Effectiveness of Lifestyle Health Promotion Interventions for Nurses: A Systematic Review

Background: Prior research has investigated various strategies to improve health, wellbeing and the job-related outcomes of nurses. However, the scope of this evidence is not clear and the types of intervention most likely to have positive outcomes are unknown. Objective: To provide an overview and synthesis of the effectiveness of interventions conducted with the goal of improving health, wellbeing and the job-related outcomes of nurses. Methods: A systematic database search was conducted from January 2000 to December 2018, with pre-defined criteria (Cochrane Central Register of Controlled Trials; MEDLINE and PubMed; EMBASE; CINAHL; PsycINFO; and BioMed Central). In total, 136 intervention studies with a total sample of 16,129 participants (range 9–3381) were included and evaluated. Data extraction, quality assessment and risk of bias analyses were performed. Results: Studies included randomised controlled trials (RCTs; n = 52, 38%), randomised crossover design studies (n = 2, 1.5%) and non-randomised pre-post studies with a control group (n = 31, 23%) and without a control group (n = 51, 37.5%). The majority of interventions focused on education, physical activity, mindfulness, or relaxation. Thirty-seven (27%) studies had a multimodal intervention approach. On average, studies had relatively small samples (median = 61; mode = 30) and were conducted predominantly in North America (USA/Canada, n = 53). The findings were mixed overall, with some studies reporting benefits and others finding no effects. Dietary habits was the most successfully improved outcome (8/9), followed by indices of body composition (20/24), physical activity (PA) (11/14), and stress (49/66), with >70% of relevant studies in each of these categories reporting improvements. The lowest success rate was for work-related outcomes (16/32). Separate analysis of RCTs indicated that interventions that focus solely on education might be less likely to result in positive outcomes than interventions targeting behavioural change. Conclusions: Interventions targeting diet, body composition, PA, or stress are most likely to have positive outcomes for nurses’ health and/or wellbeing. The methodologically strongest evidence (RCTs) is available for body composition and stress. Interventions relying solely on educational approaches are least likely to be effective. Organisational outcomes appear to be more challenging to change with lifestyle intervention, likely requiring more complex solutions including changes to the work environment. There is a need for more high-quality evidence since many studies had moderate or high risk of bias and low reporting quality.


Introduction
Nurses are at the frontline of public health and spend considerable time promoting healthy lifestyle behaviours to patients and their families. However, studies of lifestyle behaviours in nurses We excluded interventions that focused solely on health and safety initiatives, or the improvement of clinical skills (e.g., hand washing or infection control methods, patient moving and handling techniques and nurse-patient communication skills). Interventions were excluded that focused primarily on the treatment of psychological disorders (such as post-traumatic stress), although wellbeing interventions focused on stress-management and the prevention of psychological disorders or compassion fatigue were included (e.g., studies focused on nurses' stress or personal wellbeing in which factors such as compassion fatigue, communication skills or psychological disorder are measured as outcomes or are targeted as part of a multicomponent lifestyle intervention). Interventions that focused solely on organisational changes without a focus on individual health and wellbeing were excluded. Lifestyle interventions primarily targeting nurses' physical, clinical and/or psychological health, but measuring work-related outcomes, were included.

Selection Processes
Three reviewers (MN, NSh, EK) independently performed a study selection process and any duplicated records were removed. The titles and abstracts of the remaining records were screened, and full texts were sought for records which clearly referred to behavioural and/or educational lifestyle interventions for working age nurses. The full texts were then assessed for eligibility, taking into account intervention type, study population, outcomes reported, and language. The agreement on inclusion and exclusion was reached through discussion between the reviewers (MN, NSh, EK), with any disagreements resolved by a fourth reviewer (HB).

Data Extraction
The data extraction was performed independently by two reviewers (NSt, HB) and agreement was reached through discussion (i.e., both reviewers checked the data extraction table, and discussed any inconsistencies to reach a consensus; this was needed if the data were not particularly clear). The details on participants, setting, intervention, and outcome measures were extracted from each study. The methodological features of all studies were assessed using the CONSORT checklist [52,53].

Risk of Bias
Three reviewers independently reviewed and critiqued the retrieved papers (NSt, HB, EK) and any disagreements were discussed. The risk of bias was assessed using the Cochrane Handbook classification [54] for all included papers. The risk of selection, performance, detection, attrition and reporting bias were assessed. The risk of bias for non-RCT studies was assessed as high for the categories that could not be satisfied in such designs.

Method of Synthesis
The studies were summarised narratively, which is acknowledged as an appropriate approach to take when assessing data from heterogenous study designs [55]. We report narrative synthesis of the findings from the included studies, structured around the type of intervention, target population characteristics, type of outcome and intervention content. We also provide summaries of the intervention effects for each study (see Table S1 in Supplementary Material).

Included Studies
The initial search resulted in 17,126 potential articles. A review of the titles and abstracts resulted in a sample of 567 being selected for further review. The abstracts and full texts of these papers were compared against the inclusion and exclusion criteria, which resulted in 435 articles being excluded. The remaining 132 papers were hand searched. The hand search resulted in four additional papers being included. Subsequently, a total of 136 papers were included in this review. Figure 1 demonstrates the flow of the study selection process.

