The association between body fat and musculoskeletal pain: a systematic review and meta-analysis

The following databases were searched on 9th August 2017: Medline (Ovid); PubMed (non-Medline content only); Embase (OVID); Scopus; CINAHL (EBSCOhost); Cochrane Central Register of Controlled Trials; and Web of Science. All databases were searched from inception to current date. Reference lists from suitable papers were also investigated and included prior to applying exclusion and inclusion criteria. Broad MeSH terms and keywords were used, combining musculoskeletal pain and body composition. The search terms were broad to ensure capture of all relevant studies. Additional file 1 illustrates the full search strategy used for this systematic review, and minor modifications to search terms were required depending on the database searched. Database searching and registration for automatic e-alerts were also continued until the review was finalised (8th January 2018).

Following removal of duplicates, two reviewers (TPW and JBA) applied the predetermined selection criteria to all articles by reading the title and abstract alone. Where discrepancies between article selections existed, the reviewers discussed these discrepancies to form a consensus, a third reviewer was not required to arbitrate a consensus for this review. Articles were then assessed for eligibility by full-text review.

Articles from English language, peer-reviewed, scientific journals were eligible for inclusion in this review if they reported studies that examined the association between body composition and musculoskeletal pain. Studies were included if all participants were aged at least 18 years, had musculoskeletal pain recorded via self-report or questionnaire (or were controls) and had an assessment of body fat. Studies specifically investigating participants with inflammatory conditions or autoimmune diseases were excluded. Further exclusion criteria were; unclear assessment of musculoskeletal pain or body composition, letters to the editor and editorials, opinion pieces and non-English language publications.

All included articles were assessed for methodological rigour using the Epidemiology Appraisal Instrument (EAI) [27]. This tool has been shown to demonstrate good reliability and content validity [28]. A number of items from the EAI were omitted as they were not applicable to non-interventional studies (Questions 10,12,20,22-24,35,37,40) as per previous reviews of observational studies investigating musculoskeletal disorders [29, 30]. The covariates considered important for questions 11 and 36 were age, gender and a measure of psychological health. As it is not known if each question of the EAI is equally weighted, rather than providing a quality assessment score for each study, a summary score for each question is reported. A summary (%) of the number of questions a study scored ‘yes’ on is also reported.

To reduce the risk of bias, author and publication details were removed prior to data extraction. Where available the relevant data (means, medians, standard deviations (SDs), odds ratios (ORs), relative risks (RRs), confidence intervals (CIs) and p values) were recorded for each study. Where available, multivariable OR (95% CI) were extracted in preference to unadjusted OR (95% CI). For studies reporting means and standard deviations, effect sizes (Cohen’s d) and CIs were calculated. According to Cohen [31], effect sizes were interpreted as 0.2, small; 0.5, medium; and 0.8, large. Widespread pain was defined as ≥5 painful joints, which is modeled on the criteria of the American College of Rheumatology [32]. Multi-site pain was defined as > 1 but < 5 painful joints. For those studies investigating multi-site or widespread pain, the differences were calculated between the no pain group and the multi-site / widespread pain group. Meta-analysis was performed where more than one study reported on the same parameter, grouped by widespread or single-site pain location. Only the gender-specific sample size was used when entering gender-stratified data into the meta-analysis.

The OR and CIs, and SMD (Cohen’s d) were pooled for meta-analysis by the standard approach, weighted by the inverse variance method. Odds ratios and CIs were converted to SMDs for meta-analysis [33]. Statistical heterogeneity was assessed for each site using the I² statistic. Potential publication bias was assessed graphically using a funnel plot [34] and Egger’s regression intercept for low-back pain, knee pain and foot pain. Both heterogeneity and publication bias were considered, accepting the fact that the power was low because of the small number of studies for each site. Sensitivity analysis was performed via the one-study removed test (removal of individual studies out of the model in turn), which gauges each study’s impact on the overall pooled effect size. A p-value less than 0.05 (two-tailed) was considered statistically significant. All analyses were conducted using Comprehensive Meta Analysis v3.0 (Biostat, NJ, USA).

