The primary author applied the selection criteria to the list of potentially relevant reviews. The selection process was accomplished in three steps, first based on only the titles, second on the abstracts, and finally on the retrieved full papers. In case of doubt, a decision on inclusion or not was obtained by discussion and argued agreement with one of the other authors (NF).

Information on study designs included the number of original studies in the review, the number of studies concerning a specific topic, and the conclusions reached by the review authors. The references to original studies included in the reviews were obtained and the degree of overlap between the different reviews regarding included studies was recorded.

Review articles were critically appraised by two authors (JHA, NF) using AMSTAR Assessment of Multiple Systematic Reviews, an 11-item tool designed to appraise the methodological quality of systematic reviews [18], [19]. The AMSTAR items are scored as ‘Yes’, ‘No’, ‘Can't answer’ or ‘Not applicable’. Disagreements on item scores were resolved by discussion.

The AMSTAR criteria comprise:

The maximum score on AMSTAR is 11, and 0–4 indicate that the review is of low quality, 5–8, of moderate quality, and 9–11, of high quality.

We considered conflict of interest statement as of minor importance in this topic of computer use and musculoskeletal health. Therefore, our rating differed from the original AMSTAR and with a maximum obtainable score of 9 we considered 0–4 as low quality, and 5 or more as moderate to high quality.

Due to the heterogeneity of studies we did not adopt a formal grading of the synthesized evidence, e.g. like the GRADE approach [20]. We made a more qualitative assessment of (1) the overall quality of the presented evidence, (2) the overall likelihood that the observed associations were due to a causal relationship, and (3) the existence of evidence based recommendations for practice derived from intervention studies.

One of the three reviews used a quality assessment list [26], while the two others used a qualitative approach to discuss study design, exposure assessment and outcome definition. Of the total 11 studies in the reviews four were prospective in design, one was a case-referent study, one was cross-sectional with a case-referent approach, and four were cross-sectional.

The two latest reviews used an identical quality assessment list, originally derived from Dutch studies [49]–[51]. In the Norwegian review relatively small intervention studies had the same or higher quality assessment than large epidemiological studies. The two reviews [14], [25] with low quality score by AMSTAR did not include any stated quality assessment of the included studies. The review by Gerr et al. from 2006 [23] benefits from building on a 2004 review from the same group, including the same 17 studies as in 2004 [24], and adding 27 more.

The quality assessment in the intervention reviews [34]–[36] was based on explicit and comparable criteria, but Leyshon [36] rated the quality scoring in a different way. Verhagen et al. [29] used a Dephi list adopted from [52] to assess quality; Boocock et al. [33] used the modified Cochrane Musculoskeletal Injuries Group scoring system [53], while Driessen used a modified version of this system and an assessment of the overall quality of the evidence based on the GRADE approach [54]. The different approaches produced different results. As an example two studies on arm support [47], [48], were scored as studies of high quality in two reviews [34], [35], of medium quality in one review [36], and of low quality in another [30].

The three included reviews focusing specifically on carpal tunnel syndrome consistently concluded that epidemiological evidence for computer use and the occurrence of CTS is insufficient. In contrast, one of the more general reviews on risk factors [22] concluded that the risk for CTS is increased with use of a computer; especially with mouse use for more than 20 hours per week. This conclusion was, however, based on only one original study [37]. This review was also included among the carpal tunnel reviews, but since its conclusion was based on the evidence from a single study the reviews on CTS considered this evidence as insufficient to alter the overall conclusion.

At the same time, the quality assessment of the three CTS reviews in general yielded higher scores than the broad reviews of UEMSDs (Table 1).

The seven reviews had remarkably different conclusions Table 2, ranging from ‘consistent evidence’ [22], ‘extensively researched and generally well established’ [25], to ‘moderate evidence’, and ‘limited evidence’ in the two latest reviews [31], [32]. As previously mentioned the reviews are to a large extent based on different original studies, which is clearly visible in the Venn diagrams (Figure 2 and Figure 3). However, in the majority of the reviews the core body of evidence seemed to be mainly drawn from the same studies. The exception being the two reviews [14], [25] that received the lowest quality score of the in the AMSTAR rating (Table 1).

The most comprehensive review by Gerr et al. [23] concludes that there is a somewhat consistent finding of an observed association between hours of computer use and adverse hand/arm outcomes, and to a slightly lesser extent for neck/shoulder outcomes, but several methodological limitations are addressed and discussed. The review by Waersted et al. was commissioned by the Danish National board of Industrial Injuries, thus focusing on studies with clinical examinations, and they found only limited evidence for an association between computer use and specified disorders of the neck, shoulder, elbow, forearm, or wrist [29].

