Prospective research on musculoskeletal disorders in office workers (PROMO): study protocol

With respect to research question A, we expect that hand-arm symptoms are more strongly related to the duration of computer use than neck-shoulder symptoms. Previous studies showed the strongest and most consistent associations for computer use with the incidence of hand-arm symptoms [4, 6, 7, 13]. In addition, based on the same studies, we expect to find indications for a dose-response relationship between the duration of mouse use and the incidence of hand-arm symptoms.

By answering research question B, we will investigate the contribution of occupational computer use to the incidence of hand, arm, shoulder and neck symptoms, compared to the contribution of various other occupational and non-occupational risk factors. Firstly, we expect occupational computer use to be the strongest risk factor. Previous longitudinal studies, which included individual factors as well as estimates of occupational mechanical and psychosocial exposure, and leisure time exposure, have found the most consistent and strongest associations between the duration of mouse use and the incidence of hand-arm symptoms [4, 6, 7, 13]. In addition, we expect computer use to be more strongly associated with hand-arm and neck-shoulder symptoms than ergonomic factors (i.e. working posture and workstation characteristics) [24]. If ergonomic factors have a causal contribution, one would expect that the association with hand-arm and or neck-shoulder symptoms would become stronger when exposed to longer durations of computer use. Besides occupational mouse use, we expect occupational psychosocial exposure to be an independent risk factor for neck-shoulder symptoms [25].

Secondly, we expect that low levels of leisure time physical activity contribute modestly, at most, to the incidence of hand-arm and neck-shoulder symptoms. Previous longitudinal studies among office workers failed to show an association between low levels of leisure time physical activity and hand-arm and neck-shoulder symptoms [6, 7, 9, 13]. Workers exposed to high mental stress during work time and to low physical activity during leisure time were found to have an increased risk in one study [8]. However, confidence intervals were wide in this study. It should be noted that most studies among office workers have included only crude measures of leisure time exposure. In a longitudinal study among manual workers and office workers, a protective effect of sports activities on the incidence of hand, arm, shoulder and neck symptoms was reported [9]. However, specific leisure time activities might increase the risk of symptoms. Miranda and co-workers [27] reported an increased risk of incident shoulder symptoms when playing volleyball frequently. Thirdly, both female gender and previous symptoms have been reported frequently as risk factors among office workers in the published literature [female gender: [6‐9, 13, 28]; previous symptoms: [4, 5, 28]]. We will explore whether these individual factors act as effect modifiers in the associations between occupational and/or leisure time exposure, and hand-arm and neck-shoulder symptoms. In addition, we aim to explore the role of the personality trait overcommitment in the incidence of hand-arm and neck-shoulder symptoms. A longitudinal and a cross-sectional study showed indications of an increased risk of hand-arm and neck-shoulder symptoms among overcommited workers [29, 30]

A prospective cohort study is conducted, with a follow-up of 24 months. Assessment of the health outcome (symptoms and disability due to symptoms) takes place at baseline and every 3 months during follow-up. Exposure data on computer use are collected continuously during the study period, while additional exposure data and information on individual characteristics are gathered at baseline and after one year of follow-up. Participation is voluntarily and participants signed informed consent. The study design was approved by the Medical Ethics Committee of the VU University Medical Center (VUmc).

Companies with a source population of at least 500 office workers were invited to participate in the study. The study population was recruited from five different employers in the Netherlands: a brewery, a financial consultancy firm, a university, a transportation company, and an insurance company. A department within a company was included if the department had a computer network from which the software for recording the duration of computer use could be installed on individual workstations, and if at least three quarters of the employees fulfilled the inclusion criteria (see table 1).

Employees within included departments were informed about the study via distributed flyers. In addition, all these employees received an e-mail with information on study objectives and required effort for participation. As incentive for participation, workers were offered the choice between two options. Firstly, the donation of a small sum of money (20 eurocents) to a charity organization (i.e. Medecins Sans Frontieres) for each questionnaire they would fill out. Secondly, joining a lottery for a weekend holiday to a European capital. The latter option was only possible if they would fill out all the questionnaires during follow-up. Finally, a team of researchers visited the worksites and asked individual employees to participate in the study. At the same time, memo blocks were handed out as incentive. In total, 2461 out of 9161 (27%) approached employees signed informed consent. Out of these 2461 participants, 1821 (74%) filled out the first questionnaire.

Data on exposure and outcome is collected using web-based self-reports. Participants receive an e-mail containing a link to a questionnaire. By request, they can fill out a hard copy of the questionnaire. In case of non-response, participants receive a maximum of two reminders by e-mail. In addition, data on computer use is collected objectively with a software program. Participants who leave their job during follow-up, receive a paper questionnaire to their home address in order to check symptom status and the possible contribution of symptom status to turnover. This information will be used to check if a healthy worker (selection) effect has occurred: symptomatic workers might leave the study more frequently than healthy workers, leading to biased associations.

