This systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement and registered prospectively in the International Prospective Register of Systematic Reviews (PROSPERO: CRD420250650538). The research question used the PICOS strategy (P: adult shift workers (≥18 years) engaged in rotating, permanent night, or simulated shift schedules across any occupational sector; I: structured exercise; C: usual activity, wait-list, or non-exercise controls; O: sleep quality/quantity, circadian markers, and cognitive performance; S: RCT). The review addressed three core questions: (1) whether exercise improves sleep and cognitive outcomes, (2) which intervention characteristics maximize benefits, and (3) what mechanistic pathways and implementation barriers exist.

Randomized controlled trials (RCTs) of structured exercise interventions among adult shift workers (≥18 years, rotating or permanent non-day shifts; sector-specific or simulated) were included. Interventions could be any exercise method, delivered in any format or setting, provided that exercise was a principal component. Comparators were usual activity, wait-list, or non-exercise controls. Primary outcomes were sleep quality (subjective/objective), circadian alignment, and cognitive function. Studies were excluded if they (1) focused on non-shift workers, (2) used non-randomized or quasi-experimental designs, (3) were non-peer-reviewed, or (4) were not published in English.

A comprehensive search was conducted across PubMed, Scopus, Web of Science, MEDLINE, EMBASE, and Dimensions from inception to January 2025, using the keywords- shift work, exercise, sleep, circadian rhythms, and cognitive outcomes. Search strategies were devised in consultation with a medical librarian. The search terms keywords included: “shift work,” “shift worker,” “night shift,” “rotating shift,” and “day shift.,” “exercise,” “physical activity,” “exercise therapy,” “physical fitness,” and “motor activity,” “inspiratory muscle training,” “respiratory muscle training,” “breathing exercises,” “pulmonary rehabilitation,” and “respiratory muscle strengthening,” “sleep quality,” “sleep duration,” “cognitive function,” “cognitive performance,” and “neurocognitive outcomes.” (Details of the search strategy in each database is available as Supplementary file S1).

Two reviewers (FA & SA) independently screened titles/abstracts assessed full texts for eligibility and resolved disagreements by consensus. The following data were extracted in a standardized data extraction form.

Study characteristics: Author(s), year of publication, country, study design, setting (e.g., laboratory, workplace), sample size, population characteristics (e.g., age, sex, occupation, type of shift work).

Intervention details: Type of exercise, frequency, duration, intensity, supervision, timing relative to shifts.

Comparator details: Type of control intervention.

Outcome measures: Specific tools or methods used to assess sleep quality, cognitive function, and secondary outcomes, along with their respective units and time points of assessment.

Key findings: Quantitative data for all relevant outcomes (e.g., means, standard deviations, effect sizes, confidence intervals) for both intervention and control groups, at baseline and post-intervention. For studies reporting multiple time points, the longest follow-up data was prioritized. For studies reporting multiple outcomes for the same construct (e.g., different measures of sleep quality), the most clinically relevant or commonly reported measure was selected.

Risk of bias was evaluated using the Cochrane Risk of bias 2.0 tool, which examined five domains: randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selection of reported results, with each study classified as having low risk, some concerns, or high risk of bias. Methodological quality was additionally assessed using the PEDro scale, an 11-item checklist evaluating internal validity and statistical reporting, to facilitate comparison with prior exercise and rehabilitation reviews. We acknowledge that Cochrane discourages reliance on quality scales and summary scores, therefore PEDro results were interpreted descriptively alongside RoB 2.0, and not used to determine eligibility or to weight findings, noting that summary scoring may mask domain specific sources of bias. Both assessments were completed independently by two reviewers (FA & SA), and disagreements were resolved by consensus.

Meta-analysis was not conducted in our study due to statistical heterogeneity across studies in intervention type, outcome measures, and study populations. In addition, there were key differences in exercise timing (pre−/in−/post-shift), delivery context (e.g., laboratory vs. workplace), and underlying mechanisms assessed, such as circadian biomarkers. These differences made meaningful quantitative pooling inappropriate, since the resulting effect estimates would not represent comparable constructions. Instead, results were synthesized narratively, structured by outcome domain (sleep, circadian, cognitive), and further grouped according to (1) intervention modality, (2) timing relative to shift or circadian phase, and (3) participant or workplace characteristics, permitting a more valid interpretation of patterns and implementation relevance. Intervention characteristics were centrally reported and cross-compared, highlighting dose–response factors, session logistics, supervision level, and sector-specific barriers/facilitators necessary for clinical translation.

