Non-Pharmacological Interventions to Improve Chronic Disease Risk Factors and Sleep in Shift Workers: A Systematic Review and Meta-Analysis

Shift work is associated with adverse chronic health outcomes. Addressing chronic disease risk factors including biomedical risk factors, behavioural risk factors, as well as sleep and perceived health status, affords an opportunity to improve health outcomes in shift workers. The present study aimed to conduct a systematic review, qualitative synthesis, and meta-analysis of non-pharmacological interventions targeting chronic disease risk factors, including sleep, in shift workers. A total of 8465 records were retrieved; 65 publications were eligible for inclusion in qualitative analysis. Random-effects meta-analysis were conducted for eight eligible health outcomes, including a total of thirty-nine studies. Interventions resulted in increased objective sleep duration (Hedges’ g = 0.73; CI: 0.36, 1.10, k = 16), improved objective sleep efficiency (Hedges’ g = 0.48; CI: 0.20, 0.76, k = 10) and a small increase in both subjective sleep duration (Hedges’ g = 0.11; CI: −0.04, 0.27, k = 19) and sleep quality (Hedges’ g = 0.11; CI: −0.11, 0.33, k = 21). Interventions also improved perceived health status (Hedges’ g = 0.20; CI: −0.05, 0.46, k = 8), decreased systolic (Hedges’ g = 0.26; CI: −0.54, 0.02, k = 7) and diastolic (Hedges’ g = 0.06; CI: −0.23, 0.36, k = 7) blood pressure, and reduced body mass index (Hedges’ g = −0.04; CI: −0.37, 0.29, k = 9). The current study suggests interventions may improve chronic disease risk factors and sleep in shift workers; however, this could only be objectively assessed for a limited number of risk factor endpoints. Future interventions could explore the impact of non-pharmacological interventions on a broader range of chronic disease risk factors to better characterise targets for improved health outcomes in shift workers.

Note. Adapted with permission from AIHW material [31]. Shaded square with dot, indicates a direct association between risk factor and chronic disease. Non-shaded squares indicate insufficient evidence of direct relationship; however, an indirect relationship may exist.
In addition to the risk factors outlined in Table 1, sleep loss and poor sleep quality are also associated with an increased risk of chronic diseases including diabetes [32][33][34], obesity [32][33][34], metabolic syndrome [32,33], cardiovascular disease [32,[35][36][37], chronic kidney disease [38] and osteoporosis [39]. Beyond this, perceived health status has some capacity to identify individuals at risk of poorer future health outcomes [40][41][42]. Together, chronic disease risk factors, including biomedical and behavioural risk factors, sleep and perceived health status provide insight into future chronic disease risk, and by extension, areas for potential management prior to disease onset.
Given the increased risk of chronic disease associated with shift work, an understanding of beneficial interventions aimed at improving chronic disease risk factors in these employees is critical. To date, various interventions have focused on improving health and safety outcomes in shift workers. A previous literature review by Neil-Sztramko et al. [43] qualitatively reviewed 44 studies that utilised pharmacological and non-pharmacological interventions to improve indicators of chronic health effects in rotating and night shift workers. This review found that non-pharmacological interventions such as fast forwardrotating shifts, timed use of bright light and light-blocking glasses, and targeting health behaviours including physical activity and diet yielded favourable outcomes for shift workers. Pharmacological interventions were largely not efficacious. These findings suggest that non-pharmacological interventions should be implemented to improve chronic disease risk factors for shift workers. However, the review [43] did not quantitatively examine the effect of specific interventions on chronic disease risk factors, or on sleep outcomes. Such information is crucial for identifying and managing early risk before disease negatively impacts worker quality of life.

Study Selection
Titles and abstracts were screened against eligibility criteria and relevance to the review. Publications that were thought to meet the criteria were read in full by two authors (M.E.C. and A.C.R.) to determine eligibility. All papers were independently reviewed by both authors for eligibility and any discrepancies discussed. It was agreed that any discrepancies that could not be resolved by discussion would be adjudicated by a third researcher (S.A.F.); however, this situation did not arise.

Data Collection Process
Data were extracted from the articles using a pre-defined modified Cochrane data extraction sheet [50] by M.E.C. The variables extracted from each included study were: first author name, publication date, intervention design, participant characteristics (sex, age), study design, occupation, type of shift work schedule, duration of intervention, outcome measure of interest and results of studies. Where further details or clarification were needed for meta-analysis, authors were contacted wherever possible.