Data on computer use are collected at baseline and continuously during follow-up using the software program WorkPace version 3.0 (Niche Software Ltd/ErgoDirect). The program has been installed from the central network on the individual computer of the participants. The program records computer use from the moment the participant has logged in to the network. Data storage takes place on the individual computer. Periodically (i.e. after being logged in to the network for 6 hours, or during log-off) the individual data file is sent to a dedicated and secured folder on the central network. Recording is continued if a person logs in to the network on another computer. Registration of keystrokes, mouse clicks and mouse movements are stored as cumulative totals per day. Thus, separate estimates for the duration of total computer use (including reading from the screen, mouse and keyboard use), mouse use and keyboard can be retrieved. Based on the duration of the time interval between two consecutive active events (i.e. keying, mouse clicking or mouse movements), the duration of keyboard, mouse, and total computer use, as well as the duration of breaks are calculated. If a participant hits a key, moves or clicks a mouse within 30 seconds of previously hitting a key, moving or clicking the mouse, this "inter-events period" (in seconds) is stored as a usage period of total computer use. If the threshold time of 30 seconds is exceeded, the elapsed time period between two usage periods is stored as a break from total computer use. The threshold time for mouse use is 5 seconds and for keyboard use it is 2.5 seconds. Cumulative totals for several usage and break periods are stored for every day separately.

Previous research has shown good agreement between the WorkPace estimate and systematic observation for the duration of total computer use. On group level, the average duration of total computer use estimated with WorkPace is within 10% of the average duration of total computer use estimated with systematic observation [15, 31]

Self-reported data on duration of computer and mouse use, historical computer use, precision demands, use of break and exercise software, workstation characteristics and working postures while using keyboard and mouse are gathered at baseline and after one year of follow-up. Clarifying illustrations have been added to the questions on working postures for optimal validity. This questionnaire also contains questions on job tenure, job contract characteristics (e.g. working hours and working days), overtime work and work continuation during formal (lunch) breaks.

Occupational psychosocial stressors and perceived stress are measured using a translated version of the Effort-Reward Imbalance questionnaire [32] and the need for recovery scale from the Questionnaire on Perception and Judgment of Work [33, 34]. In addition, the subscale decision authority from the Job Content Questionnaire is used [35]. Mental load is assessed by a subscale of the Questionnaire on Perception and Judgment of Work [33]. Job satisfaction is measured using a single item of the Questionnaire Work and Health [36]. In addition to baseline and one-year follow-up measurements, information on job satisfaction and rapid increases of general workload is collected every 3 months.

At baseline and after one year of follow-up, physical activity during leisure time is assessed by two questions. The first question focuses on the average number of days per week in the last 3 months with moderate intensity physical activity (i.e. causing increased breathing frequency) lasting at least 30 minutes in total per day. This is done to check whether physical activity public health recommendations are met [37, 38]. The second question focuses on the average number of days per week in the last 3 months with high intensity physical activity (i.e. causing sweating), and lasting at least 20 minutes in total per day [40, 41]. Activities during leisure time involving forceful or repetitive arm or hand movements (e.g. participation in strength training of the upper extremities and racket sports, and playing music instruments) and duration of total computer use during leisure time are assessed separately with a self-administered questionnaire.

At baseline and after one year of follow-up, the personality trait overcommitment is measured within the Effort-Reward Imbalance questionnaire [32]. General health, age, gender, education, hand dominance (i.e. preferred hand for handwriting), body height and body weight are assessed with a self-administered questionnaire.

Every three months, data concerning symptoms (pain or discomfort) in the lower back, neck, shoulder, arm or hand during the past 3 months are gathered by means of a validated, modified version [41] of the Nordic Questionnaire [42]. Localization (anatomical region and side [left and or right]), duration and frequency of episodes of symptoms are recorded. The intensity of symptoms is measured using Von Korff scales [43]. Limitations in work or leisure time activities due to symptoms are measured using a Dutch translation of the scales that were used in the EPI-mouse study [44]. Data on the work-relatedness of symptoms and the presence of systematic disease or other causes of symptoms (e.g. traffic accident, burns, and rheumatic disease) are also gathered. Sick leave due to symptoms is measured by two questions that have shown adequate agreement with company records in back pain research: 97% specificity and 88% sensitivity [45]. In addition, long-term sick leave (i.e. longer than 6 weeks) is registered by the participating organizations or their occupational health service. Data on duration (days), frequency, level (full or partial sick leave) and diagnostic code are gathered. Participants who consult the occupational physician (OP) because of prolonged sick leave (i.e. 3 to 6 weeks) are diagnosed according to the Dutch guideline for the management of arm, neck and shoulder symptoms by occupational physicians [46]. All OPs connected to the participating organizations received training on diagnosing and coding hand, arm, shoulder and neck symptoms to reduce inter-observer variation in diagnosis and coding.

