The present study was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) 2020 statement (17) (Appendix 1).

Six electronic databases—EMBASE, CINAHL, Cochrane Library, PubMed, Scopus, and Web of Science—were searched for relevant studies. The search keywords included “shift work,” “night work,” “rotating shift,” “sleep,” “insomnia,” “sleep disturbance,” “sleep deprivation,” “sleep problem,” “intervention,” “treatment,” “therapy,” “counseling,” “program,” “CBT,” and/or “self-help.” Appropriate subject headings for each database (i.e., Medical Subject Headings and CINAHL headings) and free-text terms were logically combined using Boolean operators, such as AND, OR, and truncation (Appendix 2). All studies retrieved from database searches were exported to the EndNote X9 citation management software.

The inclusion criteria were as follows: (a) participants were adults (≥18 years) working in rotating night shifts, which are scheduled shifts that change over time including night-shift (00:00–05:00), (b) interventions aimed to promote sleep or improve sleep disturbances among rotating night shift workers, (c) comparators were non-intervention or any other interventions, (d) outcomes included sleep patterns, (e) study designs were randomized controlled trial (RCT) or clinical trial, (f) studies were published between January 1990 and June 2022, and (g) the full texts of studies were published in the English or Korean language. Studies were excluded in case of the following: (a) the intervention was implemented in simulated work environments and (b) the participants were non-shift workers.

All potentially eligible studies were combined and screened for duplicates. After removing the duplicates, the investigator independently screened the titles and abstracts of each study to determine its eligibility. Three authors assessed the full text of all studies that potentially met the inclusion criteria. Any reasons for exclusion were documented; disagreements were resolved by discussion. The flow diagram of the study is presented in Figure 1.

The quality of the included studies was assessed by each investigator independently using the Joanna Briggs Institute Critical Appraisal Checklist for RCTs (13 items) and quasi-experimental studies (9 items) (Appendix 3) (18). The criteria for quality assessment included the following: random sequence generation; allocation concealment; blinding of the participants, personnel, and outcome assessors; incomplete outcome data; selective reporting; and other sources of bias. Overall quality ratings were graded using the categories cited by Reilly et al. (19) (good = at least 80%, moderate = 50%−80%, poor = < 50%).

For performing the systematic review, two investigators independently evaluated each study using a data extraction form that comprised the study objective, design, participant characteristics, intervention types, follow-up duration, and outcomes (Appendix 4). The interventions were grouped into similar types, which were finally categorized into five types of interventions: pharmacological approach, light therapy, cognitive behavioral approach, aroma or alternative therapy, and shift schedule modification (Table 1).

For conducting the meta-analysis, the effectiveness of the intervention was evaluated using the outcome variable of sleep quality; this overarching outcome was calculated by averaging the effect sizes of diverse outcomes associated with sleep reported in each study, such as the total sleep duration, sleep efficiency, subjective sleep quality, and fatigue. These diverse variables were not conceptually different, thereby enabling the use of the average effect sizes within studies (48). Averaging the effect sizes of diverse outcomes in a study precludes the possibility of violating the assumption of data independence and producing imprecise standard errors (SE) and confidence intervals (CIs) from including multiple outcomes of the same participants in the meta-analysis (48).

Studies reporting sufficient statistical data for pooling the effect size calculations were included in the meta-analysis. For the studies measuring sleep quality using more than one method, such as with a diary and questionnaire, the values measured by the questionnaire were used for the meta-analysis. Because the present study aimed to identify the summary effect size of each study, for studies with more than one intervention arm, the sample size of the shared control group was divided by the number of intervention arms to avoid duplicate counting of the participants (49, 50). When the scoring of the scale was not in the same direction, the scores were converted in one direction before synthesis.

In this study, the effect sizes were calculated with Hedges' g and 95% CI using the mean values and standard deviations (SDs) of the intervention and control groups reported in the studies. If SDs were not provided, the values were derived using other information (i.e., p, t, or F statistics). Hedges' g was chosen to reduce the possibility of overestimating the effect size in very small sample sizes (51). Considering the diversity of research methods, samples, intervention types, and outcome measures in each study, a random effects model was used to calculate the summary effects and 95% CI (52). Hedges' g can be interpreted as small (0.2), medium (0.5), and large (0.8); a p-value of < 0.05 indicated a statistical significance. In the present study, a positive Hedges' g indicated that sleep interventions influence the improvement of sleep outcomes.

To investigate the effectiveness of the interventions according to intervention types, preplanned subgroup analysis was used after examining the existence of significant heterogeneity between studies. A subgroup analysis was conducted only in case of a minimum of two studies per subgroup. Heterogeneity between studies was evaluated using the forest plot of the visual test and Higgins I² homogeneity test of the quantitative test. I² of approximately 25%, 50%, and 75% were interpreted as low, medium, and high heterogeneity, respectively (53).

Publication bias of the included studies was examined using the funnel plot, Egger's test, and Duval and Tweedie's trim-and-fill method. Moreover, sensitivity analysis was performed to evaluate the robustness of the synthesized results. The Comprehensive Meta-Analysis (CMA 3.0) software was used for performing the meta-analysis.

A total of 1,972 workers participated in the included studies (Appendix 4). The studies were conducted in 13 countries, including seven European countries, four Asia-Pacific countries, and two North American countries (Appendix 3). Studies with RCT (n = 20) and quasi-experimental (n = 10) designs were conducted mostly on healthcare workers (n = 20, 60.6%). To measure the outcomes, the majority of the studies used subjective tools such as the Pittsburgh Sleep Quality Index (PSQI) and sleep diary (n = 20). Other studies utilized actigraphy (n = 2) or a combination of objective tools (i.e., polysomnography, actigraphy, and SOMNOwatch) and subjective tools.