A total of 1,360 records were identified from database searches and other sources (see the PRISMA flow diagram in Figure 1). After removing duplicate articles, 561 records remained for screening. Of these, 52 full-text articles were assessed for eligibility, and 10 randomized controlled trials (RCTs) were included in the qualitative synthesis (30–39). All included studies were published from 1988 to 2024, with 60% published since 2020.

Table 1 provides an overview of included study characteristics. Across the 10 RCTs (N = 420 participants, mean sample size 42, range 18–75), 60% were conducted in healthcare settings and 70% of enrolled participants were female. Study designs consisted of eight parallel-group RCTs, one crossover RCT, and one cluster-RCT. Most studies enrolled real shift workers from hospital or industrial sectors, whereas two used simulated shift schedules. The majority of studies investigated rotating shift workers, with permanent night and simulated shifts each comprising 20% (Table 1 for detailed study and population characteristics).

Interventions comprised aerobic-only exercise (6/10), combined aerobic-resistance (2/10), and multimodal (3/10) approaches. Sessions ranged from single bouts to programs of up to 24 weeks. Moderate-intensity aerobic exercise (60%–75% HRmax) was the predominant intervention modality, with only one trial (34) evaluating a high-intensity interval training protocol. Training frequency ranged from two to six sessions per week, with durations generally between 10 and 60 min. About half of the interventions were delivered in the workplace (supervised or semi-supervised), and the remainder were home-based or in laboratory settings. Exercise timing relative to shifts included pre-shift, during-shift, or post-shift delivery (Details of the interventions in Table 1).

When reported, compliance was assessed via training diaries, attendance records, heart rate monitoring during sessions, and wearable-derived tracking data such as accelerometer readings. However, several trials did not provide detailed adherence metrics.

Sleep outcomes were assessed using subjective and objective measures (Table 2). Only one included trial employed polysomnography (PSG), the gold-standard method for objective assessment of sleep duration and architecture. In this study, Easton et al. (35) reported no significant changes in PSG-derived total sleep time or sleep architecture following the intervention, despite observing improvements in subjective alertness and reductions in sleepiness and fatigue.

Eight of the 10 studies (80%) reported significant improvements in at least one sleep outcome following exercise intervention. Subjective sleep quality (PSQI) improved in three studies, with reductions ranging from −2.1 to −4.6 points compared to controls. Four studies reported increases in total sleep time (TST) (mean TST increases 20–70 min) based on actigraphy. Sleep efficiency improved in two studies, and wake after sleep onset (WASO) was reduced in three. Two studies, (32, 35), reported no change in sleep duration or architecture despite cognitive improvements.

Heterogeneity in outcome tools was substantial: PSQI and related subjective scales were used in 3 studies; actigraphy in 5 studies (with differing analytic criteria); and polysomnography in 2 studies (assessing different sleep architecture indicators). This variability precluded quantitative pooling and estimation of effect size. Minimal clinically important differences (MCID) for these outcome measures were not defined for shift worker populations. Additionally, PSQI findings should be interpreted cautiously as the instrument was not designed for shift-specific sleep patterns and may not reflect objective sleep continuity or circadian alignment (see Table 2 for individual study sleep outcome data and direction of effect).

Three studies (32, 35, 37) evaluated cognitive performance as a secondary outcome (Table 3). Reported domains included alertness (visual analog scale, Karolinska Sleepiness Scale), reaction time (psychomotor vigilance task), and short-term memory (SAM-test). Two studies (32, 37) found improved alertness and faster reaction times in exercise groups compared to controls. Barger et al., (32) reported significantly faster reaction times (mean slowest 10%: 543.7 ms vs. 611.0 ms, p = 0.031) with no difference in TST. Easton et al. (35) found reduced fatigue and sleepiness in flexible chronotypes exposed to activity breaks, though no objective sleep improvements were noted.

All three studies used different cognitive batteries, limiting the comparability of results. Only alertness and reaction time domains showed consistent benefit; memory improvements were less consistently reported (see Table 3 for cognitive outcome details).

Six studies reported mechanistic or biomarker endpoints. Circadian markers (melatonin phase) were assessed in two studies (31, 32), which observed exercise-associated delays in melatonin onset (2–3 h) using protocol-specific timing. Autonomic modulation was measured in two studies (38, 39) using heart rate variability (HRV), which found increased parasympathetic activity post-intervention. Markers of systemic inflammation (CRP, IL-6, TNF-α, IL-1Ra) were evaluated in five studies, with three reporting significant reductions or anti-inflammatory shifts post-exercise.