The advantages of the PROMO study in comparison with studies published so far include the long follow-up duration (two years), the repeated measurements of both exposure and outcome, and the objective measurement of the duration of computer use. These features of our study will enhance the accuracy of risk estimates. In addition, the frequent exposure and outcome assessment will provide more insight in the time window(s) of relevant exposure effects. This issue has been identified as an important topic, but has up to now not received appropriate attention [55, 56].

At the same time, observational studies, including the PROMO study, are threatened by several sources of bias. Selection bias might be present, since only one out of four workers, who were invited for the study, participated. Selection bias might hamper generalizibility of study findings to target populations. The generalizability of our findings is dependent on the distributions of effect modifiers in the target population, compared to our study population. An adjusted risk estimate in the absence of effect modification is highly generalizable, since it is by definition independent of the distribution of confounders in a target population. In the literature, few effect modifiers have been identified among office workers, making selection bias unlikely, at this moment.

Internal validity is threatened, as in most observational cohort studies, by (residual) confounding. In order to ensure that participants could fill out the baseline questionnaire within 30 minutes, we decided to restrict the item pool to the most relevant variables based on the available evidence [11, 24, 57, 58]. We cannot be certain that we did not miss relevant variables. In addition, except for the duration of computer use, all other exposures are measured by self-report. It is known that self-reported exposure estimates have more measurement error than "objective" estimates [55]. In general, this will lead to an underestimation of the true risk. However, if more than one risk factor in a multivariate analysis is measured with error, as is the case in this study, both underestimation and overestimation may occur [59]. It follows that our analysis on the contribution of various risk factors might be constrained by differences in measurement accuracy.

It has been suggested that information on physiological mechanisms is necessary to interpret epidemiological findings, since epidemiological data will always be compatible with a wide variety of underlying causal mechanisms [60]. The integration of epidemiological and physiological data might enhance the identification of causal risk factors. However, a wide range of mechanisms has been proposed to explain hand, arm, shoulder and neck symptoms among office workers, making it difficult to identify the mechanism(s) at stake. Both peripheral tissue injury and reorganization of the central nervous system have been put forward as potential mechanisms [61]. A substantial part of pathophysiological research has focused on sustained low intensity muscle activity, leading to muscle disorders and consecutive symptoms. Although supportive evidence for this mechanism is available [62], it has been criticized as well [63]. Since empirical tests of different theories are not available, the specific physiologic mechanism(s) underlying hand, arm, shoulder and neck symptoms cannot be identified yet. A possibility for further investigation is to concurrently test the predictions of different theories in epidemiological studies. According to Barr and co-workers [61], a high repetition of movements is the main risk factor. According to Visser and co-workers [62], continuous low intensity muscle contraction (caused by occupational mechanical exposure and dependent on different situational and individual factors) is the risk factor of main interest. Knardahl [63] proposes mental demands (i.e. information processing demands) during computer use, as the most important risk factor. To concurrently test these theories, the amount of unique variation in the health outcome explained by the different constructs and the amount explained by shared variance could be investigated [64]. Therefore, subgroups of workers should be defined, in which an increased risk is expected based on the constructs used in the theories. Moreover, the extent of improvement in the prediction of the health outcome by adding a construct from an alternative theory should be investigated. This improvement in prediction could be operationalized as the improvement in explained variance or the presentation of extra relevant attributable fractions for the added construct. We will use this approach in the PROMO study.

Indirect evidence of causality might also be attained from the results of primary preventive interventions. However, only a small number of such studies has been published so far [65‐67], making inferences based on this kind of evidence premature.

To increase our understanding of the etiology of hand, arm, shoulder and neck symptoms among office workers, the best way forward might be to combine multidisciplinary research efforts of observational and (lab- and field-based) experimental research. Hypotheses stemming from different research traditions should be tested (concurrently) under different situations, to make inferences about specificity and generalizability. It follows that the evidence from different levels of inference (e.g. genetics, physiology, biomechanics, individual behaviour and culture) should be integrated to optimally serve the goals of public and occupational health [68]. For a start, in the PROMO study both lab-based researchers and field-based researchers will collaborate. Moreover, a range of different hypotheses will be used to interpret the collected data.