Work-related musculoskeletal disorders (WMSDs) such as low back pain and shoulder pain constitute a substantial problem for individuals, workplaces, and societies [1-3]. At the individual level, WMSDs increase risk of poor health, sick leave, and premature exit from the labor market [1,4,5]. For workplaces, workers with WMSDs have lower workability and are more likely to have long-term sickness absence [6,7]. For the societies, WMSDs lead to substantial expenses regarding treatment, lost production, and sickness absence [1,8]. Heavy physical work is a known risk factor for developing WMSDs [9,10]. In particular, heavy lifting, pushing or pulling, and working in awkward postures have been associated with low back pain [11] and sickness absence [4,12,13]. Construction work consists of a high degree of heavy physical work [14,15]. Consequently, a reduction of the physical workload to promote sustainable working careers [16] in construction work should have high priority. Moreover, a systematic review highlighted an urgent need for interventions focusing on reducing WMSDs in construction workers [17]. In addition, most field studies in the construction industry are based on self-reported measurements [17]. Hence, a more technical approach may enable objective evaluation of the loading and provide better grounds for targeted and effective interventions.

Technological development has made it possible to obtain field measurements with surface electromyography (sEMG) [18,19], kinematics measured with inertial measurement units (IMU) [20-23], or a combination [24]. However, no studies have used sEMG, IMU, and video recordings obtained simultaneously to detect events with excessive physical workload (events) during a working day. Thus, knowledge about these specific events may be an important tool for engaging workers to combat work-related risk factors.

In participatory ergonomics (PE), the workers are involved in the decision processes. Systematic reviews have reported that PE has positive effects on musculoskeletal symptoms [25] and thereby may lead to increased productivity and reduced occupational risk factors [26]. Furthermore, a systematic review has shown that participatory responsibility concerning the identification of risk factors, development of solutions, and implementation is important to succeed in the participatory process [27]. Nevertheless, the evidence for preventing neck-shoulder and low back pain through ergonomics interventions is questionable because the number of randomized controlled trials are limited [28].

This cluster randomized controlled trial investigated whether a PE intervention with technical measurements could reduce the number of events with excessive physical workload during a working day in the construction industry We hypothesized that the PE intervention involving both managers and workers would lead to a reduction in the number of events of excessive physical workload.

This study was a 2-armed, parallel group, single-blinded, cluster randomized controlled trial with allocation concealment performed at construction sites across Denmark from May 2016 to June 2017. Clusters were defined as construction gangs. The organization of construction work, that is, working in construction gangs, was the reason for choosing a cluster design. The intervention consisted of 3 workshops based on individual technical measurements of excessive physical workload. The technical measurements to detect excessive physical workload have previously been validated in controlled laboratory settings [29] and were conducted at baseline and 3 and 6 months’ follow-up.

According to the Helsinki declaration, participants received written and oral information about the purpose and content of the study before signing the informed consent form. The study was approved by the local ethical committee of Frederiksberg and Copenhagen (H-3-2010-062) and registered with the Danish Data Protection Agency (215-57-0074) and ClinicalTrials.gov (NCT02498197). The reporting followed the CONSORT statements for cluster randomized trials [30] and CONSORT eHealth [31] (Multimedia Appendix 1). Design of the effect evaluation and process evaluation have previously been reported [32,33]. This study reports data solely from the effect evaluation.

The inclusion criterion was full-time construction work. The exclusion criteria were life-threatening diseases and hypertension >160/100 mmHg. A total of 9 participants were excluded before the baseline test. Moreover, 80 participants (15 clusters (gangs)) met the inclusion criteria and completed the baseline test. The flow of participant enrollment is illustrated in Figure 1.

The randomization was performed by a researcher who was not involved in data collection (LLA). Block randomization of the construction gangs was chosen for practical reasons and was performed continuously as the baseline tests were completed. The researchers performing the data collection were not aware of the block size or group allocation. Blinding of participants is not possible in behavioral interventions. The data analyst and the statisticians were blinded to group allocation.