Risk of Bias in Individual Studies
Quality assessment and risk of bias assessment were conducted using the Downs and Black tool 1998 [51]. This tool was chosen as it can be used with both randomised and nonrandomised studies, both included within this review. Further, the Downs and Black [51] tool has also been used for quality assessment in previous systematic reviews in shiftworking populations [8] and critical reviews of interventions targeting shift workers [43]. Item 27 was modified from "Did the study have sufficient power to detect a clinically important effect where the probability value for a difference being due to chance is less than 5%?" to "Was a power analysis performed?" 0 = no or unable to determine, 1 = yes, as in previous studies [52]. This response structure aligns with the response style for the Downs and Black tool. All studies were evaluated individually by the first author and a random sample of~20% (n = 12) was evaluated by the senior author. Comparisons between the scores of the two investigators were then compared and an agreement percentage calculated to ensure quality of evaluations.

Summary Measures
The principal summary measure used throughout the review was standardised mean difference wherever possible. Further, studies that reported statistical significance were reported qualitatively in this way.

Synthesis of Results
Meta-analyses were conducted if there were at least three eligible studies reporting an outcome of interest [53]. Further, studies included for meta-analysis were those that used similar methods for measuring the outcome of interest in order to allow for meaningful interpretation of meta-analytic results (e.g., alcohol consumption grams/per day and "alcohol risk scores" were not considered sufficiently similar measures).
For studies where insufficient information was reported in the published manuscript, two attempts were made to contact the corresponding authors, where contact details were available, between 20 July 2020 and 10 August 2020. Following attempts to contact authors, every effort was made to calculate missing data, in accordance with Cochrane guidelines [54]. For studies that only reported a p-value, mean difference was used to obtain a t-value, which was then converted to standard error, and standard deviation. Studies which presented means within figures and were of sufficient quality were processed through Plot Digitizer [55] in order to allow means and standard deviations to be extracted from figures.
Applying these approaches for calculating missing information resulted in the inclusion of more studies and thus minimised possible biased meta-analytic estimates resulting from missing data [56]. A detailed outline of statistical procedures used for each study is available upon request.
Where studies were a controlled trial, the effect size for the difference between the control and intervention group was calculated, including studies between worksites (e.g., firefighters at different stations). However, studies that utilised a control group that was not in the same occupation (e.g., participants worked in two different types of manufacturing) were assessed as within group pre-post intervention only, to avoid confounding from occupational effects. Where the study design was within groups, the effect size was calculated based on the change from baseline to follow-up period. Where there were insufficient data reported from a control group, only the intervention group data were included for meta-analysis (e.g., baseline to follow up) to allow for calculation of standard deviation.
In studies where two intervention groups were present, with one common control group, the common control group was divided by two and half the control group sample was used in the meta-analysis for each of intervention groups [54].
To allow for direct comparison of sleep duration across as many studies as possible and to allow for a clear interpretation of the effect of intervention on overall sleep duration, the decision was made to use average sleep duration effect size in meta-analysis. For calculation of total average sleep duration, effect sizes of all reported sleep durations were calculated and then combined using the Borenstein formula [57] using the MAd package [58] in R 3.6.2 [59]. Thus, a study that reported sleep duration after night shift, morning shift and afternoon shift was combined to provide one overall estimate of total sleep duration. In addition, studies that presented outcomes by groups which were not directly related to the intervention (e.g., age or marital status) were also combined, using the method described above, to allow for the accurate representation of the entire study and avoid attributing excessive weight to studies by including outcomes independently. Finally, in studies that measured variables on an inverse scale (e.g., higher score = worse sleep), data were transformed by subtracting the mean value for the total possible scale score to ensure the direction of scale was consistent with other studies [54].
Eight separate meta-analyses were performed. Meta-analyses were conducted using packages dmetar [60] and metafor [61]. Hedges' g for each for study was calculated using the esc [62] in order to minimise the effect of uneven sample sizes [63]. Random-effects models were conducted using the Sidik-Jonkman estimator for heterogeneity with the Hartung-Knapp-Sidik-Jonkman adjustment. These methods were chosen to limit error rates due to expected heterogeneity in studies [64]. Heterogeneity between studies was calculated using Tau 2 and I 2 statistics, with interpretation I 2 as~25% = low,~50% = moderate, and~75% high [65]. Further, to provide additional information about the heterogeneity between studies the prediction interval of studies was also reported [66]. Publication bias was assessed using Eggers' intercept test and visual analysis of funnel plots [67].