Designs of the Included Studies
The studies in this review included randomised controlled trials (RCTs; n = 52), randomised crossover design studies (n = 2), non-randomised pre-post studies with a control group (n = 31) and without a control group (n = 51). They all examined or compared interventions aimed at improving physical or mental health and wellbeing and/or work-related outcomes in nurses. Comparison groups included a wait-list control (n = 12), an active control (n = 27), no intervention (n = 41), or care as usual (n = 5). Fifty-one studies had no control/comparison group. Supplementary Table S1 describes the characteristics of the included studies.

Characteristics of the Samples
The sample sizes in the included studies ranged from nine to 3381 participants (total participants = 16,129; median 61; mode 30). The mean age of participants in the studies reporting this value was 39.48 years (SD = 7.18); 37 studies did not provide such data. In the majority of studies in which gender was reported (n = 101), females, on average, accounted for 91% of participants, with 23 studies reporting all female nurses. In terms of geographical location, 58 studies were conducted in North or South America (including 53 from USA/Canada), 39 in Asia, 30 in Europe, eight in Australia, and one

Designs of the Included Studies
The studies in this review included randomised controlled trials (RCTs; n = 52), randomised crossover design studies (n = 2), non-randomised pre-post studies with a control group (n = 31) and without a control group (n = 51). They all examined or compared interventions aimed at improving physical or mental health and wellbeing and/or work-related outcomes in nurses. Comparison groups included a wait-list control (n = 12), an active control (n = 27), no intervention (n = 41), or care as usual (n = 5). Fifty-one studies had no control/comparison group. Supplementary Table S1 describes the characteristics of the included studies.

Characteristics of the Samples
The sample sizes in the included studies ranged from nine to 3381 participants (total participants = 16,129; median 61; mode 30). The mean age of participants in the studies reporting this value was 39.48 years (SD = 7.18); 37 studies did not provide such data. In the majority of studies in which gender was reported (n = 101), females, on average, accounted for 91% of participants, with 23 studies reporting all female nurses. In terms of geographical location, 58 studies were conducted in North or South America (including 53 from USA/Canada), 39 in Asia, 30 in Europe, eight in Australia, and one study used a cross-cultural sample from both America and Asia. Only two studies were conducted in the UK.

Characteristics of the Interventions
The included studies were grouped according to intervention type. Many of the intervention studies included more than one intervention type (n = 37 studies).

Intervention Duration and Follow-up
Overall, the intervention length ranged from 10 min (e.g., one short massage session) to 2 years (mean 2.16 months; SD = 2.6; mode 2 months), although six studies did not provide sufficient details on intervention length. The majority of outcomes were assessed immediately after the end of the intervention, with only a few studies assessing medium or longer-term intervention effects.

Intervention Settings
The interventions were predominantly delivered in hospital wards/medical centres or ambulatory clinics (n = 123), with less common settings being a hospice (n = 2) [87,179], and residential or care homes for older people (n = 4) [74,75,98,109], as well as private home care settings [64] (See Table S1 for more details). One study had a sample which included hospital nurses as well as nurses who were municipal employees [69], another study included nurses from various settings (both community and institutional) [189], whereas another four did not specify where the nurses were employed [108,140,142,174].

Measures Used
The outcome variables were assessed by a multitude of measures, and the vast majority of measures were self-report questionnaires. The key questionnaires used are presented below.

Health Risk Factors
[i] Clinical Health Outcomes: Self-report measures of general health were more often used and included the Short-Form Survey (SF) [67,130,133,147,168,183,186], the Symptom Checklist (SCL-90) [147,150,157], the General Health Questionnaire (GHQ) [72,100,126,146,160], the Pennebaker Inventory of Limbic Languidness [176], and the Standard Shift-work Index [177].
[iii] Diet and Nutrition: Mostly measured using self-report measures such as the Health-Promoting Lifestyle Profile [HPLP; 61,67], the New South Wales (NSW) Health Survey [127], snack intake (self-report [140]), the Food Frequency Questionnaire [113] and the Rapid Block Food Screener [112]. Two studies measured cholesterol level as an outcome [83,118].
[iv] Physical Activity and Sedentary Behaviour: Most often measured by self-report questionnaires including the Health-Promoting Lifestyle Profile [HPLP; 61,67], the Yale Physical Activity Survey [112], the International Physical Activity Questionnaire (IPAQ) [83], and the Active Australia Questionnaire [127]. A small number of studies used objective measures of PA (see below). The objective measures included activity trackers (pedometers [96,112,115,123,131], or accelerometers [113]); the UKK walking test [75], or aerobic capacity using the VO 2max test [117,118].