Quality assessment in individual studies is summarized in Appendix 3. Among 10 quasi-experimental studies, four studies were determined to be of good quality (31, 34, 35, 46), and the remaining were of moderate quality (20, 22, 23, 27, 39, 45). Among 20 RCTs, only two studies had good quality (38, 44) and 11 studies had moderate quality (12, 14, 21, 24–26, 29, 30, 33, 42); seven studies had poor quality (28, 32, 36, 37, 40, 41, 47) because of allocation concealment (28, 32, 36, 37, 41, 47), blinding of the participants and researchers (28, 32, 36, 37, 40, 41), and insufficient reporting (28, 32, 36, 37, 40, 41, 47).

The interventions to promote sleep among rotating night shift workers were classified into five types: pharmacological approach (n = 7), light therapy (n = 9), cognitive behavioral approach (n = 7), aroma or alternative therapy (n = 4), and shift schedule modification (n = 3, Table 1, Appendix 3).

First, the pharmacological approach was evaluated in seven studies involving healthcare workers (n = 150), industrial or manufacturing workers (n = 50), and other occupational workers (n = 34, Table 1) (22, 26, 30, 36, 42, 43, 45). With respect to the intervention, melatonin (26, 30, 42, 43, 45), benzodiazepine (22) or non-benzodiazepine (22) class, or Gingko biloba (36) was administered as a sleep aid for main sleep (n = 7).

Second, light therapy was evaluated in nine studies involving healthcare workers (n = 242) and industrial or manufacturing workers (n = 37) (14, 20, 21, 23, 28, 29, 35, 41, 46). Interventions include intermittent bright light therapy (14, 20, 23, 28), a combination of bright light and wearing sunglasses (29, 46), wearing glasses fitted with short-wavelength filters (41), comparison of bright light and melatonin (21), and use of an eye shield (35). The intensity of the light interventions ranged from 1,500 to 10,000 lux. The individual exposure duration ranged from 10 min to 10 h; the intervention duration ranged from 4 days to 3 months. The effect of light therapy on sleep enhancement was deemed favorable in all nine studies. The glasses fitted with short-wavelength filters worn for 8 weeks demonstrated a significant improvement in sleep quality and quantity (41).

Third, a cognitive behavioral approach was implemented in seven studies involving healthcare workers (n = 388) and other occupational workers (n = 554) (12, 25, 32, 34, 38, 40, 44). The interventions included educational programs consisting of sleep physiology, sleep hygiene, and fatigue-relieving and sleep-promoting strategies (n = 4) (12, 32, 40, 44), cognitive behavioral interventions (n = 2) (25, 34), and exercise interventions (38). The interventions were delivered through face-to-face sessions (25, 32, 40, 44) and mobile technology (12, 34). The mean duration of the interventions was 9 (range, 3–24) weeks. Further, 8 weeks of aerobic exercise showed a significantly positive lasting effect on the total sleep duration measured using actigraphy in 60 female nurses (38).

Fourth, aromatherapy or alternative therapy was evaluated in four studies (24, 27, 33, 47). In two studies evaluating aromatherapy for nurses (24, 33), the interventions using Lavandula angustifolia with the inhalation method (33) and Origanum majorana with massage therapy (24) showed positive effects on sleep quality. In two other studies of alternative therapy (27, 47), transcutaneous electrical acupoint stimulation (TEAS) on acupuncture points SP6, H7, and LI4 (47) and non-alcoholic beer provided during dinner (27) for nurses positively influenced sleep quality.

Fifth, shift schedule modification was implemented in three studies involving healthcare workers (n = 62) and industrial or manufacturing workers (n = 213) (31, 37, 39) (Table 1). Interventions involved the modification of the direction of shift rotation (n = 2) (31, 37) and shift intervals (n = 1) (39).

Overall, 25 studies were included in the meta-analysis after excluding five studies with insufficient statistical data for pooled effect size calculations (23, 27, 28, 39, 44). The random effect model demonstrated that all intervention types reported a significant effect on the pooled sleep quality, which indicated a moderate effect size (Hedges' g = 0.57, 95% CI = 0.34–0.79, p < 0.001). Because heterogeneity was high (I² = 80.9%, Q = 141.26, p < 0.001), a subgroup analysis was performed according to the intervention types. The results of the meta-analysis are shown in Table 2 and Figure 2.

In the subgroup analysis, light therapy had a large effect size on the pooled sleep quality (Hedges' g = 0.86, 95% CI = 0.39–1.33). Moreover, the cognitive behavioral approach showed significant effects on sleep quality (Hedges' g = 0.60, 95% CI = 0.13–1.16). However, other interventions—namely, pharmacological approach, shift schedule modification, and aroma or alternative therapy—did not significantly affect the pooled sleep quality (Table 2).

We performed a test of publication bias to identify the validity of the results of the meta-analysis. The funnel plot showed data symmetry. Egger's test showed no significant publication bias (p = 0.072) (Figure 3); further, the trim-and-fill analysis showed no change in the effect size. In addition, the sensitivity analysis showed no change in the pooled effect size after excluding the largest weighted study (Hedges' g = 0.57, 95% CI = 0.34–0.79, p < 0.001) (29). Therefore, the overall effect of these interventions on the combined effects was robust.