Additional Analyses
Pre-specified subgroup analyses of the effect on intervention type on outcome were conducted. Subgroups were categorised according to intervention type and analysed through random-effects models to assess effect size by intervention and the statistical significance of differences.

Study Selection
The database searches identified 8465 papers. A further 54 papers were identified through hand searching of reference lists and grey literature searches, including using Google Scholar. A PRISMA flowchart for selection of studies, inclusive of updated searches, is depicted in Figure 1. Following duplicate removal, and title and abstract screening, a total of 178 papers remained for full-text review. Full-text reviews were conducted independently by two authors (M.E.C. and A.C.R.). From this review, a further 113 papers were excluded, largely as they did not meet criteria, including absence of a chronic disease risk factor, sleep or perceived health status outcome (n = 38), did not involve an intervention (n = 20) or the full text was not available (n = 18). The total number of studies eligible for inclusion in this review was 65.

Study Selection
The database searches identified 8465 papers. A further 54 papers were identified through hand searching of reference lists and grey literature searches, including using Google Scholar. A PRISMA flowchart for selection of studies, inclusive of updated searches, is depicted in Figure 1. Following duplicate removal, and title and abstract screening, a total of 178 papers remained for full-text review. Full-text reviews were conducted independently by two authors (M.E.C. and A.C.R.). From this review, a further 113 papers were excluded, largely as they did not meet criteria, including absence of a chronic disease risk factor, sleep or perceived health status outcome (n = 38), did not involve an intervention (n = 20) or the full text was not available (n = 18). The total number of studies eligible for inclusion in this review was 65.

Study Characteristics
The mean age of studies was 15.57 years (SD = 10.52), with a range of 0-42 years. The Price index (the percentage of references ≤5 years old) was 21.5% [68,69]. Four common types of non-pharmacological interventions were identified: (1) schedule changes (e.g., switching from backwards rotation to forward rotation or changing length of shifts), (2) behavioural interventions (e.g., sleep education or physical activity program), (3) controlled light exposure (e.g., intermittent bright light or light-blocking glasses), and (4) complementary therapy interventions (e.g., massage or acupuncture). The majority of interventions involved a schedule change (n = 30), behavioural change intervention (n = 17), or controlled light exposure (n = 14). Four studies involved complementary therapy interventions. Table 2 outlines the outcomes reported in each study. Subjective sleep measures were evaluated in 75% of studies (n = 49) while 34% (n = 22) utilised objective measures of sleep. Biomedical risk factors were evaluated in 39% (n = 25) and behavioural risk factors were evaluated in 42% (n = 27) of studies. An overview of the data extracted is presented in Tables 3-6.

Risk of Bias within Studies
Risk of bias assessment ratings varied between 3 and 22 from a possible score of 28, with lower scores reflecting lower quality studies; 90% agreement was achieved between the reviewers. There was considerable variability in risk of bias and quality assessment scores within studies. A full outline of the Downs and Black assessment is provided in Supplementary Material 2.