Overall Effect of the Interventions
The majority of interventions in the included studies resulted in significant improvements in at least one measured outcome, although some of the outcomes were not improved following intervention exposure. Health behaviours (including PA, diet, smoking, alcohol consumption), clinical or health outcomes, and work-related outcomes were less often measured than indices of psychological wellbeing. Overall, the strongest evidence was for (i.e., improvements reported in a high number of studies) improvements in stress, anxiety, and burnout (mostly emotional exhaustion (EE) and depersonalisation (DP)). There was some evidence for (i.e., improvements reported in a lower number of studies) personal achievement (PAch), wellbeing, compassion (satisfaction and fatigue), work functioning, PA and indices of body composition (BMI, weight). The outcomes that were less likely to change following intervention were depressive symptoms, personal accomplishment (burnout subdomain), life and job satisfaction, and job control. Based on the outcomes measured in included studies and this overall trend, it appears that lifestyle interventions were more likely to positively influence emotional-based outcomes (heavily relying on mood state, emotional valence), and less likely to positively impact cognitive-focused outcomes (such as quality of life or job-related perceptions, which are assessed more cognitively than emotionally).

Health Risk Factors
Clinical Health Outcomes Physical Symptoms and General Health: Of the included studies, 17 included a measure of general health or physical symptoms. Of these, 11 [67,72,100,133,136,146,157,160,176,177,183] demonstrated improvements in health following intervention, including physical symptoms ( [157] as measured by the Symptom Checklist-90; SCL-90), and physical health ( [67,183] as measured by Short Form-36; [72,100,146,160] as measured by the General Health Questionnaire; [133] as measured by Short Form-12; [177] as measured by the Standard Shiftwork Index), psychosomatic symptoms ( [176] as measured by the Pennebaker Inventory of Limbic Languidness), sickness and doctor's visits [160], and perception of one's health ( [136], measured with a single item). Of the 11 studies reporting improvements in a measure of health or physical symptoms, four were RCTs [100,157,176,183], three were non-randomised controlled studies [67,160,177] and four were uncontrolled studies [72,133,136,146]. Those studies demonstrating health improvements reported interventions based on mindfulness ( [146,157]-plus yoga), health [67] and coping [72] education, emotional intelligence (EI) education [100], relaxation ( [160,176] with meditation), auriculotherapy [183], sleep and relaxation [177] and PA [136].
A further six studies reported no significant change in health or physical symptoms [75,126,130,131,147,168], as measured by Short Form-36 [130,147,168], SCL-90 [147], health complaints [75], GHQ [126], or cardiovascular health (i.e., resting blood pressure, [131]). Of the six studies reporting no changes in measures of health or physical symptoms, three were RCTs [75,130,131], and three were non-randomised controlled studies [126,147,168]. The interventions failing to demonstrate positive outcomes used PA [130,131], PA with mindfulness [126], PA with stress education [75], mindfulness [147] and art-based relaxation [168]. None of the studies reported other clinical outcomes.

Diet and Nutrition
Nine studies reported nutrition (healthy eating) as an outcome, with all but one study [83] reporting positive outcomes for diet or nutrition following intervention [61,67,88,112,113,118,127,140]. Of the eight studies reporting improvements in diet or nutrition, two were RCTs [118,140], two were non-randomised controlled studies [61,67] and four were uncontrolled studies [88,112,113,127]. These interventions were predominantly based on education, including an e-health website [67,112], face-to-face education sessions [88,112], creating self-care plans [61], keeping track of one's steps and diet [127], providing physical resources (water bottle, sandwich box, healthy cookbook [88,127]), providing cooking classes [88], setting action plans for lower snack intake [140], goal setting for changes in diet [113], with five studies also incorporating a PA element (Wii exercises [112], aerobics [118], or walking [88,113,127]). All the outcomes measured in these seven studies were based on self-report methods, and included reports of fruit and vegetable intake [112,113,127], cholesterol [118], snack intake [140], breakfast consumption [127], Health-Promoting Lifestyle Profile (HPLP) score [61,67] or self-devised [88] questionnaires. The single study that showed no positive outcome was an uncontrolled study, which used blood tests to determine cholesterol level. This unsuccessful intervention combined nutritional intervention with PA.
Not all of the PA and exercise outcomes were improved by these interventions. For example, one study [75] improved leisure PA (self-report), but not aerobic fitness (objectively measured). Another study [118] failed to demonstrate improvement in aerobic fitness (objective measure of maximum oxygen uptake) but reported improvements in muscle strength (objectively measured with dynamometer). Kcal burnt per week were improved in one study [112] as well as minutes of exercise per week, but no improvements were observed in number of steps per day. Another study reported significant change in minutes spent sitting per day, but not in the MET scores [83].
Of the 11 studies reporting improvements in PA and/or sedentary behaviour, three were RCTs [75,115,118], one was a non-randomised controlled study [67] and seven were uncontrolled studies [83,88,112,119,123,127,136]. Successful interventions (even if only for some outcomes) used the following types of PA interventions: (i) worksite intervention (incl. workstation treadmill, Wii system, short video clips with energetic activities, walking meetings) with health coaching via text messaging [115], (ii) healthy lifestyle website with discussion board [67], (iii) healthy lifestyle education group sessions, website, eHealth journal, Wii system at work [112], (iv) workstation wellness intervention (to increase standing, stretching and sipping water) [119], (v) minimal-contact self-managed (setting one's own PA goal) pedometer program [123], (vi) workplace 1h/week light group exercise plus healthy lifestyle education classes [75], (vii) pedometer challenge [136], (plus recording daily steps on a website, with 10k daily steps goal) with educational classes on healthy lifestyle [83], or physical resources (water bottle, cookbook, prizes [127]), or both [88] also with extra group exercise sessions), or (viii) aerobics and resistance exercise with or without supervision. Several of the studies reporting improvements in PA included digital components to their intervention [67,112,115,123,136].
However, two other studies reported no improvement in PA, exercise or sedentary behaviour (one RCT: [76], one non-randomised controlled study: [96]) and one study reported decreased PA (daily steps and moderate-to-vigorous PA, both measured objectively) after the intervention (a non-controlled study: [113]). These unsuccessful interventions were based on: (i) worksite intervention (like in the study of [115]; also measured with activity monitor- [96], (ii) individual physiotherapy exercises with educational sessions [76]; however, this study compared the results between two groups that underwent physiotherapy exercises, with the only difference being extra educational sessions), (iii) a complex intervention (for increasing PA and diet, [113], including app for sharing recipes/tips/PA goals, Facebook groups for support, pedometer to set and monitor PA goals).
Two studies reported reductions in sedentary behaviour (e.g., minutes or hours spent sitting) [83,115]. One of these studies was RCT [115], one was an uncontrolled study [83].