The reviews in this category cover a time span of six years and there was a tendency for the most recent reviews to have higher AMSTAR score. Individual studies were included in the 7 reviews a mean of 2.04 times [SD = 1.30], but a number of core studies seemed to be common for most reviews. One study [38] was included in six of the reviews, while the papers from NUDATA [39], [40], [44] were included in five of the seven reviews. Nearly half of the original studies - 37 out of 80 - were only included in one of the reviews.

In a synthesis of the evidence presented in the seven reviews the picture is not straight forward.

There seems to be evidence in the reviews for an association between computer use and pain reporting and discomfort, especially from the distal arm and hand.

However when considering prospective studies and accepting the inclusion of clinical criteria incorporated in some of the reviews, the evidence seems moderate or even limited for an association between computer use and UEMSDs.

The one specific review on neck pain [28] was not aimed at computer use, and included only cohort studies. However, 5 studies from the total pool of studies among computer users [38], [41], [44]–[46], were included in the review and contributed to the conclusion that a wide range of workplace physical factors were risk factors for neck pain (Table 2). The multifactorial nature of neck pain was however emphasized as their main conclusion.

The six reviews are fairly consistent in concluding that there is limited evidence for specific interventions and a conspicuous lack of more high quality studies (Table 3). The most comprehensive reviews finds moderate evidence for no effects of workstation adjustments and rest breaks [34], [25], whereas some evidence supported the effect of arm support, based on 2 studies [47], [48]. The reviews had poor overlap as illustrated by the Venn diagrams in Figure 4 and Figure 5.

The reviews included are sufficient to address the objectives of the overview. The three reviews on CTS were consistent in their agreement on an insufficient evidence for computer use leading to CTS. As all reviews were considered to be of medium to high quality the conclusions about causation can be considered fairly certain.

The reviews on UEMSDs were far less specific than the CTS reviews and not all combinations of computer use and types of outcome have been studied in detail. In the majority of reviews, outcomes from the neck and upper extremities were described as “pain complaints”, defined in various ways. There was some evidence across reviews for an association between computer use and pain complaints, and this relation could be causal, but the argument for a causal relationship would be more convincing if there had been a demonstrable threshold. The review which specifically included diagnoses [32] found limited evidence for a causal relation between computer use/mouse use and tension neck syndrome and wrist tendonitis and for mouse use and forearm disorders. However, the conclusion concerning wrist tendonitis was partly based on [36], [54], and it should be noted that the original papers refrained from such a conclusion due to a very small number of cases despite a large cohort. Waersted et al. [32] further found insufficient evidence for shoulder tendonitis and epicondylitis [lateral or medial]. There was no evidence across reviews for a causal relation between computer use and specific disorders of the neck and upper extremities. The evidence across reviews of intervention studies was fairly consistent in showing limited evidence for the effect of specific interventions. At best a multi-targeted strategy could be recommended, if there was a call for an intervention.

The aim of this overview was to assess the current evidence for possible causal relationships between computer work and CTS and UEMSDs, and to synthesize possible recommendations for intervention in the work place. The reviews of risk factors did not give sufficient information to derive quantitative dose-response relationships, and the findings from the studies that indicated a relationship between duration of computer use and pain complaints could not indicate a threshold value. Without such quantifications we cannot derive the duration and/or the degree/load/force involved in the work situation that will result in an increase in the occurrence of UEMSDs and CTS. A prerequisite for the combination of information in different studies would be that the different studies had used sufficiently similar measures of exposures and outcomes.

All included reviews – except [14], [22] - reported at least some details about their literature search, selection process and quality assessment. The lack of methodological rigorousness in the two review [14,22] was reflected in the low AMSTAR score obtained by these two reviews compared with the others.

The scopes of the reviews were different. The three reviews on CTS focused on one specific outcome, and generally obtained higher quality scores than the reviews looking more broadly on UEMSDs. The study aim was rather diffuse and general in two reviews [14], [22], and again reviews with more precise aims obtained a higher quality score. The tendency for some reviews to rely exclusively on cohort studies is questionable. In studies of musculoskeletal outcomes, cross-sectional studies and other study types than cohort studies may indeed provide valuable information. The two reviews by Gerr et al. acknowledge this [23], [24], and maybe this is not fully rewarded by the AMSTAR scores. The most recent review on risk factors on the other hand included intervention studies in their appraisal of evidence for causal relationships, but this seems to be a questionable approach as long as risk factors or risk agents intervened against are not well established.