Participant Characteristics
The studies eligible for qualitative and quantitative analysis included a total of 7806 participants, 56.6% (n = 4420) male, 29.1% female (n = 2269) and the remaining 14.3% (n = 1117) did not report sex of participants. Mean age of participants, where reported, ranged from 23.4 to 52.5 years.
3.5. Results of Individual Studies 3.5.1. Schedule Change As shown in Table 3, change of shift length (38%, n = 11) and change of direction of rotation (e.g., change from days to nights instead of nights to days) (31%, n = 9) represented the most common interventions. A further 21% (n = 6) of interventions utilised increased rest or recovery periods, and 14% (n = 4) investigated a change in speed of rotation. One study compared flexible vs. 80 h week limit in medical residents working extended hours.
Hakola, 2010 schedule • Shaded squares indicate studies included in meta-analysis for corresponding outcome. (a) Outcome was not reported at follow up or was only used as a co-variate, (b) insufficient information reported for means and/or standard deviations to be calculated, (c) not eligible for inclusion in meta-analysis (follow-up study or extended working hours), and (d) meta-analysis not conducted due to insufficient studies.  HDL-C level increased LDL-C and the total HDL-C cholesterol ratio decreased      Interventions involving a change in shift length yielded variable results. Of the eleven studies, 64% (n = 7) found that change from an 8 h to a 12 h shift system (e.g., two-shift system) resulted in improvements in subjective sleep parameters, such as increased sleep duration [70,89,91,97] and improved sleep quality [88,93]. Further, a change to 12 h shifts also resulted in decreased blood pressure [91]. However, 18% of studies (n = 2) found a negative effect of change to a two-shift system on subjective sleep duration following night shift [77,97]. In addition, a change from 8 to 12 h shifts also resulted in a significant increase in BMI in male clean room workers [99]. One study compared 8, 10 and 12 h shift systems in police officers [71] and found that officers had the longest sleep duration in the 10 h condition, compared to both 8 and 12 h shifts.
Of the studies that changed the direction of shift rotation, 78% (n = 7) moved from backward to forward rotation [78,80,83,84,86,90,95]. A further 22% (n = 2) investigated a change from forward rotation to backward rotation [72,85]. A change from backward to forward rotation resulted in increased subjective sleep duration [80,83,84,90] and subjective sleep quality [78,83,84,90]. Changing to forward rotation also resulted in improvements in objective sleep efficiency [78]. Further, changing to forward rotation also demonstrated decreases in serum glucose and blood pressure, as well as improved perceived health status [90]. However, the results of change of rotation on blood pressure was not consistent across all studies, with another study finding that a change to forward rotation resulted in increased blood pressure [95]. The change from forward rotation to backward rotation was associated with improved sleep quality and improved perceived health status [85]. In Karlson et al.'s [85] study of rotation change, the change also resulted in increased recovery periods with three days off between schedules. In contrast, another study found that a change from forward to backward rotation was associated with increased sleep difficulties [72]. A change from discontinuous to continuous rotation resulted in improved sleep quality [81]. However, this change was not sustained at 4.5 year follow up [82].
Of the studies that investigated rest periods (n = 6), 83% (n = 5) involved increased time between shifts and 16% (n = 1) introduced a rest period during work hours. Introduction of a scheduled short rest period improved subjective sleep quality [75]. Further, increasing recovery time between evening and morning shifts also improved subjective sleep duration [79]. The introduction of a half-day shift prior to night shift increased objective sleep duration [87,96]. However, both studies were conducted with samples of nurses and over a very short period (3-4 days) and therefore may not be generalisable to other settings or over a longer term. Delaying starting times resulted in improved subjective and objective sleep duration, but decreased sleep quality in steel mill workers [92]. Lastly, increased rest periods resulted in improvements in blood lipids when night shifts were followed by an extra recovery day in nurses and nurses' aids [74].
Finally, an investigation of length of working week in medical residents found no difference in objective sleep duration between flexible systems and restricted maximum shift hours [73].

Behavioural Interventions
As shown in Table 4, approximately half 44% (n = 7) of the interventions that utilised behavioural change methods provided sleep and/or fatigue education sessions, 25% (n = 4) investigated prescriptive physical activity interventions and 25% (n = 4) targeted health behaviours interventions such as education to improve healthy eating and increasing physical activity. One study investigated a prescriptive bedtime of 10 h prior to shift start time.     In 71% (n = 5) of sleep education interventions, there was no significant effect of intervention for objective sleep duration one month [106,113] or four to five months [112] after intervention, subjective sleep duration [112][113][114] or subjective sleep quality [104,113]. However, sleep education utilising individualised cognitive behavioural therapy resulted in improved subjective sleep quality [106]. Further, sleep education resulted in increased subjective sleep duration when education was given both to workers and their immediate family members [105].
Prescriptive physical activity interventions involved supervised training with physical therapist [103,108,109] or weekly consultation with physical therapist via phone [111]. Prescriptive physical activity interventions demonstrated effects in favour of intervention in 75% (n = 3) of the studies, namely decreased BMI [111], increased HDL cholesterol [109], and increased subjective sleep duration [103].
Studies that investigated targeted health behaviour interventions utilised weekly team curriculum [101], individualised motivation interviewing [101], increased availability of healthy meals at work [107], a weight loss education session and online resources [110] and tailored health behaviour suggestions based on shifts being worked [115]. Two of these interventions were based on elements of Bandura's Social Cognitive theory [101,110]. All of the targeted health behavioural interventions found improvements on at least one outcome of interest. Interventions resulted in decreased BMI [110], decreased blood pressure [110], increased physical activity [110,115], increased fruit and vegetable consumption [101] and increased water intake [107]. Further, targeted health behaviour interventions also resulted in significant improvements in sleep quality [115] and perceived health status [101].
Lastly, one study evaluated the effectiveness of prescribing a bedtime ten hours prior to the start time of the following shift. The prescription of bedtime resulted in increased objective sleep duration and increased subjective sleep duration and quality [116]. Further, the sample for this study was small (n = 16) but did investigate a range of occupations (manufacturing, customer service, nursing, and public service).

