Effects on exercise and fitness outcomes of workplace physical activity interventions targeting older employees: A systematic review and meta-analysis

Background: “Active Ageing” policy to delay retirement mean that maintaining the health and fitness of older employees has become increasingly important. This systematic review summarises the characteristics and effect on exercise and fitness outcomes of workplace physical activity (PA) interventions targeting older employees. Methods: Five online databases were searched from inception to December 2018. Eligible studies were of any experimental design, included employees aged ≥50 years, had PA as an intervention component and reported PA-related outcomes. Results: Titles and abstracts of 7470 records were screened and 16 unique interventions were included (3,215 participants). Eleven studies were RCTs. Six interventions targeted multiple risk factors (n=1,586) involving screening for cardiovascular disease risk factors, but had a non-specific description of the PA intervention. Four interventions targeted nutrition and PA (n=1,127), and six intervention (n=195) focused only on PA. Seven interventions were short –term (<15 weeks), six interventions lasted 6-9 months and three interventions were long-term (10-12 months). Interventions overwhelmingly targeted aerobic PA compared to strength, balance and flexibility. No studies involved screening for falls/injury risk. Computation of effect sizes (ES) was only possible in a maximum of three RCTs per outcome. ESs were statistically non-significant for all outcomes. ESs were medium for PA behaviour (ES=0.25 95% CI: -0.07 to 0.56), muscle strength (ES=0.27, 95%CI: -0.26- 0.80), cardiorespiratory fitness (ES=0.28, 95%CI: -22 to 0.78) and flexibility (ES=0.50, 95% CI: -0.04- 1.05) and large for balance (ES=1.29, 95% CI: -0.56- 3.15). GRADE criteria-rated quality of evidence were ‘low’ due to high risk of bias, imprecision and inconsistency. Conclusions: The effect of workplace interventions for improving fitness outcomes of older employees is uncertain. There is a need for high-quality PA interventions that takes into


Background
Population ageing poses individual, social, economic and political challenges and is predicted to accelerate even further in the 21 st century [1]. The WHO's "Active Ageing" concept is a leading global policy strategy for successful ageing [2], particularly influencing retirement policies towards maintaining engagement with paid employment. In Europe and other developed countries strategies have been implemented to promote economic activities among aged workers, including incentives for late retirement, penalties for early retirement, and an increase in the age of mandatory retirement [3].
Delayed retirement means that more employees beyond the age of 55 years will remain in the workforce [4].
Because of age-related decline, an older workforce requires special considerations regarding occupational health, safety and productivity [5]. An in-depth review by Crawford et al. [6] identified a number of physiological and psychological differences between older workers and their younger counterparts, including reduced muscle strength, endurance, trunk flexibility, balance, aerobic capacity (particularly in women), tolerance to heat, increased anthropometric risks, psychological exhaustion, anxiety and depression.
Further, older employees suffer from greater prevalence of comorbidities, and more musculoskeletal problems, sickness-related absence and fatal injuries [6].
It is well established that regular physical activity (PA) is effective in attenuating agerelated physiological decline in all body systems, and preventing and managing many agerelated chronic and musculoskeletal conditions [7][8][9]. Hence, maintaining good PA habits is likely to benefit older employees, in terms of health as well as work productivity and safety. The workplace has long been considered a good setting for health promotion as most adults attend work for most of their waking hours. Hence, many PA interventions have been conducted in the workplace, with a first systematic review of their effectiveness published in 1998 [10]. However, we did not identify any systematic reviews of workplace PA interventions that addressed the specific needs of older employees in terms of their physiological or safety needs. Interventions that specifically promote muscle mass, strength, and balance are important for the prevention of sarcopenia, functional disabilities, falls and injuries, and are now part of PA recommendations for older adults worldwide [11]. One recent systematic review synthesised the evidence on health promotion interventions targeting older employees [12]. However, this review examined a variety of health promotion initiatives and may have missed interventions that focused on PA, considering that only two keywords, "fitness" and "capacity", were used to capture PA promotion. Given PA recommendations for older adults include a range of physical outcomes relevant to health beyond cardiorespiratory fitness/capacity, such as muscle strength, balance and flexibility [13,14], it is important to include broader terms when reviewing health promotion programs. Furthermore, the main outcomes included in the review by Poscia et al. were health status, wellbeing and work productivity, not PA measures. Thus, the effectiveness of workplace interventions on PA and other fitness domains of older employees is uncertain.
The aims of this systematic review were to: i) identify and characterise workplace PA interventions delivered to employees aged ≥50 years; ii) assess the methodological quality of the studies; iii) assess the effect of interventions on PA and fitness outcomes highlighted in PA recommendations for older adults.