Smoking
Only three studies reported smoking behaviour as an outcome [127,142,143]. Studies with smoking behaviour as an outcome included smoking education-based interventions [142,143], or PA combined with healthy eating education [127]. The two interventions that used education reported significant and positive effects on smoking behaviour (i.e., number of cigarettes smoked, number of people who stopped smoking, behaviour change stage). These two studies included one non-randomised controlled study [142] and one uncontrolled study [143]. The intervention that did not demonstrate a significant change in smoking behaviour was an uncontrolled study, which was focused on PA and healthy eating education [127].
Another 17 studies [59,61,83,90,91,104,112,121,130,135,145,155,160,168,169,171,190] did not report any improvements in stress. Of these 17 studies, seven were RCTs [59,91,130,135,155,169,190], four were non-randomised controlled studies [61,145,160,168] and six were uncontrolled studies [83,90,104,112,121,171]). Two of these studies reported an increase in stress following intervention [59,90]. These two studies included one RCT [59] and one non-controlled study [90] with applied cognitive-behavioural therapy (CBT) and narrative training as their modes of intervention. The interventions that were not successful in improving measures of stress used multimodal interventions, PA, education, mindfulness, and relaxation-based (art, massage) interventions.

Depression and Anxiety
Thirty-two studies measured depression and/or anxiety as an outcome. A significant decrease in depression and/or anxiety was reported in 19 studies (six studies reported decrease in solely depressive symptoms [60,93,97,118,139,177], seven in solely anxiety [164,169,171,174,178,182,187], and six in both depression and anxiety [95,143,156,157,172,176]). Of the 19 studies reporting improvements in depression and/or anxiety, 10 were RCTs [60,118,139,157,169,172,174,176,178,182], two were controlled studies [93,177] and seven were uncontrolled studies [95,97,143,156,164,171,187]. In two studies measuring both depressive symptoms and anxiety [97,139], depressive symptoms improved but anxiety did not.

Mindfulness
Only five studies reported an improvement in mindfulness following intervention [125,149,153,156,158]. Of these, one used RCT design [125], two were controlled studies [149,158] and two were uncontrolled studies [153,156]. A further four studies showed no improvements in mindfulness [91,145,148,163]. Of the four studies reporting no improvements in mindfulness, one was an RCT [91], one was a controlled study [145] and two were uncontrolled studies [148,163]. The studies that showed positive effects relied on various forms of mindfulness training [149,158], or mindfulness training as part of the intervention (with education [153,156], or yoga [125]). The studies reporting no changes in mindfulness reported interventions based on mindfulness [145,148], meditation [163] or stress education [91].
None of the included studies reported intervention outcomes regarding early retirement/intentions, staff retention rates, or staff turnover rates.

Success Rate of the Interventions
The success rate of interventions in improving the outcomes presented above is displayed in Table 2, as a percentage of included studies that measured each outcome. The highest success rate was for diet and nutrition interventions, followed closely by body composition, PA and stress/coping (all with above 70% of studies reporting at least some improvement). However, the evidence stemming from RCTs only is not clear for majority of the outcomes. Only body composition and stress coping seem to have strong RCT-based evidence for their effectiveness (in bold, Table 2).

Unintended Intervention Effects
Only ten studies (7%) reported no significant improvements in any measured outcomes following intervention [57,59,64,84,91,111,121,145,155,189]. Of these, two reported unexpected negative intervention effects, specifically, increases in burnout (EE, DP) [84] and stress [59]. Three other studies found significant improvements to some of their measured outcomes, but alongside positive outcomes they reported negative outcomes on other measures, such as increases in stress [90], a decrease in PA [113] and a decrease in emotional intelligence [73].