Controlled Light Exposure
As presented in Table 5, of the studies that investigated controlled light exposure interventions (n = 14), 50% (n = 7) evaluated the effect of intermittent bright light exposure, 36% (n = 5) utilised both bright light exposure and light-blocking glasses, while 14% (n = 2) investigated dynamic lighting in the workplace. One study investigated the effect of light-blocking glasses alone.    Of the studies that investigated intermittent bright light exposure, an increase in objective sleep duration [123,130], decrease in sleep onset latency [118] and improved subjective sleep quality [128] were observed. However, one study found that bright light exposure resulted in decreased subjective sleep duration [121], while 29% (n = 2) found no significant effects associated with bright light exposure [117,120].
In all of the studies that combined bright light exposure and light-blocking glasses, an improvement was found in at least one outcome. In 80% (n = 4) of these studies, objective sleep duration increased [119,126,129,130] and in 40% (n = 2) studies objective sleep efficiency improved [129,130]. Subjective sleep quality was more variable, with one study finding an increase [124], and another a decrease in subjective sleep quality [129].
Two studies investigated the effect of dynamic lighting in the workplace, both involving nursing populations working within hospitals. Dynamic lighting resulted in no change in sleep parameters [122,127]. However, dynamic lighting was associated with a decrease in perceived health status [127]. Lastly, one study evaluated the effect of light-blocking glasses which resulted in increased objective sleep duration and efficiency [125]. In terms of side effects, participants reported difficultly falling asleep on days off when in the treatment condition [120] and headaches [118,124,127] and eye strain [124]. However, all of these studies report a minority of participants experienced these side effects.

Complementary Interventions
As shown in Table 6, four complementary forms of intervention were included for review. Two studies investigated the effect of massage [131,132], one study investigated the effect of touch therapy [133] and one study investigated the effect of Transcutaneous Electrical Acupoint Stimulation [134]. Half (n = 2) of the studies examined blood pressure changes [132,133] and the other half (n = 2) examined subjective sleep quality [131,134]. One study investigated objective sleep duration [131]. Two studies [131,134] found significantly improved subjective sleep following intervention. Two of the studies found that control [131] and sham [134] groups also had significantly improved subjective sleep quality. One study found a significant decrease in systolic blood pressure in the control (reading) group but no change in intervention (massage) group [132].
In all four studies, the controls received a work rest break similar to that of the intervention group, but without the treatment. One study found the active control ("sham" acupressure points) group also had significant improvements in subjective sleep quality [134]. Further, a significant improvement in blood pressure in the control (reading and resting) group was found in a separate study [132]. This may suggest that there was a positive effect of a scheduled rest period in the workplace. All four studies were conducted in healthcare personnel, thus the generalisability of these results to other occupations is unknown. Therefore, future studies of complementary therapy in other occupational groups would be of benefit. Complementary therapies may confer benefit for sleep quality in nurses.

Synthesis of Quantitative Results
Eight outcomes were eligible for meta-analysis as there were a minimum of three studies reporting these outcomes. These included systolic blood pressure, diastolic blood pressure, body mass index, objective sleep duration, objective sleep efficiency, subjective sleep duration, subjective sleep quality, and perceived health status. A total of fifty-eight studies were identified as being eligible for inclusion in at least one these meta-analyses. Following the quantitative data procedure detailed in the methods section, nineteen studies had insufficient data and were subsequently excluded from meta-analyses. An outline of included and excluded studies is provided in Table 2.

Sleep Efficiency
Forest plots for sleep efficiency are shown in Figure 6. For sleep efficiency there was a medium total pooled effect size (Hedges' g = 0.48; CI: 0.20, 0.76, k = 10) with low heterogeneity (τ 2 = 0.06 and I 2 = 0.0%) and a prediction interval of −0.16; 1.11. Subgroup analysis showed a significant difference between intervention types (p = 0.02) with a medium total pooled effect size for controlled light exposure on sleep efficiency (Hedges' g = 0.59; CI: 0.38, 0.79, k = 9) and a small negative pooled effect size for behavioural interventions (Hedges' g = −0.29; CI: −0.99, 0.40, k = 1). Eggers' test of the intercept indicated no evidence of publication bias (p = 0.35, Figure 3E).