Heterogeneity in chosen biomarkers, sampling times, and laboratory methods prevented cross-study synthesis (Mechanistic outcomes are detailed in Table 4).

Across the ten included studies, five (50%) were rated as some concerns (32–35, 39) and three were rated high for risk of bias (30, 36, 37), mainly due to lack of participant and therapist blinding and incomplete outcome data; two studies were rated as low risk (31, 38). The mean PEDro score was 6.7 (range 6–8), representing moderate methodological quality. Blinding of outcome assessors was inconsistently reported, and only three trials achieved high quality (PEDro ≥ 8/10). (Figure 2 and Table 5 for risk of bias and methodological quality details).

The Meta-analysis was not performed due to the high degree of heterogeneity in intervention protocols (modality, intensity, timing), delivery setting (lab, workplace, home), participant populations (rotating, permanent, simulated shift workers), and outcome measures (PSQI, actigraphy, PSG, VAS, PVT). This prevented effect size estimation and direct comparison between studies. All results are presented as a structured narrative synthesis.

This systematic review of 10 randomized controlled trials (n = 420) examined structured exercise interventions for improving sleep and cognitive outcomes in shift workers. Despite 80% of studies reporting sleep improvements, the clinical significance remains uncertain given: (1) heterogeneous outcome measures preventing effect size estimation, (2) predominance of short-term interventions (<12 weeks) in 70% of studies, and (3) unclear minimal clinically important differences for actigraphy parameters in shift workers. Three studies demonstrated cognitive benefits, particularly alertness and reaction time improvements, though comprehensive assessment was limited. Post-shift exercise most consistently improved sleep consolidation, while during-shift exercise produced circadian phase delays. However, two studies (32, 35) found null sleep findings despite improvements in alertness, highlighting mechanistic complexity. Although polysomnography provides the gold-standard assessment of sleep, the only PSG-based study did not demonstrate changes in objective sleep duration or architecture (35).

The two trials using objective sleep assessment reported improved alertness despite no detectable changes in sleep duration or architecture. Easton et al. (35), the only PSG-based trial, found no significant changes in PSG-derived total sleep time or sleep architecture, yet participants reported improved alertness. Similarly, Barger et al. (32) showed improved alertness and reaction time without actigraphic sleep changes, though short wavelength light exposure co-intervened. This dissociation suggests exercise acts primarily as an acute alertness countermeasure rather than altering sleep physiology. Expectancy effects may also contribute. Exercise timing appears critical, post shift exercise may support sleep consolidation, whereas during shift exercise, particularly with environmental manipulations like light exposure, improves alertness and safety without measurable sleep changes. Future trials should separate exercise from light exposure and pair objective sleep with circadian phase markers.

This suggests that observed benefits may relate to alertness or fatigue-related mechanisms rather than measurable changes in sleep physiology; however, conclusions are limited by the small number of PSG-based studies. These findings are consistent with broader evidence showing that targeted sleep-promoting behaviors, such as circadian-informed lighting, melatonin supplementation, blue light blocking, and daytime napping, can enhance reaction time and cognitive efficiency in sleep-restricted conditions (40–43). The evidence base remains limited by small sample sizes (mean n = 42), short follow-up periods, and 80% of trials rated as having “some concerns” or “high” risk of bias, precluding meta-analysis.

Prior systematic reviews have established that exercise is effective for improving sleep in general populations (17), with reporting small-to-moderate effect sizes for sleep quality (Hedges’ g = 0.36 for regular exercise) across 66 studies. Similarly, Alnawwar et al. (44) documented that moderate-intensity physical activity reduces sleep latency and increases total sleep time in adults with sleep disorders. However, these reviews did not specifically address shift workers or examine exercise as a circadian realignment strategy, an important distinction given that shift workers experience unique physiological challenges, including circadian misalignment, chronic sleep restriction, and occupational safety demands absent in general populations.

Our review addresses this evidence gap by synthesizing RCT-only evidence specific to shift workers, revealing that 8 of 10 studies (80%) reported positive sleep findings comparable to the 70%–75% response rates observed in general population reviews (17). However, three key differences were identified that limit direct comparisons.