Interventions with No Significant Positive Effect
Of the 10 studies that did not report a positive change in any outcome measure following intervention, five were RCTs [57,59,91,111,155], three were controlled studies [64,84,145] and two were uncontrolled studies [121,189]. These studies are briefly described below. Noben and colleagues [111] (n = 538) reported an RCT looking at the effects of occupational health screening on work functioning. They compared outcomes between three groups; screening with referral to a physician, screening with referral to e-health resources, and screening with no feedback. All screening conditions showed improved work functioning, although there were no significant differences between groups. Müller et al. [57] (n = 46; RCT) also reported no significant differences between groups (group with education based on SOC model vs. wait-list control group) in wellbeing, work ability and job control. Menzel and Robinson [59] (n = 20; RCT) compared CBT (focused on stress and pain management) with a wait-list control. These authors reported a non-significant trend (p = 0.06) towards pain reduction together with an unexpected significant increase in stress in the CBT group. There were no significant effects for mindfulness, burnout, or stress in a study by Horner and colleagues [145] (n = 43, pre-post controlled) when comparing mindfulness training with a passive control. Hartvigsen et al. [64] (n = 255, pre-post controlled) found no significant change in lower back pain when comparing a 2-year education intervention (body mechanics, lifting techniques) with a group that attended a single instructional meeting. Similarly, Freitas et al. [121] (n = 21, non-controlled) described no significant quantitative changes in anxiety, depression, burnout or job stress when comparing pre and post scores of a group that attended a 10-min PA workplace intervention, five times a week, for three months. There were also no significant changes in outcomes measured in a study conducted by Chesak et al. [91] (n = 40, pre-post controlled), where stress, mindfulness, anxiety and resilience were compared between a group that attended two education meetings, and a group that attended a single lecture on stress. No positive effects for burnout, depression or stress were reported in a study that compared an eight-week mindfulness course with a passive control [155] (n = 45; RCT). Also, a study reporting the effects of telephone support groups on stress, coping, job satisfaction and burnout, demonstrated no significant changes in outcome measures in this pre-post non-controlled study [189] (study 2; n = 15). Lastly, Le Blanc et al. [84] (n = 304, pre-post controlled) reported an increase in burnout when compared to baseline, after group sessions devoted to forming stress reduction plans within a nursing team. It should be highlighted that this increase in burnout was smaller than that reported by the control group.
A closer analysis of these studies suggests that null findings may have occurred due to the lack of complete data sets (as can be seen above), or due to a degree of similarity between intervention and control groups [111]. The only study for which this was not the case is the study of Hartvigsen and colleagues [64]. Their results may be partly explained by, in the words of the authors themselves, the fact that "the large number of teaching sessions may have increased awareness of back problems and in fact augmented the problem in the intervention group" (p. 16).

Dropout Rates
Of all the included studies, 17 (of which seven were RCTs) did not provide clear information on dropout rate. The remaining studies reported attrition rates ranging from 0% to 75%, with a mean of 18% (SD = 16%). The study with the highest dropout rate [61] used a 3-month follow-up, where the questionnaires were left in a staff room for two months, and thus problems with matching data occurred. The mean dropout rate shows that on average, data collection was completed with 82% of participants, thus results were not likely to have been strongly affected by attrition bias. The five studies that reported the highest attrition rates (>50%) were relatively long-term interventions (i.e., multiple sessions over multiple weeks/months; focused on stress coping, aromatherapy massage, or workplace PA; [94,121,171]), were based on one long session with no refresher sessions (e.g., development of self-care plan, [61]), or on two long sessions (i.e., learning stress symptoms and coping methods [95]).

Risk of Bias Results
All the included studies (n = 136) were independently assessed for risk of bias by two reviewers (NSt, HB) with an initial agreement rate of 97.7%. A third reviewer (EK) independently assessed a 25% subsample. Disagreements were resolved by discussion between the reviewers to reach a consensus (i.e., where there was a disagreement the reviewers referred to the Cochrane Handbook, especially the definitions and examples for the bias assessment; and agreed an outcome that the most closely matched that guide). The Cochrane Handbook classification guide was followed, with reviewers assigning high, unclear or low risk level to studies in terms of six types of bias: (i) selection (random sequence generation, allocation concealment), (ii) performance (blinding of participants and personnel), (iii) detection (blinding of outcome assessment), (iv) attrition (incomplete outcome data), (v) reporting (selective reporting) and (vi) other bias.
The 'other bias' category was predominantly utilised in the current project to judge the adequateness of the sample size (here, n = 30 per condition was used as an adequate size threshold; as suggested by other authors e.g., [191,192]. It was also used to judge other aspects that may have influenced the data (such as contamination between conditions, etc.).
The results of the risk of bias analysis for all studies are displayed in Figure 2. In single group studies, blinding, randomisation and allocation concealment is not possible and therefore these studies were assessed to be at a high risk of bias in these categories. Amongst all the included studies, the highest proportion of bias was related to insufficient blinding of participants and/or personnel (111/136 studies). Other risks included lack of random sequence generation (74/136 high risk), insufficient or no allocation concealment (72/136 high risk), lack of blinding of outcome assessment (68/136 high risk), other sources of bias (64/136 high risk) and incomplete outcome data (52/136 high risk). The lowest proportion of studies with a high risk of bias was recorded for selective reporting bias (129/136 low risk). A significant proportion of studies did not adequately describe the process for collection of outcome assessment, resulting in unclear risk of bias for 54/136 studies. A high number of studies had limited reporting of allocation concealment and random sequence generation. Across all ratings, approximately 37% (352/945) of all risk ratings were low (46.6% was high, 16.4% unclear). Due to the high number of single group studies included in the review, across the seven categories of bias used only 17 studies (12.6%) were able to fulfil five or more low risk ratings, whereas only three studies (2%) reported blinding of both the personnel/participants and outcome assessment. It needs noting, however, that there is a high percentage of non-controlled one group studies in the current systematic review, which affects the risk of bias results, as for such studies blinding, randomisation and allocation concealment is not possible, and thus was assessed as presenting high risk.

