Sex differences in lifetime prevalence of low back pain: A multinational study of opposite‐sex twin pairs

Low back pain (LBP) is more likely to occur in people with a family history of this condition, highlighting the importance of accounting for familial factors when studying the individual risk of LBP. We conducted a study of opposite‐sex twin pairs investigating sex differences in LBP while accounting for (genetic and shared environmental) familial factors.


| INTRODUCTION
Low back pain (LBP) is one of the most challenging healthcare problems in our society.For perspective, LBP is the biggest cause of years lived with disability (YLD) worldwide (James et al., 2018), and the burden and costs associated with this condition are expected to increase in the coming decades (Hartvigsen et al., 2018).Approximately 80% of the adult population experience an LBP episode at some point in their lives (Walker et al., 2004).As a consequence, chronic LBP is strongly associated with an increased number of visits to healthcare providers (Shmagel et al., 2016).LBP has also been associated with social isolation, loss of employment and subsequent financial struggle (Hartvigsen et al., 2018).
Whether sex plays a role in differences in the lifetime prevalence of LBP, however, remains unresolved.A systematic review of 28 studies with data from 488,000 participants showed a global prevalence of LBP 50% higher for women than men (Meucci et al., 2015).However, potential sex differences in the lifetime prevalence of LBP are likely to be related to age, given existing evidence of a higher prevalence of LBP for females after menopause, which typically occurs around the age of 49-50 years (Wáng et al., 2016).Obesity and depression, which may occur at different rates for males and females, have been associated with a higher risk of LBP, further complicating the understanding of the epidemiologic profile of this important condition (Carroll et al., 2004;Hoy et al., 2010;Pinheiro et al., 2015;Shiri et al., 2010).
A family history of LBP (i.e., having a close relative with a previous diagnosis) also increases the risk of future LBP episodes (Matsui et al., 1997).There is consistent evidence that individual differences in the risk of lifetime prevalence of LBP in a given population are, at least partially, explained by genetic and common environmental factors shared by family members, such as twin pairs (Battié et al., 2007;Ferreira et al., 2013;Livshits et al., 2011).Similarly, genetic factors have been implicated in other musculoskeletal traits such as muscle strength, lean mass, and bone density (Seeman et al., 1996).Using a matched co-twin study design applied to data from opposite-sex twin pairs allows for the investigation of sex differences in traits and conditions of interest while controlling for such genetic and nongenetic familial confounders, such as factors shared by family members that are associated with both the exposure and the outcome (Calais-Ferreira et al., 2021).
All opposite-sex twin pairs are dizygotic; thus, they share, on average, 50% of their genetic variance, as well as early environmental factors such as in-utero exposures and social determinants of health, including education, family income, and access to healthcare (Craig et al., 2020).While previous studies have shown some differences between opposite-sex and same-sex twin pairs for traits such as birthweight and gestational age (Silventoinen et al., 2021), and educational attainment (Silventoinen et al., 2021), evidence of important differences for most health outcomes remains scarce (Ahrenfeldt et al., 2020).
Opposite-sex twin studies are, therefore, an essential part of an already comprehensive repertoire of twin study designs that can help answer relevant questions in musculoskeletal health (Calais-Ferreira et al., 2018).We have established a multinational consortium of twin registries with data from opposite-sex DZ twin pairs to (1) investigate the association between female sex and risk of compared to their male co-twins in the Spanish sample (adjusted), but a sex association was not found in the Australian nor US samples.

Conclusions:
We found no evidence of the association between sex and LBP in our merged sample.Between-population differences (i.e.cultural background or health system characteristics) are likely to be major factors leading to variation in the sex association with LBP when familial factors are accounted for.
Significance: Our study of adult opposite-sex twin pairs found no evidence of an association between female sex and lifetime prevalence of low back pain after controlling for familial factors in the merged sample from Australia, Spain and USA, contrary to findings from previous studies of unrelated individuals.Our findings indicate potentially relevant between-country genetic, cultural and environmental differences which may need to be considered for optimal and individualized strategies for the prevention and management of low back pain across the lifespan.lifetime prevalence of LBP while controlling for genetic and nongenetic familial factors, and (2) assess if this association differs by age.

| Design and setting
This study employed a cross-sectional, matched co-twin control design to data of complete DZ opposite-sex twin pairs from Twins Research Australia (TRA), the Murcia Twin Registry (MTR) in Spain, and the Washington State Twin Registry (WSTR) in the United States (US).Twins Research Australia (TRA), formerly known as the Australian Twin Registry, is a volunteer-based registry of Australian twins of all ages and zygosities (Murphy et al., 2019).Participants for the current study were recruited from those completing TRA's Health and Lifestyle Questionnaire, which was administered from 2012 to 2016, and only included complete opposite-sex twin pairs where both twins answered a binary question about a previous medical diagnosis of LBP.The MTR is a population-based registry of twins born between 1940 and 1966 in the region of Murcia, Spain, with a focus on twins aged 40 or older (Ordoñana et al., 2019).For this study, we used data from all opposite-sex twin pairs who participated in MTR's wave 3 of data collection in 2013.The WSTR is a community-based twin registry in the state of Washington, US (Duncan et al., 2019).Data from the WSTR were collected between 2009 and 2013.All pairs without complete data on LBP were excluded from the analysis.

| Variable ascertainment
The main study outcome was the lifetime prevalence of LBP (binary), ascertained through a survey question used by each of the twin registries.LBP ascertainment in the TRA and WSTR was done through questions related to a LBP medical diagnosis, while in the MTR, the question was related to seeking medical help for LBP.Table 1 provides an overview of the survey questions used for the ascertainment of LBP and covariates, including medical diagnosis of depression and current smoking status, both used as binary variables.Sex was ascertained as a binary variable at registration or at survey participation for each registry, and female sex was treated as the exposure in regression models.Bodymass-index (BMI) was derived as a continuous variable from self-reported height and weight in the registries' surveys.

| Statistical analysis
We used descriptive statistics for age, BMI, current smoking status, depression, and LBP separately for females and males.We reported p-values of paired t-tests and univariable conditional logistic regression to assess paired differences between males and females for each of these variables.We conducted dip tests (Hartigan & Hartigan, 1985) to assess modality in the age distributions.p-values were not adjusted for multiple comparisons.
We used mixed-effects logistic regression to investigate the within-pair association between female sex (compared with male sex) and lifetime prevalence of LBP while controlling for familial confounding in univariable (unadjusted) and multivariable (adjusted) models.In the multivariable models, covariates were included based on existing published evidence of their association with sex and/or LBP.Recent systematic reviews have shown that BMI (Shiri et al., 2010), and experiencing symptoms of depression (Pinheiro et al., 2015) are associated with LBP.Therefore, BMI and depression were included in the multivariable models.To address potential heterogeneity between registries, we have accounted for the withinregistry, as well as within-pair clustering, in our mixedeffects models for the analysis of the merged sample.
To study whether the within-pair association between female sex and lifetime prevalence of LBP differed by age (i.e., statistical interaction on the additive scale between female sex and age), we fitted an additional model with an interactive term between female sex and age dichotomised into <median and ≥ median years of age.The mixed-effects logistic regression models only included individuals with complete data for all covariates included in each model.
To assess heterogeneity between registries in the association of sex with LBP, we obtained the I 2 and Q statistics from a random-effects meta-analysis, and formally tested the interaction between the sex and registry.Finally, we conducted a sensitivity analysis using a different LBP outcome for the MTR data to test if the LBP ascertainment in the MTR, which was slightly different than the other two twin registries included in the study, affected our results.This new variable was obtained through the question "Have you ever suffered from chronic low back pain".All data analyses were performed using STATA statistical software version 16.0 (StataCorp., 2019).

| Ethics approvals
All recruitment and data collection procedures were approved by each registry's respective ethics committees.TRA's ethics approval was obtained from the University of Melbourne's Health Research Ethics Committee (HREC #11228).MTR's ethics approvals were obtained from the University of Murcia Research Ethics Committee (#30-1/2009).WSTR's ethics were obtained from the Washington State University IRB (#14515).Informed consent was obtained from all participants.

| RESULTS
The final study sample consisted of 238 twins from TRA, 422 from the MTR, and 930 from the WSTR, for a total sample size of 1590 participants in 795 complete pairs.The final sample was obtained after excluding a total of 130 individuals (65 pairs) from the MTR and 150 individuals (75 pairs) from the WSTR due to being part of twin pairs with missing LBP data for at least one member of the pair.Descriptive sample characteristics are described below and shown in Table 2.

| Descriptive statistics
The mean age of twins from TRA was 51.8 (standard deviation, SD = 15.3),compared with 42.4 (SD = 18.1) for the WSTR, and 56.0 (SD = 6.6) years for the MTR, yielding a mean age of 47.4 (SD = 16.5) for the merged sample.Age distributions were different between registries, so we generated histograms and kernel density curves of the age distribution of LBP cases for males and females separately by each registry and the merged sample (Figure 1).Of note, female LBP cases in the WSTR seemed to follow two distinct 'normal' distributions, with peak reported lifetime LBP prevalence at ages 25-30 and 60-65, while the LBP distribution for males was less heterogeneous.We found evidence to reject unimodality for the age distributions in the US (p < 0.001), Spanish (p = 0.003), and merged (p < 0.001) samples.T A B L E 1 Description of studied variables according to each twin registry.
BMI was higher for males compared to females in the TRA (p = 0.035), MTR (p = 0.003), and merged (p = 0.001) samples, but no difference was found in the WSTR data.Men also had a higher prevalence of current smoking in the MTR (p = 0.04) and merged (p = 0.01) samples, with 16.8% and 21.0% self-reported current smoking status for females and males for the latter, respectively.Females had a higher self-reported prevalence of medically diagnosed depression in the WSTR, MTR, and merged samples (p < 0.001), for a total of 27.7% of females reporting depression in the merged sample, compared to only 14.9% in males.In this descriptive analysis, we only found evidence of sex differences in LBP between males and females in the MTR sample, with an LBP prevalence of 46.0% for females compared to 24.6% for males.Figure 2 includes a mosaic plot of the distribution of low back pain cases and non-cases across all twin registries included in the study.

| Familial similarity
The tetrachoric within-pair correlation for LBP in the merged sample was 0.19, and the odds ratio of the association between co-twin's history of LBP and own LBP was 1.64 (95%CI: 1.32-2.03).

| Within-pair associations between sex and LBP
No within-pair associations between female sex and LBP were found in the Australian, US, or merged samples.In the Spanish sample, females had 2.37 higher odds (95% Confidence Interval: 1.48-3.78) of lifetime prevalence of LBP compared to their male co-twins after adjusting for covariates, representing a relatively small attenuation in the OR the unadjusted model (odds ratio = 2.67, 95%CI: 1.71-4.15)(Table 3).Figure 3 illustrates the adjusted withinpair associations for each registry and the merged sample.
Our sensitivity analysis (see Table S1, Supplementary Material) confirmed the main results from our withinpair analysis; we found a within-pair association between female sex and LBP in the adjusted model (aOR = 2.03, 95%CI: 1.29-3.18)for the Spanish data, with a point estimate slightly lower than in the main results.Consistent with the main results, no association was found for the merged sample.The pair-wise correlation between the two LBP measures in the Spanish sample was 0.94.

| Sex and age interactions
When including age as a binary variable (<median age vs. > = median age), we found evidence that the withinpair association between female sex and LBP differed between those two age groups for the US sample but not for the Australian, Spanish or merged samples.In the US sample, the OR was 1.80 (95%CI: 1.10-2.95)for the younger (<median age) pairs, and 0.35 (95%CI: 0.18-0.66)for the older (> = median age) pairs, p = 0.001.The interaction analysis can be found in the supplementary material (Table S2).

| Heterogeneity
In the meta-analysis between sites, we found that a large proportion of the variance in the association between sex and LBP was explained by between-registry heterogeneity (I 2 = 87.4%),with a Q test with p < 0.01.We found an interactive term of 3.13 (95%CI: 1.82-5.36,p < 0.001) on the odds ratio scale when testing the interaction between female sex and LBP when comparing the MTR with the WSTR (baseline) samples.

DISCUSSION
In this multinational LBP study of opposite-sex twin pairs using a matched design that controls for familial confounding, we did not find evidence of a sex association with LBP in the Australian, US, or merged samples.However, we found that females had more than double the odds of lifetime prevalence of LBP compared to their T A B L E 3 Univariable and multivariable within-pair association between female sex and low back pain using mixed-effects logistic regression.
F I G U R E 3 Odds ratios (and 95% confidence intervals) for the within-pair association between female sex and low back pain (LBP), adjusted for BMI and depression, in opposite-sex twin pairs from Twins Research Australia (TRA), Washington State Twin Registry (WSTR), and Murcia Twin Registry (MTR).
male co-twins in the Spanish sample, adjusting for BMI and depression.The increased odds of LBP for females in the Spanish sample were largely consistent with published evidence of Spanish non-twins with a similar mean age of 54 years, which found that the LBP prevalence for females was 25.6% higher than males, with an adjusted OR of 1.40 (95%CI: 1.34-1.52)(Palacios-Ceña et al., 2021).Moreover, this finding is also consistent with previous evidence of a higher prevalence of LBP for females after menopause (Wáng et al., 2016) and with women being more likely than men to seek care for LBP (Ferreira et al., 2010).
We found consistent evidence of heterogeneity between registries in the association between sex and LBP.Differences in the age distribution of each registry might have played a role in such observed between-country discrepancies, which was further supported by the evidence to reject unimodality for the US, Spanish and merged samples.The Spanish sample was older (mean age of 56 years, minimum age of 47 years), with most women likely to have entered menopause at the time of self-reported lifetime prevalence of LBP.The US sample had a mean age of 42 years (minimum age 19 years), with the age distribution of LBP largely divided into two distinct groups: a 20-25 age group with higher LBP prevalence for females and a 40+ age group with only a slightly higher prevalence for males.
Consistently, we found that the association between female sex and LBP differed by age (as a binary variable) in the US sample, with the association being positive for younger pairs and negative for older pairs.Differences in occupation between males and females in different countries and at different ages could be a factor explaining the distribution of LBP prevalence, given evidence that high physical workload is associated with low back and neckshoulder pain even after genetic and shared environmental confounders were accounted for (Nyman et al., 2009).
Controlling for familial confounding when studying LBP is critical, given previous examples that the associations between birthweight (Leite et al., 2019), adult obesity (Dario et al., 2016), educational attainment (Zadro et al., 2017b), and physical activity (Zadro et al., 2017a) with LBP attenuated to the null after unmeasured familial factors were accounted for.For example, a study of female twin pairs in the UK found an OR of 2.19 (95%CI: 1.33-3.60)for LBP for those with compared to those without an affected DZ co-twin (Livshits et al., 2011).
In general, cultural and lifestyle-related differences between countries, along with within-and betweenpopulation genetic variation, might offer a plausible explanation for why controlling for familial factors might produce different results in different contexts.Familial confounding might also explain why major factors that contribute to the onset of LBP, other than a previous history of LBP itself, have not yet been clearly identified (Taylor et al., 2014).Published estimates of the heritability of LBP, or the proportion of total LBP variance in a given age and population that is explained by genes, varied from 0% to 67% in a systematic review (Ferreira et al., 2013) which further highlights how the relative importance of genetic and nongenetic (shared and unique environmental) factors in explaining individual differences of LBP might be context-specific.
Furthermore, the LBP ascertainment in our study was defined as a person receiving a medical diagnosis or seeking medical attention for LBP, which is likely to be associated with access to healthcare.For this reason, the shared environment in each country (which is controlled for in our within-pair models) was likely to be different between each sample.Previous evidence that the heritability of LBP might also vary according to age (Hartvigsen et al., 2005;Hestbaek et al., 2004) and sex (Hartvigsen et al., 2009) further compounds the argument that the heritability of LBP should not be seen as a fixed estimate of genetic contributions to variance, but only a snapshot of genetically-caused variation that is dependent on environmental factors and the variance structure in each studied population.A previous study has shown that the heritability of different LBP measures (i.e.lifetime prevalence of LBP, severity of LBP, chronicity of LBP) can vary (Battié et al., 2007).Using a more heritable LBP measure could have changed our results.
Our study's main strength was the application of an innovative design with opposite-sex twin pairs, which controls for familial factors in the association between sex and LBP.While this design remains underutilized in medical research, it has contributed to evidence for sex differences in a number of different traits and conditions (Calais-Ferreira et al., 2021;Kendler & Gardner, 2014).The integration of data from twin registries from different countries with diverse genetic backgrounds and societal structural characteristics (i.e., access to healthcare) also allowed for richer analysis and interpretations of findings.
Our study also had some limitations.First, the ascertainment of the lifetime prevalence of LBP was slightly different between the MTR, which used a question related to seeking medical help for LBP, and the WSTR and TRA, which used questions related to receiving a medical diagnosis for LBP.Given the evidence that women are more likely to seek help for LBP (Ferreira et al., 2010), this might have added variability in the results of our merged sample analysis.We have attempted to address this issue in our sensitivity analysis using a different LBP definition for the MTR sample (self-reported lifetime prevalence of chronic LBP), and the results were largely consistent with our main analysis.
The cross-sectional of our study, along with the limited availability of measured individual-level potential confounders, restrict potential causal interpretations from our findings.Therefore, the odds ratios in our study should be interpreted only as associations, especially estimates related to covariates in multivariable models.
While we cannot rule out the possibility of residual genetic confounding and of individual-level environmental confounders that we did not have available data for, our study design with opposite-sex twin pairs matched on age, shared environment, and 50% of genes, offers a robust method to assess sex differences in the lifetime prevalence of LBP that are more likely to be consistent with causation.Finally, we had limited means to assess how representative our sample was of the general population of opposite-sex twin pairs, same-sex twins, or even singletons in various countries.In our withinpair analysis, only twin pairs who were discordant for LBP were included, so our findings are more directly useful for inference about people at higher familial risk of LBP.In general, observational research is more likely to recruit healthier individuals, but this was less likely to have been an issue in our study, given that both the US and Spanish samples, which formed the majority of studied twin pairs in our analysis, used population or community-based recruitment strategies (Duncan et al., 2019;Ordoñana et al., 2019).

| CONCLUSIONS
We studied opposite-sex twin pairs and did not find a within-pair association between female sex and lifetime prevalence of LBP in the merged sample, but did find this association in the Spanish sample.Between-population differences (such as cultural issues or health system characteristics) are likely to be major factors leading to variation in the sex association with LBP when familial factors are accounted for.Understanding potentially causal sex differences in LBP is important from a clinical and public health perspective because the evidence supports the concept that the effects of interventions to ameliorate LBP symptoms might differ by sex (Jensen et al., 2001).Therefore, further research is required for a better understanding of sex differences in the lifetime prevalence of LBP in the presence of familial risk factors, as their implications for improved prevention and treatment of LBP get clearer.

AUTHOR CONTRIBUTIONS
LCF co-designed the study, conducted the statistical analysis, interpreted findings, and produced the final draft of the manuscript.DP co-designed the study, interpreted findings, produced the first draft, and revised the manuscript.MBP interpreted findings and revised the manuscript.FMB interpreted findings and revised the manuscript.JRO supervised data collection, interpreted findings and revised the manuscript.GED supervised data collection, interpreted findings and revised the manuscript.JLH supervised data collection, interpreted findings, and revised the manuscript.PHF (equal senior last author) co-designed and supervised the study, interpreted findings and revised the manuscript.MF (equal senior last author) co-designed and supervised the study, interpreted findings, and revised the manuscript.

F
Age distribution (and kernel density curve) of low back pain cases from Twins Research Australia (TRA), Washington State Twin Registry (WSTR), Murcia Twin Registry (MTR), and merged sample.

F
Mosaic plot of distribution of LBP cases for males and females across samples from Twins Research Australia (TRA), Washington State Twin Registry (WSTR), and Murcia Twin Registry (MTR).sex, BMI and depression as covariates; Note: Only individuals with complete data for all covariates in each model were included in the analysis.