The intervention was carried out at gang level and consisted of workshops or reading handouts for the intervention group and control group, respectively.

The control group received handouts about WMSDs [36] and lifting guidelines from The Danish Working Environment Authority [37]. These handouts described the association between WMSDs and the impact on working life, regulations for the prevention of WMSDs, and which precautions should be taken to limit WMDSs [36]. Furthermore, the handouts described the regulations for lifting, pushing and pulling, and the risk of injuries [37].

The sample size was calculated based on the observed changes in the level of muscular activity during a working day in different occupational groups with pronounced lifting [19]. The power calculation showed that 17 participants in each group were needed to demonstrate a reduction of 20% in normalized sEMG assuming an SD of approximately 20% in normalized sEMG between individuals and a type 1 risk of 5% and power of 80%. Due to the cluster design and including an inflation factor of 1.5, 26 participants were required in each group [32]. For generalizability and risk of dropouts, we aimed to recruit 10 construction gangs of 3 to 5 individuals in each group, that is, a total of 80 participants.

t tests assessed possible group differences at baseline. The difference from baseline to follow-up between the intervention and control groups was evaluated using a linear mixed model. The number of events was log-transformed because the residuals were not normally distributed. Factors included in the model were group (intervention and control), time (baseline, first follow-up, and second follow-up), and group-by-time interaction. The analysis was adjusted for the baseline value of the outcome, age, gender, duration of measuring time, mean heart rate, number of steps, and muscle strength. Individual nested in cluster was included as a random factor. Analyses were performed using SAS statistical software (Proc Mixed, SAS version 9.4) according to the intention-to-treat principle, including all participants (n=80) regardless of loss to follow-up. The estimation method was restricted maximum likelihood with degrees of freedom based on the Kenward-Roger approximation. P levels ≤.05 were accepted as statistically significant. Outcomes are reported as within- and between-group least square mean differences with 95% CIs. Furthermore, Fischer exact test was used to test for differences in questions with categorical response variables.

Table 1 shows the baseline characteristics of the participants. Age was higher in the control group compared with the intervention group (P=.02), which was controlled for in the statistics by including age as a covariate. At the first follow-up test, 12 participants dropped out, and 4 participants did not show up for the test. At the second follow-up test, 12 participants dropped out, and 14 participants did not show up for the test. Hence, 42 participants completed the study (Figure 1). All dropouts were included in analyses.

The results showed no group-by-time interaction effect (P=.75) and (P=.51) for the number of events obtained using technical equipment in the unadjusted and adjusted analysis, respectively (Table 2). The results show a within-group difference (time effect) in the number of events from baseline to the first follow-up test (unadjusted, P=.002 and adjusted P=.05 and (unadjusted P <.001 and adjusted P<.001) for the intervention and control group, respectively. Furthermore, a within-group difference was observed from baseline to second follow-up in the unadjusted analysis for the intervention and control group (P<.01 and P<.001), respectively. The exploratory analyses confirmed the results of the primary outcome, that is, no significant group-by-time interaction (Table 2).

The analyses of heart rate and step count showed no group-by-time interaction or within-group difference. However a between-group difference was observed at baseline, first and second follow-up, and at baseline for heart rate (P<.001, P=.049, and P=.003, respectively). This between-group difference was also observed for the step count at baseline (P=.004) but not at the follow-ups. The mean heart rate was 100 (95% CI 96 to 104), 101 (95% CI 97 to 105), and 100 (95% CI 96 to 105) and 91 (95% CI 88 to 95), 95 (95% CI 92 to 99), and 91 (95% CI 87 to 95) bpm for the intervention and control group at baseline, first follow-up, and second follow-up, respectively. The mean number of steps adjusted for length of the working day were 5952 (95% CI 5517 to 6387), 5479 (95% CI 5023 to 5934), and 5980 (95% CI 5372 to 6588) and 5133 (95% CI 4791 to 5475), 5340 (95% CI 4958 to 5722), and 5320 (95% CI 4852 to 5788) steps per day for the intervention and control group at baseline, first follow-up, and second follow-up, respectively.