Quality Assessment
All studies were evaluated for methodological quality, with the use of CONSORT [52] (for RCTs) or TREND [193] (for non-randomised studies) checklists (see Table S1). Quality varied across the included studies, with the lowest score of 7.5 (out of 23) being evaluated for a quasi RCT [101], followed by 8.5 (out of 20) for a controlled study [102], and 8 (out of 18) for a pre-post uncontrolled study [180]. There were only two studies that achieved the highest possible quality rating (both were RCTs: [62,138]); none of the pre-post controlled or the pre-post non-controlled achieved the full quality score. On average, the pre-post non-controlled studies scored 12.96 quality points, whereas

Quality Assessment
All studies were evaluated for methodological quality, with the use of CONSORT [52] (for RCTs) or TREND [193] (for non-randomised studies) checklists (see Table S1). Quality varied across the included studies, with the lowest score of 7.5 (out of 23) being evaluated for a quasi RCT [101], followed by 8.5 (out of 20) for a controlled study [102], and 8 (out of 18) for a pre-post uncontrolled study [180]. There were only two studies that achieved the highest possible quality rating (both were RCTs: [62,138]); none of the pre-post controlled or the pre-post non-controlled achieved the full quality score. On average, the pre-post non-controlled studies scored 12.96 quality points, whereas the controlled studies scored 14.18. In comparison, RCTs earned on average 14.50 points. This suggests that the current literature has a high proportion of studies with low quality reporting, although there is a small number of publications that can be used as a reference point for reporting style.