A variety of sites for musculoskeletal pain were investigated, including the neck, low-back, knee and foot. The low-back was the most common region investigated, with 15 articles including this site in their analysis [18, 20, 35‐38, 43, 46‐48, 51, 52, 56‐58]. Three articles investigated the association between multi-site / widespread pain and body composition [18, 35, 36], while another investigated multiple regions, but stratified the analysis by these regions [37]. One study [54] used body composition as a predictor for any injury and thus a specific region was not investigated. Body composition was analysed with dual energy x-ray absorptiometry in 13 articles [16, 18‐20, 35, 36, 38‐43, 55], bioelectrical impedance analysis in 10 articles [37, 44‐50, 56, 57], and skin fold thickness in 5 articles [51‐54, 58]. Body composition was generally reported as a percentage of body fat (17/28 articles) [37, 39, 43‐52, 54‐58]. The cross-sectional articles consisted of; population-based (n = 7), clinic-based (n = 7), musculoskeletal pain (n = 5), occupational-based (n = 2) and unknown (n = 1). The longitudinal articles were largely population-based (n = 6) along with occupational-based (n = 1) and military-based (n = 1). The longitudinal articles varied in follow-up from 3 months [54] to > 20 years [58], but most were between 3 and 5 years.

The studies included in this systematic review reported on 12,942 participants, with studies from Asia, Europe, South America and Australia. Both men and women were represented in most studies, although gender-specific studies accounted for > 35% of the total [38, 41, 42, 46, 47, 49, 52, 54, 57, 58]. Mean age in the cross-sectional studies ranged from 20.7 years [50] to 74.4 years [41], while the longitudinal studies ranged from 19.0 years [54] to 64.6 years [16]. Most cross-sectional studies included participants with mean BMIs of < 30 kg/m², however four included participants with a mean BMI of > 30 kg/m² [18, 20, 40, 53]. The mean BMI of the participants from the longitudinal studies ranged from 20.8 kg/m² [54] to 29.6 kg/m² [19].

The results of the methodological quality assessment are provided in Additional file 2. The summary scores for each question ranged from 4 to 96%, with 14/34 questions scoring above 50%. The percentage of items in the EAI graded as ‘yes’ for each study ranged from 23 to 85%, indicating variable methodological quality of the included studies. There were common, strong themes among the studies with the clear descriptions of the aims, study design and results reported in most studies (> 85%). There were, however, a number of consistent methodological limitations; the reliability and validity of the instruments used was often under-reported, a sample size calculation was mostly not reported (96%) and the generalisability of the findings was questionable in over 80% of the included studies. Whilst there was adjustment for a number of other variables e.g. smoking, physical activity, self-reported arthritis, adjustment for all of the important confounding variables was reported in less than 30% of the articles [16, 18, 19, 35, 38, 40, 42, 56]. One article considered psychological health alone [44], one article considered age alone [58], six articles considered both age and gender [20, 36, 37, 41, 55, 57]. Only three articles [16, 40, 42] provided data that were adjusted for the important confounding variables (age, gender, psychological health) that were also used in the meta-analyses.

Meta-analysis of cross-sectional single-site and widespread pain studies found significant associations between body fat and pain (Figs. 2, 3, 4 and 5), summarised in Table 1. There was a positive medium effect size between total body fat mass and widespread pain (SMD 0.49, 95% CI, 0.37–0.61, p < 0.001 and I² < 0.001, p = 0.366). Single-site musculoskeletal pain also had positive associations with body fat. Low-back pain and body fat percentage had a combined small-medium effect size (SMD 0.34, 95% CI 0.17–0.52, p < 0.001), but there was a significant level of heterogeneity (I² = 91.21, p < 0.001). Body fat percentage and knee pain had a small effect (SMD 0.18 95% CI, 0.05–0.32, p = 0.009 and I² < 0.001, p = 0.941), while the pooled FMI and foot pain had a small effect (SMD 0.05, 95% CI, 0.03–0.06, p < 0.001 and I² < 0.001, p = 0.564).