Methods
This review followed guidance published by the Centre for Reviews and Dissemination, the Cochrane Collaboration [15], and PRISMA guidelines [16] (see Supplementary file 3) and was prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO; registration number CRD42018084863 available at https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=84863)

Inclusion and exclusion criteria
We included studies that involved PA interventions delivered at workplaces to employees aged ≥50 years. Studies that included a wider age range were eligible for inclusion if the mean age of participants was at least 50 years. Interventions could include those related to aerobic PA (e.g., walking), stretching, balance, muscle strength, yoga, tai chi, Pilates, gym workout, sport or any other form of PA/exercise. Studies that targeted other health behaviours in addition to PA were only included if PA was one of the key components of the intervention and a PA outcome was reported. Studies with any experimental design (pre-post with or without a comparison group, nonrandomised trials, RCTs and cluster RCTs) were included because of the small number of randomised controlled trials (RCT) anticipated based on a previous review [12].
Studies were excluded if the PA intervention targeted older employees, but was not delivered in the workplace (e.g., community, health practice), or if the interventions targeted the whole workplace with a wider age range and did not include specific PA outcomes for the sub-group of older employees.

Search strategies
Five databases were systematically searched from inception to the end of 2018: Medline, PreMedline, PsycInfo, CINAHL, and the Cochrane Controlled Register of Trials (CENTRAL).
Specific search strategies were developed for each database, using a combination of text terms and subject headings where applicable. All electronic searches included MeSH terms and keywords for workplace (e.g. occupations, worksite), PA (e.g., walking, postural balance, muscle strength) and intervention (e.g., health promotion, program, RCT) and were limited to humans, and age (PsycInfo 40+ all other 45+). The MeSH terms and keywords are presented in Appendix 1. One reviewer (JS) screened the titles and abstracts to identify potential papers for inclusion as well as running forward and backward citation tracking of potential studies selected for full text review. Two reviewers (JS, DM) independently conducted the final selection of papers based on full text. Disagreements were discussed and resolved by consensus.

Data extraction
Three data extraction formats were used: A table summarising the included studies was created using the subheadings source, study design, population characteristics, study duration, type of intervention, comparison group, and outcomes. A second table to address research aim 1 described characteristics of each intervention in terms of recruitment, content, intervention deliverer and process outcomes based on data extracted by DM and SM. A third table summarised the quality of the included papers based on the Cochrane risk of bias tool for RCTs and additional domains from the Robins tool for non-randomised studies with a comparison group [15]. The domains included randomisation sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data and selective outcome reporting. The first two domains were only assessed in RCTs. In non-randomised studies with a control group the additional domains of confounding and selection bias were assessed.

Quality assessments and analysis
Each domain was assessed independently by two authors (MD, FS) who assigned a judgment of either 'low', 'high' or 'unclear' risk of bias. Disagreement was resolved by consensus or, where needed, by consultation with a third reviewer (DM). We used Comprehensive Meta-Analysis software (Version 2, Biostat, Englewood, New Jersey, USA) to conduct random effects meta-analyses for each outcome where sufficient homogeneity between studies would allow for meaningful quantitative synthesis. The standardised mean difference (SMD) (Hedges' g) was calculated for each meta-analysis, standardised by postscore SD (or its estimate) and calculated using the premean and postmean and SD or, when this was unavailable, the mean change score. Effect sizes were categorised as small (0.20), medium (> 0.20-0.5) or large (0.8 or greater) [17]. For trials of effectiveness of change in moderate-to-vigorous minutes of PA we used data from accelerometers when possible, otherwise self-reported data. When energy expenditure was reported as kilocalories per week, we converted the value to minutes engaged in at least moderateintensity activity using the population mean body weight and 4 Metabolic Equivalents for Task (METs) as at least moderate-intensity exercise. Statistical heterogeneity was determined by the I 2 and χ 2 tests. Finally, we used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach to rate the overall quality of evidence for each outcome from 'very low' to 'high' [18,19], including a narrative summary of findings for outcomes without meta-analysis [19]. GRADE ratings are based on the domains of risk of bias in included studies, indirectness of evidence, imprecision, inconsistency and likelihood of publication bias.

Study selections and characteristics
Our search resulted in 7470 records which were screened by title and abstract. Of these, 59 were further assessed for eligibility based on full text ( Figure 1). Seventeen studies met the inclusion criteria, however, two publications referred to the same intervention [20,21], resulting in 16 unique interventions.