Discussion
This systematic review aimed to synthesise a substantial pool of evidence on the effects of lifestyle interventions on the physical and mental health of nurses, in addition to work-related outcomes. A total of 136 relevant studies were identified involving 16,129 participants who met all the inclusion criteria and none of the exclusion criteria.
The review suggests that interventions aimed at improving nutrition amongst nurses commonly result in improved outcomes, especially when the interventions are education-based. However, it is important to note that the total number of studies including nutrition outcomes was very limited (only nine studies, with only two RCTs). This was not the case, however, for studies measuring body composition outcomes, which also had a high level of success, but more studies in this category utilised an RCT design. Of these, 10 RCTs showed improvement, and none of the RCTs showed no improvement, which provides a particularly clear and promising finding for the influence of lifestyle interventions on indices of body composition. Similarly, stress was measured in a high number of RCTs (22), with 16 showing significant improvements on this outcome. This also supports the credibility of lifestyle interventions for reducing stress in nurses. Physical activity was also a somewhat successfully improved outcome, with four RCT studies included in that category, including three that showed improved PA outcomes. Based on the above, we suggest that there is sufficient evidence to recommend the application of lifestyle interventions targeting body composition, stress, diet, and PA. However, more RCTs are required to provide additional higher quality evidence, particularly for diet and PA.
We found some evidence for improvements in smoking behaviour (66.7%), self-efficacy (66.7%), physical health (64.7%), wellbeing/QoL (60.9%), burnout/CF (60.0%), and depression/anxiety (59.4%), although the evidence for these is not so strong. This is partly due to the lower success rate of the studies reporting on these outcomes (than for body composition, stress, diet, and PA), and also because many of the studies reporting on these outcomes had lower quality designs or demonstrated quite ambiguous RCT-based evidence. First, none of the studies reporting on smoking behaviour had an RCT design. Only three RCTs assessed self-efficacy but one of these studies showed no improvement. Similarly, in terms of physical health, wellbeing/QoL and burnout, the results provided by the RCTs were mixed (i.e., four RCTs reported improvement (RCT+), while three RCTs reported no improvement (RCT-) for physical health, for wellbeing/QoL five were RCT+, six were RCT-, whereas for burnout four were RCT+, and four were RCT-). This suggests that more high-quality research is needed measuring these outcomes, and we need to better understand what moderates the effectiveness of these interventions. Although depression/anxiety had a lower success rate across all studies measuring this outcome (59.4%), when RCTs only were considered, the findings were more promising since there were 10 RCTs reporting improvements in depression/anxiety, although six RCTs showed no change.
Finally, it is even more difficult to clearly describe the impact of interventions targeting mindfulness and work-related factors. These were the two outcomes with the lowest success rate across all studies measuring these outcomes (55.6%, and 50.0% of respective studies reported improvements on at least one relevant measure). To add to this, the RCT-based results also provided very ambiguous findings, where mindfulness was improved in one RCT, but showed no change in another, whereas work-related outcomes improved in eight RCTs, but did not show any change in another nine RCTs. Further, the type of intervention leading to improvements in these outcomes cannot be delineated. This highlights the need to more closely consider specific intervention aspects and their efficacy within targeted samples. Future research could conduct a meta-analysis of a narrower range of interventions and outcomes in order to address these questions.
Work-related outcomes, mindfulness, depression/anxiety, burnout/CF, and wellbeing/quality of life constructs were those outcomes that appeared to be less amenable to change with lifestyle intervention. It is important to consider the potential explanations for this. One potential barrier to modifying these factors is that they are complex outcomes and are influenced by multiple factors that may be more challenging to control through workplace intervention. For instance, work-related outcomes are likely to be influenced by factors that are not being targeted in lifestyle interventions, like the organisational environment and specific job stressors [196] such as work context, demands, pressure, or the perception of one's role at work, etc. One approach that produced positive results targeted empowerment, civility and trust in management [197], however it was focused on creating a supportive and empowering work-environment rather than making the individual more resilient.
Furthermore, mindfulness is a particular skill that requires intensive training to be improved. Studying mindfulness presents many issues (e.g., potential for an initial increase in distress, [198]), and so adapting brief mindfulness interventions to workplace settings brings further challenges [199]. Future studies will need to consider these aspects. Also, researchers might consider using a recently developed framework for reporting mindfulness-based interventions [200].
Similarly, depression/anxiety, as well as burnout/CF, especially when clinically significant, might require individual professional mental help or counselling to generate positive outcomes, rather than a workplace lifestyle intervention. This will likely explain why the lifestyle interventions reviewed here were less likely to produce positive improvements for these mental health outcomes. Previous work has suggested that for burnout and CF, changes in organisational culture might be particularly important [201]. It has also been shown that interventions incorporating both personal and organisational aspects have more long-term effects for burnout [202]. Additionally, there is systematic review evidence suggesting that counselling is effective in alleviating psychological problems related to work [203]. Nonetheless, more holistic approaches (incorporating reducing work-related risk factors for mental health, developing positive aspects of work and employees, and addressing mental health problems irrespective of their cause) have been recently advocated [204]. Such initiatives need exploring more, as burnout has been identified as a leading cause of work-related mental health issues (e.g., [202]).
Lastly, wellbeing and quality of life are complex multidimensional concepts, which have been acknowledged as being difficult to change (e.g., [205]). Such factors may take a significant length of time to change; thus, short-term modifications to one's health or lifestyle behaviour may not have immediate effects upon an individual's overall perception of their life or general wellbeing, as such behaviours may need to be sustained for much longer periods to influence the more fundamental nature of wellbeing and quality of life. The majority of studies measured outcomes immediately post-intervention and did not assess outcomes in the medium or longer term when any changes to these outcomes may be more likely to have taken effect. It might also be the case that nurses who work in a particularly demanding work environment (with long shifts, problems with understaffing and over-utilisation of the health systems) do not perceive small individual changes (e.g., to health behaviours or psychological health) as salient enough to improve their overall quality of life. Given that shift-working nurses report lower quality of life than the general population [206,207], it is not surprising that improving their quality of life might be difficult to change at an individual level, and might require more complex changes at the organisational level. It is also true that quality of life as a concept has often been misunderstood in healthcare research [208], which might have affected the results presented here. In summary, all five of these outcomes might benefit from complex interventions that take a more holistic approach and pay attention to the conceptualisation and measurement issues. The variability in the measurement scales that were used to assess these constructs (as presented in the results section) provides additional evidence, both for the lack of consistency in measurement approaches, and lack of a consensus as to how to best measure these outcomes.

Results Specific to RCTs
Despite the inclusion of both RCTs and non-randomised studies in this review, results from RCTs only did not vary considerably from the findings based on the wider spectrum of the evidence reviewed. Similar types of interventions resulted in improvements, or no effects. Likewise, results relating to organisational outcomes showed little amenability to change. It is important to highlight that education-based RCTs were the only intervention type not to show any significant improvements in any outcome, whilst smoking-focused RCTs demonstrated only short-term effects. It seems likely that the provision of education-only might be the least beneficial to nurses and their organisations, and studies reporting on interventions targeting smoking behaviour are too few to draw meaningful conclusions. This is in agreement with the psychological literature, which warns that merely possessing knowledge does not necessarily lead to change in behaviour (e.g., attitude-behaviour gap; e.g., [209][210][211][212]. Moreover, it corroborates other findings from the nursing literature, suggesting that nurses, despite their training, knowledge and skills in health promotion practice, often do not practice what they preach (e.g., [7,8,42]). Thus, it might be crucial, if relying on education-based interventions, to offer them within a more multimodal context, which also focuses on aspects of behaviour change.