The results from the other outcomes are presented in Tables 3 and 4. In the intervention group, the results showed a within-group decrease in general fatigue after a typical working day (P=.001; Table 3) from baseline to second follow-up and in influence on own work from baseline to first follow-up (P=.04; Table 3). The remainder of the other outcomes showed no effect from the intervention (Tables 3 and 4).

No adverse events were reported.

This study is the first to detect events with excessive physical workload using only technical measurements, individual thresholds, and applying these measurements in a PE intervention. The results of this cluster randomized controlled trial showed that a PE intervention did not decrease the number of events during a working day. Other outcomes showed positive effects on influence on own work and general fatigue after a typical working day, but not on pain, perceived workload, and how often heavy lifting was performed.

Technical measurements have the advantage of being objective. Furthermore, it builds on from standardized analytical procedures of the raw data, rather than, for example, self-reports or visual observations [52,53]. We have recently shown that the intraday reliability for sEMG during lifting tasks is acceptable in laboratory conditions [38]. Furthermore, the method for detecting events of the lower back and shoulder based on sEMG and IMU from the upper back has shown high accuracy in a laboratory setting [29]. Thus, we are certain the measurement method per se was not cause of the nonsignificant findings.

There was no group-by-time interaction for the primary outcome, which was the change in the number of events from baseline to follow-up. However, the number of events increased over time in both groups, and because this technical detection was used for the first time in a field study, we performed exploratory analyses with the detection of events based on 150% of the sEMG obtained during the reference lifts and 50% MVCs (Table 2). This confirmed that there was no effect of the intervention on the primary outcome. However, the exploratory analysis did not show a within-group difference as seen in the preplanned analysis [32]. This could be related to the threshold, which might have been too low, and we might have seen a more stable normalization factor by using, for example, 30 kg as reference value. However, the within-group increase in both groups could also be related to the organization of construction work, which is characterized by a distinctive variation regarding work pace, work tasks, and work processes. This variation makes it challenging to conduct intervention studies with field measurements in the construction industry as the inherent variance will necessitate a larger sample size than anticipated based on laboratory measurements. As the participants were further into the process of their current construction project during the follow-up than at baseline, this may have increased the work pace due to incentive reasons. However, analyses of heart rate and step count did not support this speculation. Another possibility could be that the participants were more aware of the measurements at baseline and therefore acting more carefully to avoid heavy lifting. This effect may have diminished during follow-up.

The other outcomes showed effects on influence at work and general fatigue after a typical working day. The difference in influence on own work was only seen at first follow-up and was primarily driven by the control group experiencing a decrease in influence. This may indicate that the control group felt neglected compared with the intervention group who attended the workshops and had the opportunity to bring forward new ideas. The second follow-up was 3 months after the last workshop. Therefore, the feeling of being neglected might have eased off, likely because the tangible consequences of the intervention were often only “increased attention to physically stressful work” as shown in Multimedia Appendix 2, rather than real changes in the working process or technical assistive devices. The decrease in general fatigue after a typical working day in the intervention group indicates that some effect occurred in response to the intervention despite not being effective in reducing events. It can be speculated that the implemented solutions led to work that reduced light loads, repetitive work, or made the work processes more efficient in general and thus less physically fatiguing. A review has shown reductions in physical work demands and musculoskeletal symptoms if mechanical lifting devices are introduced at workplaces [54], and other studies have shown a decreased discomfort [55] or ergonomic improvement when introducing devices for raised bricklaying that may decrease the physical workload during construction work [56]. As the majority of the implemented suggestions concerned technical assistive devices, it can be speculated that the increased use of assistive devices may partially explain the decrease in general fatigue.

The majority of the suggestions were related to assistive devices (Multimedia Appendix 2) and are in accordance with previous findings, where the workers identified ergonomic solutions using assistive devices to reduce WMSDs, but the support from the contractors to implement these was lacking [57]. Accordingly, other studies suggest that support from the management is critical for providing changes in the construction industry [58,59]. In this study, the management was often not willing to support the suggestions if they involved increased costs. Hence, more support from the management might have had a positive effect [60]. The intervention might have failed in involving the management as we underestimated the challenge of obtaining economical and persistent commitment from the management. However, this seems to be a highly common but underaddressed issue in participatory research [27,61].