Participants
In total, 3  Interventions overwhelmingly focused on aerobic PA compared to other domains relevant for older people such as balance, strength and flexibility. In all multiple risk factor interventions some form of 'risk assessment' was applied before the targeted behaviour was chosen by or for participants, such as a nurses' check-up for CVD risks [34] or selfappraisal of risk using questionnaires [22,24], or assessment carried out during the first educational session [33]. However, none of the risk factor assessments considered screening risk factors specific to older people such as falls and injuries. In addition, nearly all of these multiple risk factor studies used a generic description of the PA intervention, such as 'exercise on their own' [31], 'exercise training' [25], referrals to exercise classes [34], or choosing PA goals [22,24]. In the multiple risk factors studies by Hughes et al.

PA types and fitness domain
[22] and Low et al.
[25] the interventions also offered on-site self-managed activities such as walking groups or using worksite facilities. The dual PA and nutrition interventions focused on aerobic regimens such as steps accumulation [21, 32,35], counselling to shape PA goals [36] or an aerobic workout on site [21]. The study by Strijk et al. [21] also offered yoga sessions, an activity that is recognised as multi-dimensional and enhances muscle strength, balance, and mobility [37]. The PA-only interventions focused either on aerobic regimens such as use of a treadmill workstation [29], self-managed aerobic walking [27] or Nordic walking [26] as well as non-aerobic fitness domains, such as multidimensional PA (tai-chi) [38], and progressive strength and balance exercises [30]. Nordic walking is primarily aerobic activity that may increase upper body strength and flexibility, hence, these domains were marked as well in Table 2 [26].

Delivery modes
Three delivery modes were noted ( Table 2). In seven studies the PA was delivered on site during working hours [21, 23, 25, 26, 29, 30], or on site but before working hours [28], in five studies participants were counselled in the workplace in face-to-face meetings [31][32][33], or by telephone calls (COACH-arm) [22], or a combination of both [36], but the PA sessions took place in employees' discretionary time. In four studies the PA was selfmanaged with little or no supervision [24,27,34,35]. The study by Hughes et al. [22] had two intervention arms; one involved personal contact with students trained in behaviour change coaching ("COACH"), while the other arm was self-managed through a web-based intervention ("RealAge") with no other contact [22]. The study by Cook et al. [24] was also a web-based self-managed intervention (HealthyPast50).

Intervention reach
Program reach (i.e., proportion of study population recruited to the trials) varied substantially. When recruitment methods involved a targeted strategy (TS) based on prior 'screening or eligibility criteria', a strategy reported in five studies [27,[32][33][34]36], intervention reach was highly variable. In the Japanese study 92% of the targeted eligible employees were recruited [32], while two interventions in the UK [27, 34] the Taiwanese study [33] and the Italian study [36] using the same strategy achieved much lower recruitment rates (ranging from 20% to 35%). When the method of recruitment was workplace advertisement (WA, eight studies) [20, 22-26, 28, 31], reach was hard to estimate due to lack of information, with the exception of two studies: Strijk 30,35], the proportion of targeted employees who attended was not reported, nor was the proportion who were eligible and agreed to participate.

Studies' quality
Included RCTs and non-randomised studies were generally of poor quality (Supplementary Table 2). The randomisation process and concealment of allocation was mostly poorly ]. An additional major concern was the risk of selection bias due to attrition given that most trials (78%) had incomplete outcome data, and intention-to-treat analysis was done in only four RCTs [21, 24, 28, 29]. The five included studies that were not RCTs, had a high risk of bias across almost all domains. Table 3  and Nordic walking [26] and found a moderate sized effect (0.50) without heterogeneity, but that was also statistically non-significant (95% CI -0.04 to 1.05). GRADE ratings of the certainty or quality of evidence for each outcome were low to very low mostly due to the high risk of bias of included studies, high heterogeneity (inconsistency), and imprecision (all pooled CIs crossed 0 and were statistically non-significant).