Summary of Evidence
The present review provides the first comprehensive qualitative and quantitative analysis of the scope and effectiveness of non-pharmacological interventions for chronic disease risk factors, sleep outcomes and perceived health status in shift workers. The present review showed that schedule change, behavioural change, controlled light exposure and complementary therapy interventions have been used to improve chronic disease risk factors and sleep in shift workers.

Meta-Analytic Findings
In considering biomedical risk factors, only the outcomes of blood pressure and BMI had sufficient studies to be included for meta-analysis to date. For behavioural risk factors, there were insufficient studies for meta-analysis across all risk factors. Both objective sleep measures and subjective sleep measures had sufficient literature to allow for metaanalysis. There were also sufficient studies examining perceived health status for metaanalysis. Heterogeneity was common, likely reflecting the varied methods of intervention utilised and differences between occupational settings.
In studies with sufficient data for meta-analysis, there appear to be small favourable effects of interventions on biomedical risk factors, being blood pressure (both systolic and diastolic) and BMI. Further, subgroup analysis showed a significant difference of effect by intervention type, with behavioural interventions resulting in favourable effects. For objective sleep measures, there were large favourable effects of intervention on sleep duration and medium effects on sleep efficiency. In addition, subgroup analysis showed that controlled light exposure had a significant effect on sleep efficiency compared to behav-

Summary of Evidence
The present review provides the first comprehensive qualitative and quantitative analysis of the scope and effectiveness of non-pharmacological interventions for chronic disease risk factors, sleep outcomes and perceived health status in shift workers. The present review showed that schedule change, behavioural change, controlled light exposure and complementary therapy interventions have been used to improve chronic disease risk factors and sleep in shift workers.

Meta-Analytic Findings
In considering biomedical risk factors, only the outcomes of blood pressure and BMI had sufficient studies to be included for meta-analysis to date. For behavioural risk factors, there were insufficient studies for meta-analysis across all risk factors. Both objective sleep measures and subjective sleep measures had sufficient literature to allow for meta-analysis. There were also sufficient studies examining perceived health status for meta-analysis. Heterogeneity was common, likely reflecting the varied methods of intervention utilised and differences between occupational settings.
In studies with sufficient data for meta-analysis, there appear to be small favourable effects of interventions on biomedical risk factors, being blood pressure (both systolic and diastolic) and BMI. Further, subgroup analysis showed a significant difference of effect by intervention type, with behavioural interventions resulting in favourable effects. For objec-tive sleep measures, there were large favourable effects of intervention on sleep duration and medium effects on sleep efficiency. In addition, subgroup analysis showed that controlled light exposure had a significant effect on sleep efficiency compared to behavioural interventions. However, subjective measures of sleep and perceived health status only showed small effects of intervention, with no differences between intervention types.