Quality Concerns
Many of the included studies were assessed as presenting low methodological quality which may have limited their ability to uncover intervention effects. While there were many randomised controlled trials included in this review (RCTs; n = 54; two with crossover design), there was a higher number of non-randomised studies (n = 82) that had pre-post designs (n = 31 with a control group, n = 51 without a control group). It is difficult to determine the effectiveness of interventions that have been tested using non-randomised designs, and these studies by their nature had higher risk of bias. The methodological quality of the studies varied substantially (including those tested in RCT designs), and there was a high number of studies that did not report enough information to make an assessment of quality in certain areas (see Figure 2 e.g., 'blinding outcome assessment' was unclear in a large number of studies). Future studies should adhere to CONSORT guidelines when designing, running and reporting intervention studies.
The main methodological concerns observed in the included studies were (i) absence of a control group or inclusion of a 'non-active' comparator group, which may obscure the actual effectiveness levels of the interventions; (ii) drop-out rate, with some studies reporting very high attrition from the research study, which may limit the true effect of an intervention; (iii) use of voluntary and small samples; with many studies not being randomised, and/or having very limited numbers of participants. In addition, only a very small sample of studies looked into long-term effects (>6 months) of the interventions, which limits the interpretation of their effectiveness after the intervention period is over.

Review Limitations
Efforts were made to minimise limitations such as the inclusion of risk of bias analysis and presentation of the quality assessments of the included studies. There are still, however, certain limitations that need to be taken into account when interpreting the result presented. The review was limited to articles published in the English language. Searches were undertaken by a single researcher, although there were two researchers involved in the overall process. There is a possibility therefore that some relevant literature was missed due to human error, or due to its publication in a language other than English. Studies with non-controlled designs were included in the review and assessed using the same stringent criteria, which increases the proportion of studies assessed as having a higher risk of bias. The fact that only 54/136 intervention studies utilised an RCT design needs to be taken into consideration when interpreting our findings, although to account for this, we have presented results separately for RCTs. Nevertheless, given the lack of recent reviews on lifestyle interventions and therefore the unknown scope and quality of evidence in this area, it was deemed important to employ broad inclusion criteria to capture details of relevant intervention studies and highlight the vast number of studies published in this field with low quality research designs and reporting. Our search criteria generated a large number of articles reporting the outcomes of a diverse range of interventions. However, it is possible that some articles were missed where particular search terms were not in our criteria (e.g., we did not specify 'back pain' or 'musculoskeletal' interventions in our search terms, although the review identified some articles with interventions in this area). There may be scope for a review focusing specifically on musculoskeletal interventions in nurses and/or other healthcare professionals. It is possible that our results were affected by publication bias, as null findings are less likely to be published. Lastly, a meta-analytical approach was not considered feasible due to the exceptionally high heterogeneity of outcomes and intervention modes used in the included studies.

Future Directions
One of the main issues identified in this review is the length of the interventions and timing of the post-intervention measurement. As reported above, the most common time frame for interventions was two months. Even though this seems like a considerable length, the habit formation literature suggests that it can take many weeks of daily repetition to establish a habit (e.g., [213,214]), which may not be possible with interventions running on a one time per week basis. Whilst daily home practice was stipulated by many of the interventions, it is not always clear whether this home practice actually occurred. It might be worthwhile to design and test interventions that maximise the 'dosage' of intervention by offering additional resources such as support group elements, or mobile support or reminders, in addition to encouraging and recording daily behaviours. For many studies, due to a low quality of reporting, it was not possible to determine the influence of factors such as intervention fidelity (relating to engagement and delivery), adherence and actual versus intended dosage, or attrition. Many studies did not report any theoretical framework or model for the intervention and the majority of studies did not use formal reporting guidelines in their publications.

Conclusions
In summary, this systematic review provides a comprehensive synthesis of the literature investigating workplace lifestyle interventions aimed at improving individual physical and mental health, and/or organisational outcomes in working-age nurses. This review highlights that there are significant methodological limitations in the published literature, with low quality of reporting regarding mostly interventions and research processes. This needs to be addressed in future studies with the increased use of standardised tools and checklists to inform intervention design and reporting. Tentative conclusions are drawn from a vast pool of research with mixed designs, heterogeneity of outcome measures, with a significantly smaller pool of higher quality RCT evidence. Overall, this review suggests that workplace lifestyle interventions targeting nurses are likely to have positive effects on a range of individual health and lifestyle factors such as diet and nutrition, body composition, PA and job-related stress. Findings for mindfulness, wellbeing/QoL, burnout/CF, depression/anxiety and work-related outcomes are more mixed, and may require novel, or more complex organisational approaches. Similar work needs to be undertaken among other groups of healthcare professionals, such as medics, whose health may have direct implications for the healthcare of their patients.