Discussion
To our knowledge this is the first systematic review and meta-analysis of workplace PA interventions specifically designed for older employees. Although few interventions demonstrated significant effect on one fitness domain, collectively we found no definitive evidence of effectiveness as pooled effects remained non-significant. Methodological quality was generally poor for most of the included studies. In addition, we did not identify any program that fully addressed the WHO's broader PA recommendations, which highlight the need to incorporate balance exercises (≥3 times a week) and muscle strengthening (≥2 times a week) along with the aerobic recommendations [39]. Only one study examined whether participants met aerobic recommendations and only two trials targeted balance or muscle strength as main outcomes. Therefore, there is a key mismatch between the aims and outcome measures of interventions designed by the public health research community and those recommended as ideal for older adults. Additionally, this review suggests that the initial level of participation (i.e., intervention reach) was generally low, which was also noted in another systematic review of workplace health promotion interventions [40].
High heterogeneity makes the pooled estimates difficult to interpret as it is unclear whether population, intervention implementation or methodological quality issues underlie the heterogeneous results across studies. For example, two cluster RCTs reported changes at 6 months in predicted maximal oxygen uptake; the Japanese study reported a significant medium effect size based on a sub-maximal stationary bike test [32] and the Dutch study [21] reported a small non-significant effect on the 2km walking test. The difference between these two measures could explain the high heterogeneity as well as differences related to completion rate; high in the Japanese study (85% and 93%) and low in the Dutch study (56% and 66%) in control and intervention, respectively.
Multiple risk factor interventions incorporated screening for cardiovascular risk factors, with no focus on other age-related health problems, such as falls, which is preventable by appropriate exercise. Such screening could be carried out either through a single selfreport question (e.g., history of falls) or through assessment of physiological fall risk (e.g., poor balance, impaired leg strength). Risk of falls and related injuries is an important safety issue in older workers because the consequences of falls and injuries are greater in older employees compared to their younger counterparts [6]. Further, despite emerging evidence that resistance training is as beneficial as aerobic training for the prevention and management of cardio-metabolic risks [41,42], none of the multiple risk factor interventions explicitly included strength training to reduce CVD/metabolic risk. Strength training was promoted to employees with osteoarthritis, in accordance with clinical guidelines [28], but only one intervention specifically targeted muscle strength to generally healthy older employees [30], supporting the recent claim that strength training has been a neglected aspect of PA promotion [43]. From the mid-forties onwards, adults lose 14% to 16% of their isokinetic leg strength each decade [44], which is a greater decline than that observed in cardiovascular capacity (~10% per decade) [45].
Considering that employees aged 50 years and over are likely to stay at work for another decade or more, neglecting this domain is unwarranted.
The included PA programs varied substantially in terms of the PA modality, delivery methods and duration, but none stood out as a clearly superior strategy for older employees. Compared with self-managed programs carried out in participants' discretionary time, on-site supervised sessions during the workday [21, 23, 26, 30] tended to demonstrate high retention rates (71% to 100%) and good compliance. However, onsite PA programs were usually shorter in duration below 15 weeks. On-site pragmatic approaches such as treadmill workstations, short breaks for balance and strength training or web-based interventions, can be easily integrated into the workday. However, the effectiveness of treadmill workstations was disappointing [29], and the evidence for other approaches was inconsistent [22,24,30]. These interventions should be further tested by high-quality replication studies before translational research can be conducted.
The most alarming finding is the poor quality of the RCTs we identified, with the exception of the study by Bergman et al. [29]. Study quality could be improved by appropriate randomisation, intention to treat analysis, and the use of both objective and subjective measures of PA. For example, most multiple risk factor interventions included objective measures of CVD risks (i.e., blood tests or anthropometrics), but failed to use objective fitness tests or activity trackers. Further, to improve the directness of evidence, the selection of outcome measures should better reflect the fitness dimension(s) that the proposed PA is likely to affect, such as measuring balance and strength outcomes in interventions involving yoga or tai-chi or aerobic fitness in Nordic walking interventions.
Research demonstrates increases in leisure-time PA after transition to retirement compared to during full employment [46,47]. However, a systematic review has found disparities by socioeconomic status (SES); employees with low SES tend to be less active after retirement whereas those with high SES tend to be more active after retirement [47].
Focus groups with older retirees revealed that retirees from lower SES appear to place lower value on the importance of leisure-time PA [48]. Therefore, improving PA whilst in the workforce may be of even greater importance to older people with low SES. In this review we identified three studies that specifically targeted low paid employees in manual occupations (i.e., factory employees) in Japan and the UK [27, 32,34]. Only the study from Japan, which included social and environmental support in addition to behavioural counselling, demonstrated significant positive effects. However, this study was a nonrandomised study with a high risk of bias for almost all domains. Further work is therefore needed to ascertain what types of interventions can most effectively improve PA outcomes in older employees with low SES.

Strengths and limitations
This review has applied guideline-informed, rigorous methods to synthesise and evaluate current evidence around workplace PA interventions for older people.

Ethics approval and consent to participation
Not applicable for systematic review

Competing interests
No competing interests to declare.

Funding
There was no funding for this research.     Low >>ΟΟ The pooled effect size was derived from accelerometer data in intention to treat analysis (2-studies) and one selfreport non-randomised trial and therefore reduce overall quality One addit (Hughes e included to calcula mixed mo interactio study rep betweensignifican