Qualitative Findings
The present review indicates that interventions aimed at improving sleep in shift workers are well represented, albeit diverse in their approaches and measurement of other chronic disease risk factors. This is unsurprising given the immense evidence of the negative sleep effects associated with shift work [1][2][3][28][29][30]. However, consistent measures of biomedical and behavioural risk factors were present in relatively few studies. This indicates a need for future research which considers risk factors associated with chronic disease, beyond sleep outcomes, if we are to inform a strong evidence base for early (pre-disease) intervention in shift workers.
Findings differed across outcome and intervention type and therefore, it is not possible to suggest that any one type of intervention may be best to improve overall chronic disease risks for shift workers. This is not surprising-while shift work may be a common feature for these workers, individual differences in occupation, sociodemographic characteristics, job demands, and health status likely call for unique interventions. Broadly, objective sleep measures were improved by controlled light exposure including intermittent bright light [123,130], bright light and light-blocking goggles [119,126,129,130], light-blocking goggles alone [125], schedule change which increased recovery periods [87,92,96] and prescriptive sleep scheduling [116]. Subjective measures of sleep were improved by a change to 12 h shifts (e.g., two-shift system) [70,88,89,91,93,97], a change to forward rotation schedules [78,80,83,90], increased rest or recovery periods [75,79], individualised behavioural intervention aimed at improving sleep and fatigue [106,115], and some complementary therapies (e.g., electrical acupoint stimulation and aromatherapy massage) [131,134].
Of those biomedical risk factors reported, most did not change significantly, except for a decrease in blood pressure associated with change from backward to forward rotation [90], a change to 12 h shifts [91] and a behavioural intervention targeting weight loss and healthy eating [110]. Further, cholesterol showed some improvements in schedules that increased recovery periods [74] and a prescribed physical activity program [109]. BMI significantly decreased in behavioural interventions targeting weight loss through healthy eating and increased physical activity [110] and prescribed physical activity intervention [111].
When evaluating behavioural risk factors, an increase in physical activity was associated with prescribed physical activity intervention [111], interventions based on Social Cognitive theory [110] and personalised behavioural interventions [115]. The only study reporting a significant difference in nutritional intake was a behavioural intervention based on Social Cognitive theory [101]. Finally, perceived health status was improved with behavioural intervention based on Social Cognitive theory [101], change to a slower rotation schedule [76] and change to slow backward rotation schedule [85].
Schedule change interventions were commonly used. This is largely unsurprising, as an extensive body of literature has shown that schedule characteristics are associated with health and safety outcomes [135][136][137][138][139][140]. Results of these interventions favoured two shift changes, a forward-rotating shift system and increased length of recovery periods. While it is positive that schedule changes can benefit individual worker health, one of the major limitations of schedule changes is that they require management to be actively involved and require change at an organisational level, which is not always simple. Further, comparison of interventions involving schedule changes are somewhat hampered by heterogeneity in the shift systems being worked before workers underwent respective interventions. Studies utilised differing shift designs at baseline, therefore further complicating the ability to compare across all schedule change interventions. Thus, while changing schedules to optimal timing for workers is likely beneficial for both safety and health outcomes, additional individual interventions may be needed for workers.
Behavioural interventions resulted in some improvements in sleep and other health indicators. Importantly, behavioural interventions that tailored the information provided to workers were more likely to result in positive changes. This is consistent with previous studies comparing individualised to non-individualised interventions for health behaviours [141,142]. An example is the study by Van Drongelen et al. [115], in which a control group received general information regarding fatigue and health behaviours, while the intervention group received personalised information. The findings suggest that without tailoring educational resources, providing information about fatigue, sleep behaviours or healthy diet may not be effective in a shift-working population. It is possible that these findings can be attributed to the complex nature of behavioural change. Shift workers may not know that they personally are at risk of poorer health outcomes and thus may not interpret the information provided as relevant to them. Thus, the use of a behavioural change framework for future interventions may be effective. Further, where behavioural interventions were informed by an existing behavioural model, studies yielded significant improvements on various outcomes indicating the possible utility of adapting existing behavioural models for use in shift-working populations to improve chronic disease risk factors.
Qualitative analysis indicates that controlled light therapy was beneficial for shift workers, particularly for improved objective sleep parameters. This was supported by meta-analysis. However, the strength and timing of bright light varied greatly between studies. Unfortunately, there were not enough studies to conduct subgroup analysis by light timing or strength. Consequently, there remains limited evidence to date to establish the most suitable timing or intensity of light administration to provide clear guidelines for shift workers. Additionally, the available studies were largely conducted on nursing populations or those working offshore, therefore limiting our understanding of controlled light exposure in other occupational settings. The overrepresentation of these occupations may simply indicate that the use of controlled light may be more practical in these settings where nurses have common staff facilitates (e.g., break room) and offshore industrial workers living on location. Feasibility for other shift-working populations is unclear and warrants further consideration.
It is also important to note that controlled light appeared to be associated with some negative side effects (e.g., headaches or eyestrain) [118,120,124,127], decreased subjective sleep quality [129] and deterioration in perceived health status [127]. Future studies should include sufficient measures of possible negative side effects when conducting controlled light interventions. It appears that an important balance will need to be found between use of light therapy and well-being from the individual worker's perspective. Taking an individualised approach to light therapy such as adjusting the timing and strength of light exposure according to shift schedule [143] may be an option; however, future studies should first consider feasibility (and sustainability) of personalised approaches in a robust, experimental manner.
Complementary therapies, whilst limited in number, showed interesting results for health risk indicators in shift workers. Both studies that investigated subjective sleep quality found a significant positive effect of intervention on sleep quality. However, a limitation of the literature on complementary therapies in shift workers is that all were conducted within healthcare populations, limiting the generalisability. It is unclear whether the same benefits would be observed in other occupations as no complementary interventions used in other occupational groups were identified. While the literature here is limited, it is feasible that complementary therapies may afford benefits for sleep in shift workers. Future studies could consider shift workers from other occupations to identify specific worker populations for whom complementary therapies confer greatest benefit.
The present systematic review is strengthened by an integration of qualitative and quantitative analysis. Further, use of chronic disease criteria in combination with sleep and perceived health status allowed for a robust understanding of interventions targeting improvements to chronic disease risk factors in shift workers. Beyond this, for meta-analytic data, the present review evaluated effect sizes by outcome, and also by intervention type. This allowed for quantitative understanding of the impact of type of intervention on various outcomes. Finally, the present quantitative analysis was based on the combination of effect sizes when sleep outcomes were reported by shift, allowing for comparison of total sleep duration and quality across all available studies. This technique overcomes some reporting barriers which make qualitative comparisons difficult, while also allowing for a consistent measure of the impact of intervention on total sleep time.