The association between knee pain and body fat percentage was not significant when the data pertaining to women from Scott et al. [39] was removed from the meta-analysis, (SMD 0.16, 95% CI -0.02-0.34, p = 0.075), suggesting that the relationship may be mediated by gender. All other sites remained significant when one study was removed from the respective model.

No significant publication bias was detected for studies reporting foot pain or knee pain, with Egger’s regression intercept (95% CI) of 0.75 (− 2.38–3.87), p = 0.412 and − 0.61 (− 10.87–9.66), p = 0.589, respectively. There was however a potential for publication bias detected for studies reporting low-back pain with Egger’s regression intercept (95% CI) of 3.44 (1.57–5.33), p = 0.004. Widespread pain was reported in only two studies and was therefore not amenable for the funnel plot test or Egger’s regression intercept.

The cross-sectional studies not included in the meta-analysis (due to the type of data or the parameter used) were generally concordant with the overall findings (Table 1), with the reasons for exclusion in Additional file 3. Multi-site pain (3 sites) was associated with FMI in the study by Brady et al. [18]. Neck pain was associated with body fat percentage in one study [45], while temporomandibular pain was not [44]. The large study by Chou et al. [38] that investigated low-back pain used the same sample as reported by Butterworth et al. [42] who investigated foot pain, with both finding FMI, but not FFMI, to be significantly associated with pain. Celan et al. [46] studied the relationship between body fat percentage and low-back pain, but the only data provided were mean body fat percentage, without confidence intervals or standard deviations and therefore these data were not amenable for the meta-analysis. Iizuka et al. [37] investigated multiple regions separately (neck / shoulder, back and low-back) and their associations between body fat percentage. Whilst we felt it appropriate to include the low-back region in the meta-analysis, we did not include the neck / shoulder and the back region with the other studies given the difficulty with delineating these regions, particularly the low-back region from the back region, but we did include the neck / shoulder region in Table 1. Other studies investigating low-back pain generally found increased fat mass was associated with pain. The smaller studies that investigated both knee and shin pain found non-significant associations between pain and body fat mass.

Findings from the longitudinal studies (Table 2) were consistent with the overall theme identified in the cross-sectional studies, finding increased levels of body fat predicted future musculoskeletal pain. Higher baseline FMI was predictive of foot pain in the short term (less than 3 years) [16, 19] in data from both a community cohort (OR 1.06, 95% CI 1.02–1.11) and a musculoskeletal study (OR 1.28, 95% CI 1.04–1.57). In the knee, Jin et al. [55] found an association between increased fat mass and an increased relative risk (RR) of pain in either lying in bed, (RR 1.47, 95% CI 1.12–1.93) or sitting (RR 1.46, 95% CI 1.10–1.95), although knee pain when weight-bearing was not associated with fat mass. More frequent knee pain at 5.1 years follow-up was positively associated with higher total fat mass, and there was an increased risk (95% CI) of consistent (RR 1.89, 95% CI 1.43–2.51) and fluctuating knee pain (RR 1.78, 95% CI 1.41–2.25). A five-year longitudinal study by Pan et al. [35] found a significant trend across three time-points for fat mass and multisite pain, with the number of painful sites significantly associated with total body fat mass over 5 years. There was, however, some discordance between the relationship of body composition and low-back pain, but the larger studies found fat mass to be a predictor of increased pain and disability following multiple adjustments [56‐58]. A twin study by Dario et al. [57] did not find a significant relationship between body fat and the risk of chronic low-back pain in women (n = 314), however a larger study (n = 4986) by Hussain et al. [56] found higher body fat at baseline to be predictive of both high pain intensity and high disability in women and men at 5 years follow-up. Hashimoto et al. [58] also found that men in the fourth quartile of body fat percentage had a significant risk of chronic back pain at > 20 years follow up when adjusting for age, smoking, alcohol consumption and maximal oxygen uptake (OR 2.12, 95% CI 1.13–3.98). One study found that the risk of developing injury during a three-month training program increased in women with an increased body fat percentage (OR 1.16, 95% CI 1.00–1.34) [54].