A multi-trait approach improves polygenic risk scores for chronic back pain across population-based and clinically ascertained samples

A multi-trait approach improves polygenic risk scores for chronic back pain across population-based and clinically ascertained samples | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search A multi-trait approach improves polygenic risk scores for chronic back pain across population-based and clinically ascertained samples View ORCID Profile Rachael O. Osagie , Goodarz Koli Farhood , View ORCID Profile Marc Parisien , Amandeep Kaur , Hsuan Megan Tsao , Benjamin Kaufman , Justin Pelletier , Claude Bhérer , View ORCID Profile Audrey V. Grant , Carolina B. Meloto doi: https://doi.org/10.1101/2025.08.27.25334588 Rachael O. Osagie 1 Faculty of Dental Medicine and Oral Health Sciences, McGill University , QC, Canada 2 The Alan Edwards Centre for Research on Pain, McGill University , QC, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Rachael O. Osagie Goodarz Koli Farhood 1 Faculty of Dental Medicine and Oral Health Sciences, McGill University , QC, Canada 2 The Alan Edwards Centre for Research on Pain, McGill University , QC, Canada 3 Department of Anesthesia, Faculty of Medicine and Health Sciences, McGill University , Montreal, QC, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Marc Parisien 1 Faculty of Dental Medicine and Oral Health Sciences, McGill University , QC, Canada 2 The Alan Edwards Centre for Research on Pain, McGill University , QC, Canada 3 Department of Anesthesia, Faculty of Medicine and Health Sciences, McGill University , Montreal, QC, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Marc Parisien Amandeep Kaur 2 The Alan Edwards Centre for Research on Pain, McGill University , QC, Canada 4 Department of Human Genetics, McGill University , QC, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Hsuan Megan Tsao 2 The Alan Edwards Centre for Research on Pain, McGill University , QC, Canada 4 Department of Human Genetics, McGill University , QC, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Benjamin Kaufman 4 Department of Human Genetics, McGill University , QC, Canada 5 Canada Excellence Research Chair in Genomic Medicine 6 Victor Phillip Dahdaleh Institute of Genomic Medicine at McGill University Find this author on Google Scholar Find this author on PubMed Search for this author on this site Justin Pelletier 4 Department of Human Genetics, McGill University , QC, Canada 5 Canada Excellence Research Chair in Genomic Medicine 6 Victor Phillip Dahdaleh Institute of Genomic Medicine at McGill University Find this author on Google Scholar Find this author on PubMed Search for this author on this site Claude Bhérer 4 Department of Human Genetics, McGill University , QC, Canada 5 Canada Excellence Research Chair in Genomic Medicine 6 Victor Phillip Dahdaleh Institute of Genomic Medicine at McGill University Find this author on Google Scholar Find this author on PubMed Search for this author on this site Audrey V. Grant 1 Faculty of Dental Medicine and Oral Health Sciences, McGill University , QC, Canada 2 The Alan Edwards Centre for Research on Pain, McGill University , QC, Canada 3 Department of Anesthesia, Faculty of Medicine and Health Sciences, McGill University , Montreal, QC, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Audrey V. Grant For correspondence: audrey.grant{at}mcgill.ca Carolina B. Meloto 1 Faculty of Dental Medicine and Oral Health Sciences, McGill University , QC, Canada 2 The Alan Edwards Centre for Research on Pain, McGill University , QC, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site Abstract Full Text Info/History Metrics Supplementary material Data/Code Preview PDF Abstract Chronic back pain (CBP) is a complex, heritable condition, and a leading cause of global disability. Previous genome-wide (GW) CBP polygenic risk scores (PRS) derived from a large-scale cohort have shown low discrimination without clinical validation. To improve PRS performance and clinical relevance, we applied Multi-Trait Analysis of GWAS (MTAG) to summary statistics from five genetically correlated traits of European-ancestry individuals with UK Biobank (UKB) CBP as the primary trait, including dorsalgia and chronic musculoskeletal pain (N(effective)=492,717). For comparison, we also constructed a single-trait PRS using UK CBP-only GW data (N=234,013). PRS construction parameters were optimized in an independent large-scale cohort, the Canadian Longitudinal Study on Aging (CLSA) via five-fold cross-validation using LD clumping and p-value thresholding. With covariate adjustment, the MTAG-PRS achieved an AUC of 0.603 (AUC = 0.621; AUPRC = 0.346; R² = 0.051) that was slightly better than the UKB-only PRS (AUC = 0.604; AUPRC = 0.330; R² = 0.038). External validation in CBP cases and controls from another large-scale cohort CARTaGENE) confirmed the MTAG-PRS robustness (AUC = 0.638; AUPRC = 0.335; R² = 0.064). Validation in clinician-ascertained CBP cases (GENE-PAR study) contrasted against an independent subset of CARTaGENE controls improved the MTAG-PRS performance beyond the threshold for clinical utility (AUC = 0.785; AUPRC = 0.616; R² = 0.306). GENE-PAR CBP cases in the top decile PRS also displayed greater burden of CBP symptoms. These findings demonstrate that leveraging genetic pleiotropy, coupled with rigorous phenotyping, moved CBP PRS to clinical utility. Introduction Chronic back pain (CBP) is a prevalent, complex, and heritable condition that imposes a substantial burden on individuals, healthcare systems, and society at large. Globally, it is a leading cause of disability with far-reaching socioeconomic consequences[ 16 ]. Twin and family studies estimate CBP heritability at 40% to 68%[ 5 ; 8 ; 43 ; 59 ]. Nevertheless, polygenic risk scores (PRS), which aggregate risk alleles weighted by their estimated effect sizes, currently explain only a modest fraction of this heritability[ 58 ; 73 ], consistent with CBP’s highly polygenic architecture and the additional contribution of non-genetic influences, such as environmental and psychosocial factors[ 22 ; 26 ; 27 ; 68 ; 69 ]. The first genome-wide CBP PRS[ 73 ], derived from a single-trait UK Biobank study, achieved modest discrimination (AUC ≈ 0.56) and was evaluated only in population-based samples, underscoring the need for clinical validation. Genome-wide association studies (GWAS) have progressively identified risk loci associated with CBP, expanding from the initial discovery of three loci ( SOX5 , CCDC26/GSDMC , DCC ) in ∼158,000 Europeans to over 100 genomic regions in a multi-ancestry meta-analysis of more than 550,000 individuals[ 8 ; 67 ; 69 ]. However, as with many complex traits, these loci individually explain only a small fraction of the phenotypic variance. Empirical observations across diverse complex traits suggest that individual SNPs typically account for less than 0.05% of a trait’s variance, underscoring a highly polygenic architecture where genetic risk is distributed across thousands of small-effect variants[ 10 ; 66 ]. The highly distributed nature of the genetic signal not only necessitates large sample sizes but also highlights the value of multi-trait genetic analysis approaches, which leverage genetic correlation with related traits to increase discovery power and improve risk prediction[ 3 ]. CBP shares substantial, genome-wide genetic correlations with anthropometric, musculoskeletal, and neuropsychiatric traits[ 7 ; 26 ; 79 ]. Multi-trait GWAS frameworks[ 30 ; 37 ; 42 ; 60 ; 74 ] leverage this overlap by jointly analysing correlated phenotypes. Multi-Trait Analysis of GWAS (MTAG) is a meta-analysis approach with a primary trait of interest and other secondary traits that boost power to detect association with the primary trait. MTAG models the genetic covariance between traits, increases the effective sample size, elevates sub-threshold associations to genome-wide significance, and returns trait-specific effect estimates suitable for PRS construction[ 48 ; 74 ]. Empirical applications of multi-trait GWAS methods to back pain phenotypes have reported additional loci and improvements in predictive accuracy[ 6 ; 7 ; 79 ]. However, translation of MTAG-derived CBP summary estimates into clinically informative PRS for physician-ascertained cohorts remains unexplored. Therefore, we sought to leverage traits genetically correlated to CBP using MTAG to derive a CBP PRS and assess its predictive accuracy and phenotype variance explained in a large population-based sample. Next, we evaluated the external validity and clinical relevance of the MTAG-derived PRS in an independent, clinician-ascertained sample, including comparisons of pain-related clinical features across genetic risk strata. MATERIALS AND METHODS Study Overview We combined five GWAS summary statistics, i.e., CBP from the United Kingdom Biobank[ 28 ] (UKB-CBP), CBP from the Cohorts for Heart and Aging Research in Genomic Epidemiology[ 69 ] (CHARGE-UKB CBP), dorsalgia from the FinnGen study[ 44 ], dorsalgia from the Million Veteran Program[ 38 ], and chronic musculoskeletal pain (CMSKP) from the UKB (UKB CMSKP), and applied MTAG to boost CBP locus discovery and account for population overlap. We then used the MTAG-derived CBP summary statistics to build a PRS in the Canadian Longitudinal Study on Aging (CLSA) via five-fold cross-validation, selected the optimal clumping parameters based on the area under the receiver operating characteristic curve (AUC), and applied those parameters to score all CLSA participants. Performance was further evaluated first in CARTaGENE, a population-based case-control cohort, and then in a separate dataset composed of clinician-ascertained CBP cases (GENE-PAR) and an independent subset of CARTaGENE controls. Finally, within the clinician-ascertained sample (GENE-PAR), we compared LBP characteristics and comorbid symptoms burden between the top and bottom deciles of the PRS. Study Populations This study leveraged genome-wide data from four consortia, yielding five discovery GWAS summary statistics that served as base data for PRS construction. All discovery GWAS analyses and subsequent PRS analyses were restricted to participants of European ancestry ( Supplementary Figure 1 ). A summary of cohort characteristics, including sample sizes and case definitions, is provided in Supplementary Table 1 . We also describe the three target datasets used for PRS evaluation, including one internal and two external datasets. Discovery Dataset 1: The United Kingdom Biobank – Chronic Back Pain (UKB-CBP) The UKB is a large-scale, multimodal population-based study that involves approximately half a million participants aged 40-69 years at recruitment (2006-2010). UKB genotyping, imputation, and quality control (QC) are described elsewhere[ 28 ]. We defined two case-control phenotypes using baseline, self-reported pain measures. Briefly, CBP cases (n = 70,643) were individuals who reported back pain that interfered with usual activities in the last month (Field 6159) and, in a conditional follow-up question, indicated that their back pain had lasted for three months or longer (Field 3571). Controls (n = 163,854) were participants who reported no pain at any body site. Individuals who reported back pain lasting less than 3 months or selected “prefer not to answer” were excluded. Discovery Dataset 2: The United Kingdom Biobank – Chronic Musculoskeletal Pain (UKB-CMSKP) In the UKB, CMSKP cases (n = 156,235) were defined as chronic pain (≥3 months) in at least one of four sites (back, knee, hip, neck/shoulder). Controls (n = 154,045) reported no pain at any site. UKB genotyping, imputation, and quality control (QC) are described elsewhere[ 28 ]. Discovery Dataset 3: The Cohorts for Heart and Aging Research in Genomic Epidemiology - Chronic Back Pain (CHARGE - UKB) The CHARGE - UKB meta-analysis pooled data from 16 population-based cohorts (15 CHARGE cohorts and one from the UKB interim release; n = 29,531 cases, n = 128,494 controls) and harmonized CBP definition across cohorts as follows: (1) ≥3 months of back pain, (2) ≥6 months of back pain, or (3) ≥1 month of back pain per year in consecutive years (≈ 12 months total). Controls were those with no back pain or pain of insufficient duration[ 69 ]. Discovery Dataset 4: FinnGen Study The FinnGen project ( https://www.finngen.fi/en ) integrates genotype data with national longitudinal health registry data for over 500,000 Finnish individuals. Phenotype definitions in FinnGen were defined using the Finnish version of the International Classification of Diseases (ICD-10) and harmonized with ICD-8 and ICD-9 where applicable to incorporate historical health records[ 44 ]. Dorsalgia cases were defined using the curated FinnGen endpoint M13_DORSALGIA, which corresponds to ICD-10 code M54 and encompasses a range of back pain diagnoses, including low back pain, sciatica, radiculopathy, cervicalgia, thoracic spine pain, and unspecified dorsalgia. We utilized GWAS summary statistics from Data Freeze 12, comprising 83,888 cases of dorsalgia and 353,224 controls. Controls were defined as individuals without any diagnosis under the broader M13_DORSOPATHY category. Genotyping, imputation, and quality control procedures have been described previously[ 44 ]. Discovery Dataset 5: The Veterans Affairs (VA) Million Veteran Program (MVP) We used GWAS summary statistics for dorsalgia in MVP. These data were obtained from a publicly available meta-analysis conducted by FinnGen, which combined results across MVP, FinnGen, and UKB cohorts (available at https://mvp-ukbb.finngen.fi ). Specifically, we extracted the summary statistics corresponding to the MVP cohort by selecting the columns labeled “MVP_EUR” within the meta-analysis results. This subset included 221,759 dorsalgia cases and 187,718 controls. Phenotype harmonization across cohorts was conducted by the FinnGen consortium. Endpoints from UKB were included only if they exactly matched FinnGen definitions, ensuring strict alignment. FinnGen permitted inclusion of imperfectly matching endpoints if sample overlap exceeded 90%, while MVP endpoints were included regardless of matching status. Concordance of effect sizes at top loci was evaluated across cohorts to support phenotype alignment; however, no formal heterogeneity statistics or exclusion thresholds based on effect size discordance were reported. Genotyping, imputation, and QC procedures in MVP have been described elsewhere[ 38 ]. Target Dataset 1: The Canadian Longitudinal Study on Aging (CLSA) The CLSA is a multimodal, prospective study that enrolled ∼50,000 subjects aged 45-85 years, primarily of European ancestry (93%). Details on CLSA genotyping, imputation, and QC are described elsewhere[ 24 ; 64 ]. At baseline, the comprehensive CLSA cohort comprised 30,097 participants with complete questionnaire data. Of these, 5,109 participants reported back pain on most days for at least one month; indicated that the pain had persisted for three months or longer (duration ≥ 0.25 years); and confirmed that the pain had occurred within the previous year. Pain-free controls (n = 13,577) reported no history of back pain of this duration and stated that they were usually free of pain or discomfort. After merging the genetic and phenotypic data, we retained 3,201 participants with CBP and 9,484 control participants of European ancestry for the analysis. Target Dataset 2: CARTaGENE CARTaGENE is a prospective, population-based cohort designed to investigate the link between genetic factors and chronic diseases in Quebec men and women aged 40-69 at recruitment[ 4 ]. Participants completed questionnaires, underwent physical measurements, and provided biological samples. Sociodemographic and lifestyle questionnaires were self-administered, and a trained interviewer administered the health questionnaire. Genotyping and QC information are described elsewhere[ 4 ; 41 ]. We used data collected about pain to identify individuals with CBP and pain-free controls. CBP cases (n = 632) were defined as participants who answered “yes” to the question: “Has a doctor ever told you that you have chronic back pain?” within the musculoskeletal/connective-tissue domain. Pain-free controls (n = 3,339) were participants who (1) reported being “usually free of pain or discomfort”, and (2) answered “no” to any doctor-diagnosed condition in the musculoskeletal/connective-tissue list. This dataset was used in two independent validation contrasts to calculate PRS performance metrics. The CARTaGENE - controls subset comprised pain-free individuals only that were used as controls for GENE-PAR cases, given that the latter is a case-only cohort and both populations are from the same geolocation and share similar genetic architecture ( Supplementary Figure 2 ). After removing related individuals, we randomly sampled 864 CARTaGENE pain-free controls to compose this subset, achieving a 1:3 case-control ratio[ 40 ] relative to the 288 GENE-PAR chronic low back pain cases (n in the GENE-PAR + CARTaGENE – controls = 1,152). The CARTaGENE – case-control subset was comprised of cases and controls from CARTaGENE only. From the remaining 2,475 pain-free controls, we randomly sampled 2,319 to be contrasted against the 632 CARTaGENE cases to achieve a similar 1:3 case-control ratio (n in the CARTaGENE – case-control subset = 2,951) and ensure comparable PRS metrics. Participants with “don’t know” or “prefer not to answer” responses to any relevant item were excluded. Target Dataset 3: The GENEtic Variants Associated with PAin Reduction (GENE-PAR) Study The GENE-PAR Study comprises a subset of individuals with clinician-ascertained chronic low back pain (CLBP), i.e., with a clinical diagnosis of CLBP given by a pain specialist, who participated in the Quebec Pain Registry (QPR, www.quebecpainregistry.com )[ 15 ]. Briefly, our team contacted QPR CLBP participants via mail and collected data and samples from them between 2015-2016. Participation in the GENE-PAR Study entailed donating a saliva sample, completing a questionnaire (the Canadian adaptation of the NIH Minimum Dataset for LBP research[ 20 ; 45 ]), and returning both to our team using a pre-paid envelope included with the mail. In total, we recruited 349 QPR CLBP participants, of which 327 provided completed questionnaires. We confirmed their CLBP status as per the NIH classification criteria using the first two questions of the Canadian adaptation of the NIH Minimum Dataset for LBP research (i.e., LBP that has been an ongoing problem for at least 3 months and has resulted in a problem on at least half of the days in the past 6 months)[ 20 ; 21 ]. Additional data collected with the Minimum Dataset included: average pain intensity in the past 7 days (0-10 numerical rating scale), the presence or absence of pain that spreads down the leg (sciatica pain), the presence or absence of one or more bothersome painful comorbidities (stomach pain, pain in arms, legs, or joints, headaches, and widespread pain; being bothered a little or a lot by one of these conditions characterized the presence of bothersome painful comorbidity), history of LBP surgery (i.e., having ever had a LBP surgery), pain interference (the PROMIS Short-Form-4a pain interference score)[ 1 ], ongoing treatments for LBP (opioids, infiltrations/injections, exercise therapy, or psychological counselling), LBP-related work absenteeism (yes to being off work or unemployed for 1 month or more due to LBP), LBP-related workers’ compensation benefits (yes to receiving or having applied for disability or workers’ compensation benefits because of LBP), physical function (the PROMIS Short-Form-4a physical function score)[ 1 ], depression (the PROMIS Short-Form-4a emotional distress or depression score)[ 1 ], sleep disturbance (the PROMIS Short-Form-4a sleep disturbance score)[ 1 ], kinesiophobia (agreeing to “it is not really safe for a person with my low back problem to be physically active”)[ 12 ; 34 ], pain catastrophizing (agreeing to “I feel that my low back pain is terrible and it’s never going to get any better”)[ 12 ; 34 ], history of LBP-related lawsuits and legal claims (being involved in a lawsuit or legal claim related to LBP), substance abuse (answering sometimes or often to “have you consumed alcohol or used drugs more than you meant to?” or to “have you felt you wanted or needed to cut down on your drinking or drug abuse?”). Next, the minimum dataset includes questions on sociodemographic data (i.e., age, sex, race, educational attainment, and employment status), as well as smoking status, and height and weight that were used to derive the body mass index (BMI). Neuropathic pain features were screened with the Douleur Neuropathique-4 (DN-4) questionnaire[ 9 ], and somatic symptom burden was quantified using the Somatization subscale of the Symptom Checklist-90 (SCL-90)[ 19 ]. These measures were used to characterize this clinical group of individuals with CLBP and to evaluate how their PRS scores relate to CLBP symptoms and comorbid symptoms. DNA was extracted by the Genome Quebec Innovation Centre using QIAsymphony. Genotyping will be done by the Genome Quebec Innovation Centre, using the Axiom® Precision Medicine Research Array (Thermo Fisher Scientific). Arrays will be spiked with technical replicates to ascertain the integrity of genotyping plates. Following genetic QC, 288 unrelated participants remained for PRS construction. Target datasets 2 and 3: Genotype Imputation To combine GENE-PAR cases with CARTaGENE controls for PRS construction, we imputed GENE-PAR and CARTaGENE genetic data ( de novo ) as described below. Prior to imputation, chip controls from the GENE-PAR genotype dataset were removed, along with individuals of ambiguous genetic sex, using PLINK v1.96[ 63 ]. The CARTaGENE[ 4 ] genotype dataset was arbitrarily divided into two halves (N1 = 15,500, N2 = 13,056) to accommodate the Michigan Imputation Server’s limit on the number of samples per job[ 18 ]. Variants with a minor-allele frequency under 1%, a variant call rate below 95%, and a significant departure from Hardy-Weinberg Equilibrium (HWE) (p < 10 -25 ) were filtered out[ 72 ]. Samples with a call rate under 95% were excluded from the analysis. Strand alignment and normalization were performed using bcftools, and chromosome names were reformatted to meet the requirements of the Michigan Imputation Server [ 17 ; 18 ]. Genotype imputation was performed on three datasets (the GENE-PAR cases and the two CARTaGENE batches) using the Michigan Imputation Server using the TopMed reference panel (TopMed-r-3)[ 18 ; 71 ]. The Michigan Imputation Server employs a hybrid approach, utilizing Eagle v2 for pre-imputation haplotype phasing and Minimac2 for genotype imputation [ 29 ; 47 ]. In each of the three imputed datasets, variants with imputation quality scores (R²) below 0.6 were excluded. The three file sets were then merged using bcftools, resulting in 19,618,805 shared variants. Principal component analysis was performed on a pruned version of the merged GENE-PAR + CARTaGENE dataset (LD correlation R² > 0.5, window size 50 kb, step size 5 kb; MAF>5%; SNP missingness > 1%; HWE p < 10 -25 ) using PLINK v1.96 and FlashPCA v2[ 2 ; 63 ]. To identify and exclude batch-effected variants in the new merged dataset, a pairwise genome-wide association study was run between samples originating from GENEPAR and those from CARTaGENE[ 72 ]. A logistic regression was run with the first 20 principal components as covariates using PLINK v1.96[ 63 ; 72 ]. The logistic regression scheme to test for batch-effected variants in the merged imputed dataset is as shown below: Computational and statistical analysis GWAS summary statistics We assembled summary statistics for five European-ancestry GWAS (UKB CBP & CMSKP; CHARGE-UKB CBP; FinnGen dorsalgia; MVP dorsalgia). We performed UKB CBP and CMSKP scans using REGENIE v3.2.2[ 52 ], fitting linear mixed-effects models of SNP dosage on phenotype and adjusting for age, age squared, sex, genotyping batch, and PCs 1-40. CHARGE-UKB CBP, FinnGen, and MVP dorsalgia summary statistics were accessed from https://gwasarchive.org , https://r12.finngen.fi , and https://mvp-ukbb.finngen.fi , respectively. SNP-based heritability and Genetic correlations with related traits We applied linkage disequilibrium score regression (LDSC; v1.0.1)[ 13 ] to estimate the genetic correlation between UKB CBP, CHARGE-UKB CBP, FinnGen dorsalgia, MVP dorsalgia, and UKB CMSKP. LDSC estimates genetic correlation by exploiting the relationship between GWAS test statistics and linkage disequilibrium (LD), whereby a SNP’s association statistic reflects both its own causal effect and the effects of variants in LD with it. To ensure consistency across datasets, the effect alleles and non-effect alleles from the GWAS datasets were harmonized using the munge_sumstats.py script. We then used the ldsc.py script to compute SNP heritability and genetic correlations, referencing the pre-computed LD scores from the 1000 Genomes European data[ 32 ]. The detailed analytical procedure can be found at https://github.com/bulik/ldsc/wiki/Heritability-and-Genetic-Correlation . Multi-trait analysis To identify the optimal combination of genetically correlated traits for inclusion in our Multi-Trait Analysis of GWAS (MTAG), we systematically evaluated various trait groupings. Each combination was assessed based on predefined criteria, including improvements in the MTAG mean chi-square statistic, GWAS-equivalent sample size, and enhancements in polygenic risk score (PRS) performance across multiple clumping parameters. Based on these evaluations, we selected a five-trait model comprising UKB baseline CBP, CHARGE-UKB CBP, FinnGen dorsalgia, MVP dorsalgia, and UKB CMSKP for subsequent analyses. MTAG jointly analyzes GWAS summary statistics from genetically correlated traits while accounting for sample overlap using bivariate linkage disequilibrium score regression (LDSC)[ 13 ; 74 ]. This approach provides trait-specific SNP effect estimates and pairwise genetic correlations among the input traits. We first harmonized all GWAS summary statistics to the same genomic coordinate system using the command line tool liftOver ( https://genome-store.ucsc.edu/ ). The final MTAG analysis was restricted to the intersection of variants shared across all five datasets. To evaluate the credibility of MTAG results, we assessed the maximum false discovery rate (MaxFDR) for each trait. MaxFDR reflects the worst-case expected proportion of false discoveries among genome-wide significant findings under the MTAG model[ 74 ]. In addition, following a hold-out approach used in prior large-scale GWAS[ 46 ] we conducted two replication analyses to evaluate the robustness and added value of the MTAG results. In the first analysis, we utilized genome-wide significant SNPs (p < 5 × 10⁻⁸) from the MTAG output for the UKB CBP trait, employing a four-trait MTAG model that excluded MVP dorsalgia to ensure sample independence. In the second analysis, we focused on genome-wide significant SNPs from the MTAG-enhanced CHARGE-UKB CBP results, selected because CHARGE-UKB was the lowest-powered input GWAS in the five-trait model. For both analyses, SNPs were matched by variant ID to the MVP dorsalgia GWAS summary statistics. A SNP was considered replicated if it showed a consistent direction of effect and nominal significance (p < 0.05) in the MVP dataset. Polygenic Risk Score Development We constructed polygenic risk scores (PRS) using the clumping and thresholding (C+T) approach[ 39 ] as implemented in PRSice-2[ 14 ]. After aligning alleles and filtering to autosomal SNPs with MAF ≥ 0.05, a total of 3,757,833 variants were retained for PRS construction. We performed 5-fold cross-validation over 45 PRS models defined by all combinations of LD clump r² thresholds (0.1, 0.5, and 0.7) and window sizes (w c ) (250 kb, 500 kb, 1000 kb) as hyperparameters[ 23 ; 62 ]. Within each fold and clumping configuration, PRSice-2 simultaneously calculated scores at eight predefined p value thresholds specified with --bar-levels (P T = 1, 0.5, 0.4, 0.3, 0.2, 0.1, 0.05, 0.001); the P T maximizing mean AUC across folds was selected. All PRS were standardized to Z-scores (mean = 0, SD = 1) within each fold. We then recalculated for the full sample using the optimal r², w c, and P T settings. We compared mean age between cases and controls in CLSA and GENE-PAR + CARTaGENE controls using Welch’s two-sample unequal variances t-tests to determine whether age bias by phenotype warranted inclusion of a quadratic age term in downstream models. To assess the association between PRS and CBP, we applied logistic regression models in the CLSA cohort, adjusting for age, age squared, sex, and the first 10 principal components (PCs) of ancestry. The model was specified as: To allow for comparison with the recalculated PRS based on the MTAG boosted summary statistics, we derived a second PRS in the CLSA from a single-trait UKB CBP GWAS. Incremental predictive utility was assessed via Nagelkerke’s pseudo-R²: (i) covariates only; (ii) covariates + UKB PRS; (iii) covariates + MTAG PRS. The MTAG gain was defined as the difference between (iii) and (ii), with standard errors estimated by 1000-replicate bootstrap. Discrimination was assessed using AUC by pROC[ 65 ] and area under the precision-recall curve (AUPRC) by PRROC[ 33 ]. AUCs were compared by DeLong’s test and calibration was examined using Brier scores[ 11 ]. To further evaluate predictive performance, we stratified participants into PRS quantiles and estimated odds ratios for each quantile compared to the middle 40-60% group. We additionally assessed case enrichment in the top 5% of the PRS distribution. Predictive performance was evaluated using standardized metrics[ 76 ]. External PRS validation We first assessed the MTAG-derived PRS in the CARTaGENE – case-control subset (n = 2,951; 632 cases) and subsequently in the GENE-PAR + CARTaGENE – controls dataset (n = 1,152; 288 cases). Individual scores were calculated in PRSice-2[ 14 ] using the optimized parameters (r² = 0.5; w c = 1000 kb; P T < 0.1) and standardized to Z-scores for each analysis. We compared standardized PRS between cases and controls using Welch’s two-sample unequal variances t-test and reported the mean difference with 95% CI. To compare genetic risk between groups, we contrasted PRS distributions in cases versus controls via Welch’s two-sample t-test, reporting mean differences with 95% confidence intervals. We then fitted three logistic regression models for CBP status: (1) PRS only (per 1-SD increase); (2) demographics-adjusted (PRS, age, age², sex); and (3) fully adjusted (demographics + PCs 1-10). For each model, we report odds ratios per 1-SD PRS increment, Nagelkerke’s pseudo-R² to quantify variance explained and discrimination metrics (AUC by pROC[ 65 ]; AUPRC by PRROC[ 33 ]). Brier scores[ 11 ] were estimated for all PRS models. We next stratified participants into PRS deciles and tested enrichment of CBP cases in the top versus bottom decile using Fisher’s exact test, deriving decile-specific odds ratios from the resulting 2×2 contingency tables. Finally, we compared clinical and psychosocial characteristics between the bottom and top PRS deciles among GENE-PAR cases [ 25 ; 56 ; 75 ]. Continuous measures were compared by Welch’s unequal variances t-tests, and categorical measures by Fisher’s exact tests. Results GWAS summary statistics The UKB CBP GWAS identified 52 independent significant SNPs (P < 5 × 10⁻⁸), corresponding to 31 lead variants across 30 genomic loci. Estimated effective sample sizes were 234,013 for UKB CBP, 158,010 for CHARGE-UKB CBP, 437,112 for FinnGen dorsalgia, 409,477 for MVP dorsalgia, and 310,280 for UKB CMSKP. A summary of the number of significant single nucleotide polymorphisms (SNPs; P < 5×10⁻⁸), lead SNPs, genomic loci, heritability (h²), and genomic inflation factor (lambda GC) for each trait is detailed in Supplementary Table 2 . SNP-based heritability and Genetic correlations with related traits We used LD Score Regression (LDSC) to estimate the SNP-based heritability ( h 2 ) and pairwise genetic correlations (rg) between CBP in the UKB and four genetically related traits: CHARGE-UKB CBP, dorsalgia from the FinnGen study, dorsalgia from the Million Veteran Program, and UKB CMSKP. The SNP-based heritability estimates on the observed scale varied across traits, with the highest being the UKB CBP phenotype ( h ² = 0.1193, SE = 0.0053). Trait-specific heritability estimates were as follows: CHARGE-UKB CBP: h² = 0.0356 (SE = 0.0034); FinnGen dorsalgia: h ² = 0.0535 (SE = 0.0022); MVP dorsalgia: h ² = 0.0698 (SE = 0.0031); and CMSKP (UKB): h ² = 0.0775 (SE = 0.0034) ( Table 1 ). All traits exhibited acceptable LDSC intercepts (ranging from 1.0003 to 1.2587), indicating that inflation in the test statistics was primarily due to polygenicity rather than confounding bias. Each of the traits showed a strong and statistically significant genetic correlation with UKB CBP, with all rg values ∼ 0.7 ( Figure 1 ). Download figure Open in new tab Figure 1. SNP-based heritability and genetic correlations across back pain-related traits. ( A.) SNP-based heritability estimates (on the observed scale) for chronic back pain (CBP) in the UK Biobank, CBP from the CHARGE-UKB meta-analysis, dorsalgia phenotypes from FinnGen and MVP, and chronic musculoskeletal pain (CMSKP) from the UKB. Bars represent point estimates with standard errors. ( B .) Genetic correlations (rg) between UKB CBP and related traits estimated using bivariate LD Score Regression. Values shown within cells indicate the genetic correlation and standard error. All correlations were statistically significant (p < 0.05). View this table: View inline View popup Download powerpoint Table 1. Predictive Performance of Polygenic Risk Scores (PRS) across Discovery (Base) and Validation (Target) datasets Multi-trait analysis Applying MTAG to these five summary statistics increased the mean chi-square (χ²) for the UKB CBP trait from 1.393 to 1.826, reflecting a genome-wide strengthening of association signals for CBP and indicating improved power and more precise estimation of SNP effects due to the integration of genetically correlated traits. The effective sample size for UKB CBP nearly doubled from 234,013 to 492,694 and the MTAG-boosted CBP phenotype showed an observed-scale h² of 0.2599 (SE = 0.0080). The MTAG weight factors revealed the strongest contribution from UKB CBP itself (1.416), followed by CMSKP (1.169), MVP dorsalgia (1.094), and FinnGen dorsalgia (1.062), with CHARGE-UKB CBP contributing modestly (0.786). Across all five traits, a total of 5,482,652 overlapping SNPs were retained and used in the joint analysis. The MTAG-boosted analysis detected 390 independent significant SNPs in the UKB CBP, including 179 lead SNPs across 156 genomic loci ( Supplementary Table 2 ). To assess type I error control under the MTAG model, we examined the maximum false discovery rate (MaxFDR) for each trait. All traits exhibited low MaxFDR values, consistent with well-controlled false positive rates. Specifically, MaxFDR was 0.00107 for MTAG-boosted UKB CBP, 0.00148 for CHARGE-UKB CBP, 0.00286 for FinnGen dorsalgia, 0.00272 for MVP dorsalgia, and 0.00062 for CMSKP. Among the 4,438 SNPs reaching genome-wide significance in the MTAG-boosted UKB CBP (excluding MVP), 4,429 (99.8%) were available in MVP dorsalgia. Of these, 3,544 (79.9%) exhibited concordant direction of effect and nominal replication significance (P < 0.05). Similarly, for the MTAG-boosted CHARGE-UKB CBP trait, 4,386 of 4,395 significant SNPs (99.8%) were present in MVP, with 3,502 (79.7%) meeting the same replication criteria. Polygenic risk score development CBP PRS were derived in the CLSA using five-fold cross-validation to optimize LD clumping parameters and p-value thresholds. Across 45 PRS models (r² = 0.1, 0.5, 0.7; window = 250, 500, 1,000 kb), the best discrimination was achieved at r² = 0.5 and window = 1,000 kb (AUC = 0.6297 ± 0.0135, Supplementary Table 3 ). The corresponding PRS-only model yielded an AUC of 0.6028 ± 0.0115, and the optimal score included 82,768 ± 19 SNPs (range: 82,741-82,794). Each 1 SD increase in PRS was associated with an OR of 1.455 (95% CI: 1.394-1.518) for CBP. Table 1 summarizes discrimination, explained variances, and overall model calibration for all PRS models. In the CLSA, cases (n = 3,201) and controls (n = 9,484) had nearly identical mean ages (62.1 ± 9.8 vs. 62.2 ± 9.9 years; p = 0.488). The distribution of standardized PRS values in the CLSA revealed a rightward shift among cases compared to controls, indicating that individuals with CBP had higher polygenic scores on average. Cases had a mean PRS 0.40 SD higher than controls (p = 7.6×10⁻¹⁹) ( Figure 2A ). In the unadjusted model, the MTAG-PRS achieved an area under the ROC curve (AUC) of 0.603 (95% CI: 0.588–0.618; AUPRC = 0.329; Nagelkerke R² = 0.037) and a Brier score of 0.184, modestly higher than the UKB CBP-only PRS (AUC = 0.582; AUPRC = 0.312; R² = 0.023; Brier score = 0.186). Fully adjusting the model by adding age, age², sex, and PCs 1-10 increased the MTAG-PRS AUC to 0.617 (AUPRC = 0.343; R² = 0.047), versus 0.600 (AUPRC = 0.328; R² = 0.034) for the UKB-PRS. Figure 2B shows the ROC of the fully adjusted MTAG-PRS (AUC = 0.621; AUPRC = 0.346, Figure 2C ; R² = 0.051), which was slightly better than that of UKB CBP-only PRS (AUC = 0.604; AUPRC = 0.330; R² = 0.038). The MTAG-boosted PRS explained an additional 2.15% of the variance in CBP beyond the covariate-only model, while the additional variance explained by the UKB CBP-only PRS was 1.34%. This corresponds to a 0.81% gain in predictive power. Individuals in the top 5% of the MTAG PRS distribution had more than twice the odds of reporting CBP compared to the middle 40-60% group (OR = 2.11; 95% CI: 1.75-2.53) ( Figure 2D ). Download figure Open in new tab Figure 2. Internal Validation of Polygenic Risk Scores for Chronic Back Pain in the CLSA Cohort (A. ) Density distribution of standardized PRS at optimized parameters (clump r2 =0.5; clump kb =1000; p < 0.1; 82,762 SNPs included) for all cases and controls. (Δmean = 0.36; P = 7.6× 10⁻ 71 , Welch’s t-test). ( B .) ROC curve for the fully adjusted model (AUC = 0.621). ( C .) Precision-recall curve for the same model (AUPRC = 0.35). ( D .) Odds ratios for chronic back pain by PRS quantile bin, comparing MTAG-PRS (red) and UKB-only PRS (blue). The middle quantile (40-60%) was used as the reference group (OR = 1). Error bars represent 95% confidence intervals. External PRS validation Within the CARTaGENE – case-control sample (n = 2,951), CBP cases (n = 632) were significantly older than controls (n = 2,319), with mean ages of 56.98 ± 7.85 versus 54.89 ± 8.05 years, respectively (p = 4.62 × 10⁻⁹). The standardized PRS was significantly higher in cases than controls with a mean difference of 0.40; P = 1.7× 10⁻ 19 ( Supplementary Figure 3A ). The MTAG-PRS achieved an AUC = 0.615 (AUPRC = 0.290; R² = 0.041; Brier score = 0.164) in the unadjusted model, increasing to AUC = 0.636 (AUPRC = 0.329; R² = 0.061; Brier score = 0.161) with demographics and to AUC = 0.638 (AUPRC = 0.335; R² = 0.064; Brier score = 0.161) after full adjustment ( Table 1 , Supplementary Figure 3B, Supplementary Figure 3C ). In the combined GENE-PAR + CARTaGENE – controls dataset, the overall mean age was 56.3 ± 9.1 years (n = 1,152), and this remained identical (56.3 ± 9.1 years) after excluding eight participants with missing covariate data (n = 1,144). Cases were significantly older than controls (60.6 ± 10.3 vs. 54.8 ± 8.2 years; p = 2.8 × 10⁻¹⁶). The standardized PRS was significantly higher in cases than controls (0.441 ± 2.58 vs −0.147 ± 2.40; p = 7.91 × 10⁻¹⁷; Figure 3A ). The MTAG-PRS alone yielded AUC = 0.670 (AUPRC = 0.415; R² = 0.102) ( Table 1 ). Demographics adjustment raised the AUC to 0.760 (AUPRC = 0.584; R² = 0.260), and the fully adjusted model achieved AUC = 0.785 (AUPRC = 0.616; R² = 0.306) ( Figure 3B ; Figure 3C ). Download figure Open in new tab Figure 3. External validation of the chronic back pain polygenic risk score (PRS) in GENE-PAR+CARTAGENE cohort ( A. ) Density of standardized PRS in controls (blue) and cases (red), with dashed lines at group means (Δmean = 0.59; P = 7.9× 10⁻ 17 , Welch’s t-test). ( B.) ROC curve for the fully adjusted model (AUC = 0.785). ( C.) Precision–recall curve for the same model (AUPRC = 0.62). ( D .) Odds of chronic back pain GENE-PAR participants present in the top vs bottom 10% of the PRS distribution (OR = 6.70, 95% CI: 3.60-13.10). Clinical PRS Validation When we stratified the GENE-PAR + CARTaGENE – controls by PRS decile (n = 116 per decile), 51.7% of individuals in the top decile (60 of 116) were cases, compared with 13.8% (16 of 116) in the bottom decile. This enrichment corresponded to an OR of 6.64 (95% CI 3.60-13.10; p = 7.88 × 10⁻¹⁰) for top versus bottom decile ( Figure 3D ). Clinical and psychosocial characteristics of the 76 participants in the GENE-PAR dataset in the extreme PRS deciles (top decile, n = 60; bottom decile, n = 16) are presented in Supplementary Table 4. Participants in the top decile had significantly greater pain interference (65.46 ± 5.73 vs 60.94 ± 6.94; p = 0.026), greater pain intensity (6.33 ± 1.95 vs 4.69 ± 2.39; p = 0.022), were more likely to report pain catastrophizing (66.7% vs 31.2%; p = 0.02), and reported a greater burden of somatic symptoms (SCL “no-pain” sub-scale = 1.57 ± 0.82 vs 1.12 ± 0.72; p = 0.040) compared with those in the bottom decile ( Table 2 ). Seemingly counterintuitively, participants in the top PRS decile also showed better physical function (mean ± SD 38.3 ± 5.8 vs 35.3 ± 4.8; p = 0.047). View this table: View inline View popup Download powerpoint Table 2. Subset of Clinical and Psychosocial Characteristics of GENE-PAR participants in the top and bottom 10% of the PRS distribution Discussion We demonstrated that integrating genetically correlated traits via MTAG enhances PRS performance for CBP. By combining five European-ancestry GWAS (UKB CBP, CHARGE-UKB CBP, FinnGen dorsalgia, MVP dorsalgia, and UKB CMSKP) and accounting for overlapping samples, we nearly doubled the sample size for CBP cases (going from 234,013 to 492,717) and more than doubled its SNP-heritability (going from 11.9% to 26.0% on the observed scale). MTAG identified 156 independent CBP loci, more than fivefold the 30 detected with the UKB CBP-only data. Incorporating genetically correlated phenotypes, namely back pain without a duration criterion (dorsalgia) and chronic musculoskeletal pain, defined using the same duration threshold as the CBP primary trait at 3 or more months across multiple musculoskeletal sites, increased power for otherwise subthreshold CBP signals. Translating these MTAG-derived discoveries into PRS development, we observed substantial improvements in predictive performance across both population-based and clinician-ascertained samples, despite differences in phenotype definitions and sample composition. The larger effective sample size and increased power from MTAG enabled the inclusion of more informative SNPs with stronger effect estimates, which collectively enhanced the ability of PRS to discriminate CBP cases from controls and explained a greater proportion of phenotypic variance. In contrast, earlier CBP GWAS identified only three loci (SOX5, CCDC26/GSDMC, DCC ) in ∼158,000 Europeans[ 69 ], and subsequent meta-analyses conducted by CBP consortia expanded this number to just over 100 loci [ 67 ]. Given CBP’s extreme polygenicity, discovery improves via larger samples and by leveraging cross-trait genetic correlation via MTAG to achieve locus-specific boosts in power where trait effects align. Despite substantial discovery of CBP loci, the predictive performance of CBP PRS remains modest; to date, the only prediction-focused study reported an AUC of 0.56 evaluated in large-scale population-based samples[ 73 ]. In the CLSA, our MTAG-derived PRS produced a slight improvement in AUC (0.603 vs. 0.582), AUPRC (0.329 vs. 0.312), and Nagelkerke’s pseudo-R² (0.037 vs. 0.023) compared to the UKB-only PRS. After adjusting for age, sex, and the first 10 principal component vectors, discrimination rose to AUC = 0.621 (R² = 0.051) versus 0.604 (R² = 0.038) for the UKB CBP-only score. Although numerically small, an absolute AUC gain of ∼0.02 represents a tangible advance in risk stratification for CBP, yielding more high-risk individuals correctly identified at matched specificity. In fact, the MTAG-derived PRS exceeded the ∼0.003 AUC increase produced by an improved knee osteoarthritis PRS[ 54 ]. Brier score improved only marginally (0.184 vs. 0.186; fully adjusted = 0.182). Importantly, in the CARTaGENE-only analysis, the fully adjusted model also showed gains (AUC = 0.638, AUPRC = 0.335, R² = 0.064, Brier = 0.161), although largely less substantial and similar to those seen in the CLSA. External validation in the dataset including clinician-ascertained CLBP cases (GENE-PAR) and CARTaGENE controls yielded substantially greater performance. According to the framework proposed by Swets (1988) for evaluating the clinical utility of predictive tests[ 70 ], our MTAG-PRS alone approached the levels of clinical utility (AUC > 0.7), displaying an AUC of 0.670 (R² = 0.102). Following full adjustment, the AUC surpassed the level of potential clinical utility (0.785; R² = 0.306), an unprecedented finding in chronic pain. Brier score also showed an improvement compared to that obtained in the CLSA (Brier = 0.141 vs. 0.182 in CLSA), indicating better overall accuracy of the predicted probabilities. Combined, these findings demonstrate that CBP PRS benefits substantially from clinical characterization of CBP status. A crucial determinant of PRS accuracy is discovery GWAS sample size: larger training datasets reduce standard errors and sharpen SNP-effect estimates[ 3 ; 57 ]. Leveraging genetically correlated traits (e.g., via MTAG or multi-PGS) further enlarges effective sample size and improves prediction[ 3 ]. These gains also appear when training on broader phenotypes, where increased N boosts PRS performance despite greater phenotypic heterogeneity. The corollary is a trade-off between shallow phenotyping (large N, lower specificity) and deep phenotyping (smaller N, higher specificity). The trade-off between sample size and extent of phenotyping was recently explored in the context of major depressive disorder (MDD) for the base dataset. Using large, minimally phenotyped meta-analysis GWAS summary data (∼246k cases) yielded stronger PRS associations with clinically diagnosed major depressive disorder (MDD) (OR≈1.75 per SD) than a much smaller strictly diagnosed GWAS (OR≈1.14); however, when sample size was equalized, the clinically defined GWAS produced the most specific predictor for clinical MDD[ 53 ]. When considering the target dataset, other studies of various health outcomes indicated that PRSs perform better in rigorously phenotyped samples[ 51 ; 77 ]. Consistent with prior findings in other phenotypes, while enlarging the base GWAS improved locus discovery and slightly increased PRS performance, our findings showed that CBP-specific clinical phenotyping in the target dataset drove a substantial increase in the CBP-PRS performance. Another non-exclusive explanation for higher PRS performance in the target dataset is that recruitment for the GENE-PAR Study may have captured individuals with more severe chronic back pain, as they may have been more prone to respond to an invitation to participate. This could enrich genetic liability and inflate apparent PRS performance, constituting ascertainment bias due to severity enrichment. Despite this, GENE-PAR likely aligns well with the intended target population for a clinical setting PRS application, which is the downstream goal of this work. In addition, in this combined dataset, (GENE-PAR + CARTaGENE – controls), participants spanned a wide age range, and CBP cases were on average substantially older than controls, creating a clear age-related imbalance. Given that CBP prevalence typically follows an “inverted-U” trajectory - rising through mid-life, peaking in the early 60s, and declining thereafter[ 35 ; 36 ], incorporating both linear and quadratic age terms captured this non-linear risk pattern and meaningfully improved model discrimination. By contrast, in CLSA, where participants were older and the age distribution was narrower, there was no significant difference in age between cases and controls, and including age² contributed little additional predictive value. The CARTaGENE subset, which had intermediate age variability, showed smaller but still detectable gains when modeling age non-linearly. Together, these findings suggest that accounting for non-linear age effects enhances prediction in samples with broader age distributions, whereas its impact is minimal in older, more homogeneous populations. PRS decile analysis in the GENE-PAR + CARTaGENE controls also showed that individuals in the top decile had sixfold higher CBP odds versus those in the bottom decile. Notably, among GENE-PAR CLBP cases, top-decile individuals reported greater pain intensity, pain interference, catastrophizing, and a greater burden of somatic symptoms. These measures have consistently been shown to inform cLBP outcomes, including the risk of chronicity[ 34 ; 56 ]. Pragmatically, our findings suggest that PRSs can be used to stratify patients and ultimately inform targeted allocation of healthcare resources or early intervention strategies to minimize the burden of chronic pain. Despite the abovementioned greater burden of CLBP symptoms, participants in the highest PRS decile displayed better physical function. This has been reported before and suggests that these individuals develop adaptive mechanisms of coping or resilience that enable them to remain active despite a heightened symptom burden[ 50 ; 56 ]. Our study has limitations that must be considered. MTAG assumes a single, genome-wide covariance structure of SNP effects; loci with trait-specific architectures may be misestimated. PRS are ancestry-dependent and portability of PRS using a European base population on African ancestry target populations is particularly poor[ 49 ]. The method proposed here can be extended to other ancestries as GWAS of CBP become available. In this study, we constructed PRS via clumping-and-thresholding because it is computationally efficient and enables the systematic exploration of multiple tuning parameters across large-scale datasets. However, other approaches that model linkage disequilibrium more comprehensively have shown improved predictive performance in other complex traits[ 31 ; 55 ; 61 ; 78 ]. Future studies could evaluate whether such strategies further enhance CBP risk prediction within the MTAG framework. Finally, despite significant improvements, our PRS explains only a fraction of CBP heritability, reflecting the influence of environmental, psychosocial, and gene-environment interactions; thus, PRS should complement, not replace, established clinical and lifestyle risk factors in comprehensive CBP risk models. Our study benefits from several methodological strengths. By integrating five genetically correlated pain-related traits using MTAG, we increased discovery power without requiring additional individual-level data. Independent GWAS replication, excluding MVP dorsalgia from the MTAG analysis, showed that ∼80% of lead SNPs remained directionally consistent and nominally significant in MVP, supporting the robustness of our findings. We further optimized PRS construction through five-fold cross-validation in CLSA, minimizing overfitting and improving parameter selection. Finally, external validation in both population-based and clinician-ascertained samples demonstrated the generalizability of our MTAG-derived PRS. In summary, our study demonstrates that integrating genetically correlated traits through MTAG substantially enhances discovery and predictive performance for chronic back pain. By increasing effective sample size and SNP-heritability, we achieved unprecedented levels of PRS discrimination, particularly in clinically characterized samples. These findings establish a proof-of-principle that multi-trait approaches, coupled with rigorous phenotyping, can move CBP PRS toward thresholds of clinical utility. Continued expansion across diverse ancestries, adoption of advanced statistical methods, and integration with clinical and environmental factors will be essential to translate these genetic insights into tools that improve prevention, risk stratification, and personalized management of chronic back pain. Data Availability The data used in this study were obtained from multiple cohorts. UK Biobank data are available to bona fide researchers through application to the UK Biobank ). CLSA data are accessible to qualified researchers through application to the Canadian Longitudinal Study on Aging ). CARTaGENE data are available through application to CARTaGENE. FinnGen summary statistics are publicly available through the FinnGen project (https://www.finngen.fi/en ). MVP summary statistics are available at https://mvp-ukbb.finngen.fi . CHARGE-UKB meta-analysis summary statistics are available at the GWAS Archive (https://gwasarchive.org ) under "Chronic back pain" Summary statistics generated in this study are available from the corresponding author upon reasonable request. ACKNOWLEDGMENTS The authors would like to acknowledge the Canadian Institutes of Health Research (CIHR PJT 190092), that have funded this study. The authors also would like to acknowledge the LAEF for the LAEF PhD Studentship awarded to R.O. The authors have no conflict of interest to declare. The authors thank the participants and investigators of data sources used in this study, including the UK Biobank resource (application #20802), the CLSA (application #190213), the FinnGen study, the MVP, the CHARGE and PainOmics consortia, the CARTaGENE and GENE-PAR. Footnotes ↵ * Co-senior authors. REFERENCES [1]. ↵ PROMIS® (Patient-Reported Outcomes Measurement Information System) , 2023 . [2]. ↵ Abraham G , Qiu Y , Inouye M . FlashPCA2: principal component analysis of Biobank-scale genotype datasets . Bioinformatics 2017 ; 33 ( 17 ): 2776 – 2778 . OpenUrl CrossRef PubMed [3]. ↵ Albinana C , Zhu Z , Schork AJ , Ingason A , Aschard H , Brikell I , Bulik CM , Petersen LV , Agerbo E , Grove J , Nordentoft M , Hougaard DM , Werge T , Borglum AD , Mortensen PB , McGrath JJ , Neale BM , Prive F , Vilhjalmsson BJ . Multi-PGS enhances polygenic prediction by combining 937 polygenic scores . Nat Commun 2023 ; 14 ( 1 ): 4702 . OpenUrl CrossRef PubMed 4. ↵ Awadalla P , Boileau C , Payette Y , Idaghdour Y , Goulet JP , Knoppers B , Hamet P , Laberge C , Project CA . Cohort profile of the CARTaGENE study: Quebec’s population-based biobank for public health and personalized genomics . Int J Epidemiol 2013 ; 42 ( 5 ): 1285 - 1299 . OpenUrl CrossRef PubMed Web of Science [5]. ↵ Battie MC , Videman T , Levalahti E , Gill K , Kaprio J . Heritability of low back pain and the role of disc degeneration . Pain 2007 ; 131 ( 3 ): 272 – 280 . OpenUrl CrossRef PubMed Web of Science [6]. ↵ Belonogova NM , Elgaeva EE , Zorkoltseva IV , Kirichenko AV , Svishcheva GR , Freidin MB , Williams FMK , Suri P , Axenovich TI , Tsepilov YA . A multi-trait approach identified 7 novel genes for back pain . Pain Rep 2025 ; 10 ( 1 ): e1218 . OpenUrl [7]. ↵ Bjornsdottir G , Stefansdottir L , Thorleifsson G , Sulem P , Norland K , Ferkingstad E , Oddsson A , Zink F , Lund SH , Nawaz MS , Bragi Walters G , Skuladottir AT , Gudjonsson SA , Einarsson G , Halldorsson GH , Bjarnadottir V , Sveinbjornsson G , Helgadottir A , Styrkarsdottir U , Gudmundsson LJ , Pedersen OB , Hansen TF , Werge T , Banasik K , Troelsen A , Skou ST , Thorner LW , Erikstrup C , