Identifying lower limb specific and generalised joint hypermobility in adults: validation of the Lower Limb Assessment Score

A total of 112 participants were recruited across three groups that were selected to represent varying levels of joint hypermobility. They comprised people with: i) normal mobility (controls), ii) likely hypermobility and iii) known symptomatic hypermobility. The control group were healthy individuals who had not undertaken long-term training likely to affect joint mobility such as dance, gymnastics and acrobatics. These participants were recruited from the University of Sydney Cumberland Campus using flyers on noticeboards. The likely hypermobile group were comprised of elite dancers who were either studying dance at tertiary level with a majority of their training in ballet or who were currently dancing professionally (working as a dancer full-time). They were recruited via flyers, emails, and verbal invitations by a liaison officer from the Western Australian Academy of Performing Arts (Edith Cowan University) and the West Australian Ballet (Perth). The known hypermobile group were people with a prior medical diagnosis of JHS/EDS-HT under the previous nosology, and were recruited from Australian JHS/EDS-HT patient support groups on Facebook. For all groups, participants between the ages of 18–40 years of either gender were included. Volunteers older than 40 years were excluded due to the possibility that joint range could be affected by degenerative disease [3]. Participants across all groups were also excluded if diagnosed with another hereditary disorder of connective tissue (e.g. Marfan syndrome, osteogenesis imperfecta or other subtypes of EDS), or if they had a current orthopaedic condition that might restrict normal range of motion.

All involvement was voluntary and participants provided written informed consent prior to data collection. Ethical approval was gained from the Sydney University [Protocol No. 2015/465] and Edith Cowan University [Protocol No. 13174] Human Research Ethics Committees.

A survey was used to collect information on age, gender, ethnicity, clinical history, current presentation and self-perceived hypermobility. All participants attended a single physical testing session where they were assessed using i) the LLAS (Additional file 1) to provide a maximum score out of 12 for each limb [13] and ii) the Beighton Score (Additional file 2), providing a single score out of 9 [3]. All assessments were completed in the same order without prior warm up and were conducted by experienced physiotherapists (CC and LN) with over 30 years of combined clinical expertise in the domains of sports and musculoskeletal physiotherapy. Each participant was also categorised as hypermobile, borderline hypermobile or normal according to the clinical opinion of these physiotherapists. Given there remains some subjectivity in the sole use of the Beighton score for determining GJH, expert clinical opinion was used [1]. This clinical judgment was based on a comprehensive history, the validated five-part historical hypermobility questionnaire [16], Beighton score, and the physical assessment of peripheral joints, not using the LLAS. Cut-off scores of ≥4/9 and ≥5/9 on the Beighton score were used for men and women respectively in the control and known hypermobile cohorts [2]. A higher score of ≥6/9 was considered GJH in the likely hypermobile cohort to account for the debated palms to floor measure in dancers [17, 18].

In addition, to determine the inter-rater reliability of the LLAS assessments the data from the first 20 control and known hypermobile participants assessed by both examiners were analysed. Given that the objective was to determine whether the LLAS is a robust measure for clinical practice, intra-rater reliability was not of major concern for this study. Furthermore, logistically it was not possible for many of the EDS-HT/JHS participants to return for a second testing session. The inter-rater reliability was assessed using a two-way random absolute agreement, single measures intraclass correlation coefficient. This revealed good to excellent inter-rater reliability (ICC_2,1 = 0.85, 95% CI = 0.67 to 0.94, p < 0.001) [19]. Given the small sample size, a Bland and Altman plot was also used similarly demonstrating no significant difference between the results of the two examiners (limits of agreement = −2.94 and 2.44, 95% C.I. = −0.89 to 0.39, p = 0.43). Therefore, a single examiner was used to assess the remaining participants.

Statistical analysis of collected data was performed using SPSS Version 21 (IBM, NY, USA). The data were inspected for outliers and tested for normality using calculations of skewness. Descriptive statistical analysis was performed on age, gender and ethnicity. Pearson’s correlation coefficient was used to test the association between the LLAS and age. Pearson’s (r) correlations of ≤0.35 were considered to be weak, 0.36–0.67 moderate, and ≥0.68 strong [20]. A paired samples t-test was used to compare the LLAS scores obtained from the left and right limbs.

To test the null hypothesis of no significant difference being detectable between the three groups when using the LLAS, a multivariate analysis of variance (MANOVA) was performed, followed by post-hoc analyses (Tukey HSD) to reveal where the significant differences occurred between the three groups. Effect sizes (partial η²) greater than 0.01 were considered small differences, greater than 0.06 were considered medium differences, and values greater than 0.14 were representative of a large difference between groups [21].

The cut-off score to represent hypermobility when using the LLAS was estimated using median and inter-quartile ranges for each group. The sensitivity and specificity for each level of the LLAS was tested using Receiver Operator Characteristic (ROC) curve analysis. A cut-off point that maximised specificity whilst optimising sensitivity was selected in order to minimise false positive errors [22]. The positive and negative likelihood ratios were calculated for the determined cut-off point, combining the utility of sensitivity and specificity [22]. Clinical opinion was used as the gold standard for this analysis, and the adults who were initially categorised as “borderline hypermobile,” were reclassified as normal [13]. This was based on the decision that if the clinician was uncertain of the classification, then the participant was less likely to be clinically hypermobile. Finally, the percent agreement between the LLAS (using the newly established cut-off point) and clinical opinion was calculated to determine if it could be used to accurately predict GJH. Clinical opinion was used as the gold standard for the identification of GJH.

For this study, a p-value <0.05 was considered significant. It was powered at 90% to minimise the possibility of type 2 errors, with α at 2.5% (type 1 error rate). The probability of GJH in the general population ranges from 15 to 30% [7, 8] so an average would be 15% (i.e. 0.15 pA), whilst the probability of hypermobility in the experimental cohort (likely hypermobile and diagnosed JHS/EDS-HS groups combined) is high, with a conservative estimate of 60% (i.e. 0.6 pB). Therefore, a minimum of 31 controls and 62 hypermobile participants (31 in likely and 31 in known) were needed to be able to reject the null hypothesis that the LLAS is equivocal in both groups at this level of power (PS Power and Sample Size Calculations Version 3.0, 2009).

The percentage of adults at each level of the LLAS (total score/12) was examined (Fig. 1).

Observation of the distribution of scores in Fig. 1 suggests that the LLAS was able to discriminate clearly between all groups, with overlap mostly at scores of four and six out of 12. The MANOVA confirmed a statistically significant difference between the three groups when assessed using the LLAS (Pillai’s Trace = 0.45, F(4.0, 218.0) = 15.8, p < 0.001). There was a large difference between the groups with an effect size (partial η²) of 0.23. The post-hoc Tukey analysis revealed statistically significant mean differences in LLAS scores between all three groups: comparing the controls to the likely hypermobile cohort (MD = −1.95, 95% CI = −2.93 to −0.97, p < 0.001) and to the known hypermobile cohort (MD = −3.43, 95% CI = −4.47 to −2.39, p < 0.001), and comparing the likely hypermobile cohort to the known hypermobile cohort (MD = −1.48, 95% CI = −2.52 to −0.44, p = 0.003).

The median and interquartile ranges for the LLAS across the three clinically defined groups are represented in Fig. 2. This figure suggests that a cut-off score of seven would clearly differentiate the controls from the known hypermobiles, with minimal overlap with the borderline individuals. This observation was confirmed by the ROC Curve presented in Fig. 3. The most suitable cut-off score was seven, where sensitivity was 0.68 and specificity was 0.86 (Table 2). This produced a positive likelihood ratio of 4.84 and a negative likelihood ratio of 0.37.

Given the predominance of Caucasian individuals in the study’s cohort, the same analysis was repeated for the 88 Caucasian participants. The most suitable cut-off score for this group was seven, with a sensitivity of 0.68 and specificity of 0.84.

Having identified a cut-off score of ≥7/12 for the identification of lower limb hypermobility, the LLAS identified lower limb hypermobility in one out of the 40 (3%) control, 16/40 (40%) likely hypermobile and 21/32 (66%) known hypermobile participants. High levels of agreement were found between the LLAS and clinical opinion across each cohort, especially in the control group at 98% agreement, followed by the likely hypermobile group at 70% and the known hypermobile group at 69% (Table 3).

Difficulties arise when validating the LLAS due to the lack of a gold standard for GJH classification. While the Beighton score has gained international acceptance for the identification of GJH, the variation in cut-off scores used and the lack of a standardised protocol limits the comparability across reliability studies [27, 28]. Clinical opinion was therefore used in this study as the most appropriate alternative for comparison, however future research into a standardised investigative procedure for GJH is warranted.

The generalisability of the cut-off score in this study is limited to a predominantly young female and Caucasian adult population. A higher proportion of male participants and a higher mean age would also be useful to more accurately reflect the older adult population. Additionally, this study was not designed to investigate the LLAS cut-off scores in differing ethnicities. Therefore investigation into the use of the LLAS across varying racial groups would be beneficial. Further, whilst the recruitment across three groups in this study was not equal as intended, the sample size allowed the study to remain sufficiently powered to detect between group differences.

Future research into the clinimetric properties of the LLAS would be of value to determine which of the 12 tests are clinically most useful, and might assist the revision of the LLAS into a more efficient assessment tool. Finally, as the examiners in this study are experienced musculoskeletal physiotherapists, the inter-rater reliability of the LLAS when implemented by less experienced assessors should be investigated.