable before any alarms had occurred, which they attributed to apprehension induced by the prospect of being woken.

The laboratory work removes the call entirely. Across three counterbalanced four-night protocols run at the Appleton Institute sleep laboratory in Adelaide between January 2016 and June 2017, 70 male participants aged 20 to 35 slept in a time-isolated facility with lights out at 2300 h and waking at 0700 h. The methods are explicit that despite being on-call on relevant nights, participants were only ever called at the end of their eight-hour sleep opportunity (Sprajcer et al., 2020). What that design establishes is narrow but useful: nothing woke anyone during the night, so any difference between those nights belongs to anticipation rather than to disturbance. The protocol ran an adaptation night, a control night and two counterbalanced on-call nights, and the contrast reported below is the one between the two on-call conditions.

Read the sample sizes carefully before quoting them. The pooled paper gives participant counts of 24, 22 and 24 for the three studies in its methods, then states elsewhere in the same section that the number was 24 for each, an internal inconsistency in the published article (Sprajcer et al., 2020). Either way the arithmetic is small: roughly two dozen healthy young men per condition, none of them women, older workers, or people already carrying a sleep problem.

One of those laboratory studies manipulated how likely a call was. Participants spent an adaptation night, a control night and two counterbalanced on-call nights, told on one of the on-call nights that they would definitely be called and on the other that they may be called. Neither on-call night was interrupted, because calls came only at the end of the sleep opportunity. Compared with the night when a call was certain, the uncertain night brought significantly less non-rapid eye movement sleep and stage two sleep, significantly more rapid eye movement sleep, and poorer cognitive performance the next day, even though participants felt more anxious before bed when the call was certain (Sprajcer et al., 2018).

That dissociation is the practical finding, and the authors state it plainly: self-reported anxiety may not be linked with sleep outcomes while on-call. The pooled reanalysis points the same way. Trait anxiety did not differ across the three studies, F(2, 143) = 1.63, p = 0.200, while pre-bed state anxiety differed sharply by what was being anticipated, F(2, 212) = 45.44, p < 0.001, and was highest at 40.8 in the study that anticipated a stressful task rather than an uncertain call, against 33.6 for call likelihood and 32.4 for the chance of missing the call (Sprajcer et al., 2020). Feelings and sleep are not tracking each other here.

The laboratory studies are counterbalanced within-participant experiments, so they support causal statements about the anticipation manipulation, but only in a sleep laboratory, in screened healthy young men, at sample sizes near two dozen. What the contrast reported here licenses is a statement about what changes when expectation changes, tested in that setting and that sample. The field evidence is five engineers recorded in 1988 with the equipment of that era, and the review above it is observational and cannot separate call nights from quiet ones. The review, the laboratory program and the management review below all come from one Australian research program with overlapping authors rather than independent replications, with the Karolinska field study as the separate line of evidence.

On-call exposure is wider than a named rota. In a cross-sectional online survey of 2,044 Australian adults run in March and April 2019, 45.5% of the 956 working respondents analyzed reported on-call work in the preceding month (Sprajcer et al., 2021). It reached 63.1% among 18 to 24 year olds, 62.6% among people holding multiple jobs against 42.8% for single-job holders, and 65.2% in accommodation and food services, 61.4% in manufacturing, and 56.3% in both agriculture, forestry and fishing and retail trade. These are self-reported figures from one country at one point in time, so use them to scope the question rather than as a global rate.

The management side is thinner than the volume of guidance suggests. A gray literature search produced 65 relevant guidance materials, and thematic analysis found eight consistent strategies: regularity and predictability of shifts, a fatigue management policy, prescriptive rule sets, fitness for work assessment, on-the-day control measures, risk assessment, training and education, and call management. A parallel systematic review identified 17 original studies on effectiveness (Bumpstead et al., 2025). The authors conclude that very little research exists outside some prescriptive hours of work limitations, that those limitations generally reduced fatigue but often had the unintended consequence of increasing workload in ways that may inadvertently increase overall risk, and that training, education and call management such as protected naps during on-call periods also had some supporting evidence.