It is noteworthy that all the reviews of intervention studies scored relatively high on the AMSTAR criteria, but that may actually cast some doubt about the appropriateness for using AMSTAR in public health studies. In our opinion two of the intervention reviews [34], [35], which originated from the same group, had a higher quality than the others, based on their completeness, thorough search strategies, and rigorousness of interpretation, but this was not fully reflected in the AMSTAR scoring.

The quality of the evidence regarding computer use and CTS was considered high. The three reviews were consistent in their conclusions and the results seem well founded. For UEMSDs the reported evidence ranged from insufficient over limited to consistent and well established, and the quality of the evidence was considered moderate. It cannot be excluded that future studies will change or modify the conclusions on UEMSDs. The reviews of interventions were consistent in their conclusions on the lack of high quality studies and the shortage of evidence based recommendations applicable in today's office environment. The quality of the evidence for this statement was considered high and there is no reason to believe that important information from intervention studies has been missed.

Authors of all the included reviews were contacted to ensure that any substantial reviews or ongoing projects was not overlooked, and we are not aware of any research group currently performing comprehensive reviews or meta-analysis. Five reviews were excluded due to language. However, based on the English abstracts included in the reviews and their reference lists we do not believe substantial information was missed. The whole area of grey literature was not formally searched, but we screened some reports for studies, which were not accounted for in the included reviews. The 2001 consensus report from National Research Council [58] included 9 of the 80 papers on risk factors covered by the six reviews in this paper. A Norwegian report on work as causal for musculoskeletal disorders from 2008 obtained information from 7of the risk factor studies, 11 from the intervention studies, and they included 6 of our 17 reviews [59], and an European report from 2008 [60] extracted their information from 2 of the reviews in this paper [26], [31]. From these three examples it can be concluded that commissioned reports attempting to cover exposures and outcomes very broadly, ends up basing their evidence on a smaller number of original studies than most of the reviews in this paper.

It has been shown that the single largest source of error in relation to sensitivity in searching is errors in conceptualization of the search [61]. The very first step in evidence based medicine is to formulate questions that can be answered [62]. If nothing else, it became clear that the included reviews with the lowest AMSTAR score also asked the most imprecise questions.

Our search strategy involved a librarian with many years of professional experience, and we searched both in Medline, Embase, CINAHL and Web of Science. A complementary search in the Cochrane database did not reveal additional reviews. We consider our search strategy to cover the important information for our research questions.

There was a poor overlap between the studies selected in the reviews on risk factors of UEMSDs and intervention studies, whereas the overlap for assessing risk factors for CTS was much higher. Probably the bias in this overview is lower than in most of the included reviews, because the information from all the reviews was taken into account.

The AMSTAR scoring was performed independently by two authors and there were very few disagreements, which all could be solved by discussion. In the synthesis of the overall evidence we made a qualitative assessment rather than a formal one. This could be criticized, but we find measures such as GRADE developed to suit public health studies less well than studies on pharmaceutical interventions. A formal system to synthesize the evidence into conclusions would have been preferable, but to our knowledge nobody has set up such a system and tried it out in relation to reviews of etiology or public health interventions.

We could not identify other overviews with the same aim as our own.

The research conducted in the last decade has clearly led to more modest conclusions on the relation between computer work and the risk of CTS and UEMSDs, and demonstrated the lack of well documented recommendations concerning interventions in the office environment. This development is a cause for concern. If there were a causal relation between computer work and CTS and UEMSDs, it would have been expected that the introduction of better study designs, larger studies, and more rigorousness in analysis, had strengthened the evidence, but apparently the opposite has been the case. The explanation for this is not straight forward. It's seems unlikely that changes in the office environment over the last 10 to 15 years should have reduced previous exposure levels and made computer related MSDs a thing of the past. It may be argued that office ergonomics and work postures have been improved, but at the same time we have witnessed work intensification and an increased mouse use.

An alternative explanation would be that the alarming reports from earlier studies reflected a public anxiety of new technology, more medicalization of society and the radical change of work organization in the office environment rather than a traditional association between high physical work place exposure and adverse health outcomes. Perception of pain and subjective feelings of discomfort in association with computer work is not a trivial issue, whether the association is causal or not. The effects in terms of loss of productivity, reduced well-being at work and a negative influence on psychosocial work conditions are substantial and should be of major concern for everyone involved in working environment issues and occupational health. The solutions, however, appear to be complex, involving a system approach, focusing on safety management systems and work organizational changes based on participation and dialogue. In this perspective, equating pain and discomfort from computer work with musculoskeletal injuries in traditional industrial work seems to be an unfruitful approach.