Nineteen of the twenty primary studies identified by this review recorded that a greater proportion of veterans of the Persian Gulf War of 1991 reported painful symptoms compared to other military service personnel who were not deployed to the Gulf War. For all five sites of bodily pain, each of the summary estimates from the meta-analyses indicated deployment to the Gulf War was associated with increased odds of reporting painful symptoms. Gulf deployment was most strongly associated with abdominal pain, with Gulf veterans being more than three times more likely to report such pain.
Unfortunately the majority of studies included in this review did not investigate whether Gulf War veterans report more symptoms of severe pain than non-Gulf veterans. Only Kang et al and Kelsall et al reported prevalence estimates separately for moderate to severe symptoms. Kang et al found that a greater proportion of the Gulf veterans than non-Gulf veterans reported more severe symptoms of joint pain, but they did not differ from non-Gulf veterans in the severity of symptoms of back, muscle or abdominal pain . Kelsall et al reported that more of the Gulf veterans suffered from general muscle aches and pains that were more severe in nature but that their degree of symptoms of back or joint pain was the same as non-Gulf veterans .
Statistical heterogeneity between study estimates for a particular site of pain was significant (with the exception of chest pain). Variation in each of the following characteristics across studies probably contributed to this heterogeneity: sampling strategy (single military units versus stratified random samples), degree of differential response rates between Gulf and non-Gulf veterans, method of symptom ascertainment, measured period of prevalence and specific definition of symptoms.
At least the statistical heterogeneity that arose probably reflects heterogeneity in the strength of association rather than the direction of association.
The prevalence of painful symptoms amongst Gulf War veterans was most often reported to be between approximately 20% and 40% depending on the site of pain and exact definition of measurement. In contrast, a recent population-based survey of young adults aged 18 to 25 years in the UK observed that 66.9% (95%CI 63.7% to 70.1%) reported any pain within the previous six months, although a low response rate (37%) means the estimates should be interpreted with caution . The prevalence of pain amongst military personnel when compared to the general population might be expected to be relatively low due to a "healthy worker" effect, but conversely the increased risk of pain received through injuries during military training may contribute to the prevalence of pain in military populations. This highlights the need in study samples for a relevant comparison between veterans deployed to the Gulf War and either veterans deployed elsewhere or non-deployed military personnel.
Limitations of primary research
Sampling of participants
In a cross-sectional survey it is important to derive a random sample of all those subjects who are potentially eligible in order to generate a representative sample of the larger population of interest. Those studies which selected a random sample of veterans from either US, British, Canadian, Danish or Australian military personnel databases are likely to have fulfilled this criterion [17, 20, 21, 23, 26, 28, 29]. However, those studies which sampled more opportunistically from individual military units are more prone to selection bias [3, 11, 12, 18].
In general, most of the studies achieved a satisfactory response rate amongst veterans of the Gulf War. However the response rate amongst non-Gulf veterans unfortunately tended to be systematically lower in most studies for which data were available. Differences in response rates between the exposed and unexposed groups can lead to bias if the responders are systematically different to non-responders. Unwin et al  intensively followed up a random selection of non-responders and found that those with more symptoms responded earlier but there was no significant interaction between deployment, late response and health outcome. So the prevalence estimate of symptoms might be a biased overestimate, but relative measures of effect as reported in this review should be less prone to bias. Kelsall et al  also suggested that response bias is unlikely to fully explain any differences observed between Gulf and non-Gulf veterans. They reported that odds ratios from a prediction model which assumed full participation and accounted for age, rank and service were only marginally lower than corresponding odds ratios observed for participants.
All of the studies relied on the veterans' self reported symptoms of pain which would be prone both to random measurement error and more importantly to measurement bias. Two studies included symptom items in their questionnaires which were not thought to have any physiological basis but were designed to estimate the level of over-reporting of symptoms amongst Gulf War veterans. For example Knoke et al  found that 1.2% of Gulf veterans versus 0.2% of non-Gulf veterans reported symptoms of 'earlobe pain', whilst Sostek et al  reported a significantly greater proportion of Gulf veterans reported a 'change in the colour of fingernails'. These results suggest that at least some of the association between Gulf deployment and reporting of painful symptoms might be explained by systematic over-reporting of symptoms amongst Gulf veterans.
In an attempt to minimise the measurement error and possible bias that might be associated with the reliance on symptom checklists, Simmons et al  introduced the use of open-ended questions enquiring about any new medical problems or changes in general health since 1990. This method of data collection was indeed associated with lower overall prevalence of symptoms but still demonstrated greater reporting of symptoms amongst Gulf War veterans relative to non-Gulf veterans.
A few of the earliest studies did not attempt to control for potential confounders in any way and therefore may have inflated estimates of risk [3, 11, 12]. Some studies accounted for the effect of gender by restricting their analysis to a single sex [21, 22, 24, 28, 29], whilst some studies made adjustments for a number of confounding variables in the analysis of the data [18, 19, 21, 22, 25, 27–29]. However, the later and larger studies tended to control for potential confounders more thoroughly in the sampling design of the study by matching veterans on age, sex and at least some aspect of military status [13–17, 20–22, 26, 28, 29].
Since the meta-analyses are based on the raw prevalence data from each study, potential confounders could only be partially accounted for in the resulting summary odds ratios if individual studies stratified both the Gulf and non-Gulf samples on age, sex or military status. To estimate the size of the possible effect of confounding on our reported summary estimates, it would be useful to compare the unadjusted and adjusted results from any of the primary studies. However very few of the primary studies report both raw and adjusted results. Unwin et al reported an unadjusted OR of 2.8 (95%CI 2.5–3.2) for joint pain in male Gulf veterans versus era controls which was reduced to an OR of 2.2 (95%CI 2.0–2.6) after adjusting for age, smoking, alcohol consumption, marital status, educational attainment, rank, employment status and civilian or military status on follow-up. It might seem reasonable to assume that our unadjusted summary ORs arising from the meta-analyses might be similarly overestimating the true association between painful symptoms and Gulf deployment.
Strengths and limitations of this review
This review benefits from a sensitive search strategy based on both published material and on grey literature such as conference abstracts and preliminary reports. Furthermore, inclusion and exclusion criteria were independently assessed by two reviewers. However, failure to identify some studies is always a possibility in systematic reviews. The majority of the primary studies that we identified reported Gulf deployment to be independently associated with symptoms of pain. The absence of many studies with negative findings raises the possibility of the existence of publication bias. However in order to affect the weighted summary estimates derived from the meta-analyses, any statistically significant negative results that are currently missing from the review would have to have been based in large study samples and these would have been more likely to be published. Therefore the likelihood of publication bias being present which would actually alter the conclusions of the review is small.
This review could only investigate symptoms of pain in sites reported by the primary studies. Published papers might have been limited to reporting only the most frequently recorded symptoms rather than all measured symptoms [14, 21, 22, 29], and therefore the association between Gulf deployment and symptoms of pain in other unreported sites is unknown. However, given the consistency in the results for all measured sites of pain included in this review, it might seem unlikely that Gulf deployment would have a dramatically different association with any unreported site of pain.
This review has been limited to investigating the association between Gulf deployment versus non-deployment and reporting of pain. We chose not to examine the association between specific environmental exposures of the war (e.g. threat of chemical warfare agents, non-routine immunisations) and reporting of symptoms due to the problems associated with the inaccuracy of such self-reported exposures.
In this review we were not attempting to measure the possible underlying biological or socio-cultural mechanisms which could explain the observed association between Gulf deployment and symptoms of pain. However, the experience of being deployed into a potentially life threatening situation is obviously extremely stressful, and psychological stress can manifest itself in a range of physiological symptoms, including pain .