There may be several contributing factors to the high physical demands of construction work, of which work organization plays an important role. Construction work in Denmark is characterized by being organized in small working units, often on a group-based wage, which can be associated with an increased risk factor for WMSDs [62] and can induce a group pressure within the gang to get the work done at a certain time without taking pain into consideration [63]. Studies suggest that both structural and cultural changes are necessary to create changes in the construction industry [57,64]. The lack of effect from the intervention in this study might be related to the culture in the construction industry where WMSDs are an accepted part of being a construction worker [65,66].

With the rapid technological development, this method could be integrated into portable devices connected to, for example, mobile phones and thereby provide the worker with direct feedback to prevent work tasks with excessive physical workload.

A strength of this study is the cluster randomized design, making it possible to intervene at gang level, thus reducing the risk of a number of biases associated with nonrandomized studies. However, there are also known challenges of conducting behavioral randomized controlled trials, for example, blinding of participants or potential participants and supervisors who do not accept randomization [67]. Another strength is the use of technical measurements to quantify the workload rather than relying on self-reports or observations.

A limitation of this study is that the number of dropouts was higher than expected, resulting in reduced statistical power. The terms of employment in the construction industry are dominated by short-term contracts, which resulted in a relatively high turnover of workers in the participating gangs in this study. This affected the number of participants employed over the entire intervention period. In research involving randomized controlled trial, it is preferable to have a stable group of participants. However, to our knowledge, no participants drop out of the study due to a lack of willingness to participate but were missing at random, and all participant were included in the intention-to-treat analysis. To fully control a randomized controlled trial in the construction industry showed to be extremely difficult due to, for example, variation in work tasks performed during a working day and sudden changes based on unpredictable incidents at the construction site. On the other hand, a considerable strength of this study is that the measurements have been conducted during actual construction work. Furthermore, it is a limitation that the size of the individual clusters was larger than anticipated, which also reduced the statistical power due to intracluster correlation. In general, the variance in measurements was also higher than expected. Due to these factors, future studies would need to recruit a much larger sample size to be randomized. On the other hand, the results of this study do not indicate that a relevant between-group difference would be reached even with a larger sample size.

The loss of data from sEMG, especially from the erector spinae muscles at baseline, was also a limitation. This loss of sEMG data was primarily caused by electrodes that slipped off, and future studies should minimize this loss of data by securing the sEMG cables such that excessive sweating of the participants does not compromise the skin-electrode impedance. Larger band aids over the electrodes or performing the measurements during the cooler season of the year when sweating is not a big issue might also help.

The inherent variation in daily working tasks at the construction sites is a practical challenge because the necessary sample size can easily grow to a level that is not realistic to achieve. We tried to control this by having close contact with the construction sites and conducting measures on the workers during similar working tasks, but this was not always possible. However, we compensated for this by controlling for steps and heart rate in the analysis.

During the recruitment, we were in contact with many small-scale construction companies that were unable to participate because their job tasks did not permit the long follow-up time in this study. Hence, we only included large-scale construction companies; thus, one should be cautious about generalization of our results to small-scale construction companies. On the other hand, changes are often even more difficult to implement in smaller companies where resources are scarce. It is, therefore, unlikely that inclusion of smaller companies would have changed the main conclusion of the study.

Finally, the difference in age and, partly, experience between the intervention and control group could be limitations to the study. Therefore, we controlled for age in the statistical analysis, but it cannot be ruled out that a more experienced intervention group could have increased the implementation rate of the suggested solution and thus reduced the number of events with excessive physical workload following the intervention.

This PE intervention with 3 workshops did not reduce the number of events with excessive physical workload during construction work. An exploratory analysis using higher thresholds confirmed the results. The intervention group experienced a reduced general fatigue and an increased influence on own work following the intervention, compared with the control group.