Limitations
A substantial number of studies could not be included in meta-analyses due to insufficient reporting or data that were not compatible with measures of standardised mean difference (see Table 2). Therefore, while these meta-analytic outcomes provide preliminary understanding of the effect of interventions on some important health measures in shift workers, it is necessary to acknowledge not all studies have been included. This highlights the importance of future data sharing and open science practices to allow for accurate evaluation of intervention studies.
The review is limited by the minimal investigation of biomedical and behavioural risk factors associated with downstream chronic disease in shift work interventions. Given that each of these risk factors is known to contribute to the development of various chronic disease, it is important that such risk factors (beyond sleep outcomes) are targeted within a shift-working population. Another limitation within the current studies is the relatively limited studies that utilised an individualised design. Personalisation of behavioural interventions [141,142] and controlled light exposure [143] have been shown to have positive effects for individuals and should be considered for future intervention studies.
Quality assessment and risk of bias indicated significant variability in reporting, external validity, study bias and possible confounding. The average estimates of these scores were impacted by studies that scored particularly low on the assessment scale [86,105]. Beyond this, a difference among studies is not unexpected as the review included both randomised trials and observational studies and did not limit studies to recent publications. Nonetheless, it is crucial that future studies and subsequent publications aim to limit bias and confounding wherever possible and ensure greater transparency with reporting in publications.
Another limitation of the studies within the present review is that many interventions were only reported in one occupational group, and thus generalisability to other occupational groups may be limited. This highlights a pressing need for expanded intervention studies in shift-working populations in the future in order to address the known long-term chronic disease outcomes associated with this work pattern. Further, studies differed in both baseline characteristics and intervention types making meaningful comparison challenging. It will be useful for future studies to consider these existing interventions when designing new studies in order to allow for comparison with existing literature. Lastly, none of the current interventions address all the chronic disease risk factors outlined, therefore limiting our ability to compare current interventions on all risk factors. Future studies may consider including additional outcomes to allow for analysis of the effect of study across multiple chronic disease risk factors.

Conclusions
The findings of this systematic review and meta-analysis suggest that some chronic disease risk factors, particularly sleep outcomes, can be improved with interventions.
Objective sleep outcomes are improved with a medium to large effect, with controlled light exposure and schedule change producing large favourable effects. Subjective sleep duration is improved with a small effect size, with behavioural interventions contributing the largest positive effect. Subjective sleep quality is also improved, with a small effect size, with complementary therapies producing the largest effect. Qualitative analysis suggests that two-shift, forward-rotating schedules, increased recovery periods, tailored behavioural interventions, interventions based on existing behavioural theoretical frameworks and intermittent bright light and light-blocking goggles, improve some chronic disease risk factors, sleep and perceived health status in shift workers. The limited volume of interventions targeting biomedical and behavioural risk factors in shift workers indicates a need for further studies within this area.
Future studies should consider the efficacy of existing interventions, as analysed in this review. The implementation of promising interventions such as two-shift schedules, increased rest periods, bright light exposure combined with light-blocking goggles, and tailored behavioural interventions in different occupational groups and across larger samples would provide additional understanding of the feasibility and effectiveness of these interventions for all shift workers. Importantly, future studies should consider targeting various chronic disease outcomes, aim to take all practical steps to limit bias and confounding, and to report as many study details as possible in order to assist in the further development and enhancement of interventions to improve chronic disease, sleep and health outcomes in shift workers.