First, shift workers may respond differently to exercise timing: our findings suggest post-shift exercise serves a transitional role facilitating sleep onset, whereas during-shift exercise acts as an alerting countermeasure with phase-shifting properties (31, 32). Timing considerations of this nature are largely unaddressed in day-worker studies.

Second, two studies with null sleep findings (32, 35) nonetheless demonstrated significant cognitive improvements (alertness, reaction time), suggesting exercise may benefit shift workers through pathways independent of sleep restoration. Barger et al. (32) found significant alertness improvements (p < 0.0001) and faster reaction times (543.7 vs. 611.0 ms, p = 0.031) despite no change in actigraphic sleep duration (6.7 vs. 6.9 h, ns), indicating exercise may enhance daytime function via direct neurophysiological mechanisms rather than sleep-mediated recovery. Similarly, Easton et al. (35) observed that breaking up sitting with light-intensity walking improved early-night alertness in flexible chronotypes despite no polysomnographic sleep improvements, potentially explained by acute arousal effects and individual circadian tolerance differences.

Third, adherence barriers are heightened in shift workers: fatigue, irregular schedules, and competing recovery needs reduce physical activity levels by 20% compared to day workers (27), complicating intervention sustainability.

Meta-analysis was deemed inappropriate due to substantial heterogeneity across multiple dimensions: (1) intervention timing (pre/during/post-shift) targeting different mechanisms (alertness vs. circadian realignment vs. sleep consolidation), (2) delivery context (laboratory-based efficacy studies vs. workplace feasibility studies measuring different constructs), (3) outcome measurement (PSQI, actigraphy, PSG each capturing distinct sleep dimensions), and (4) participant characteristics (simulated vs. operational shift work; rotating vs. permanent nights). Pooling these heterogeneous studies would produce misleading effect estimates, diminishing the visibility of clinically meaningful subgroup differences.

For clinical translation, this heterogeneity indicates that exercise is not a uniform intervention timing, modality, and individual factors (chronotype, shift pattern) critically moderate effectiveness, necessitating personalized prescription rather than one-size-fits-all recommendations.

The inability to conduct meta-analysis due to measurement heterogeneity prevents effect size quantification, limiting our ability to define clinically meaningful benchmarks. For example, does a 30-min increase in actigraphic total sleep time translate to reduced occupational errors in safety-critical industries? Without standardized outcomes and minimal clinically important difference (MCID) thresholds specific to shift workers, the practical significance of observed sleep improvements remains unclear.

Across the ten included randomized controlled trials, risk of bias assessment indicated that eight studies (80%) were rated “some concerns” or “high” risk for overall risk, primarily due to challenges with participant and therapist blinding (n = 8) and incomplete outcome data (n = 5). Only two studies achieved low risk of bias across all domains. Methodological quality, evaluated using the PEDro scale, was moderate overall, with a mean score of 6.7 out of 10 (range 6–8 across studies). These limitations are common in behavioral intervention research, but they diminish confidence in effect estimates and raise the potential for performance bias, especially for subjective sleep and cognitive outcomes. While objective measures, such as actigraphy and laboratory-based cognitive tests, help mitigate some bias, the inability to fully blind most exercise protocols and variable adherence reporting remain important barriers to drawing reliable, generalizable conclusions. Greater methodological standards in future research especially complete follow-up, improved objective measurement, and transparent reporting is needed to strengthen confidence in study findings.

Clinicians managing shift workers with sleep complaints may suggest but should not prescribe moderate-intensity aerobic exercise (2–3 sessions/week, 30–60 min) as a complementary strategy, recognizing that evidence remains preliminary. Timing appears critical: post-shift exercise may facilitate sleep transition, while pre-shift exercise may enhance on-shift alertness, though individualization by chronotype and shift pattern is likely necessary. Importantly, the null sleep findings in Barger et al., (32) and Easton et al., (35) alongside positive cognitive outcomes suggest exercise may benefit shift workers even without measurable sleep restoration a finding with potential relevance for workers unable to achieve adequate sleep duration due to operational constraints.

Employers should view these findings as preliminary evidence supporting workplace exercise program feasibility but should not mandate participation given the limited evidence base (only 10 heterogeneous RCTs, n = 420 total, 80% with bias concerns). Workplace-based interventions (50% of studies) achieved higher adherence than home-based programs, suggesting organizational support (e.g., on-site facilities, protected time) may be important facilitators, though formal cost-effectiveness analyses are absent.

Research needs include: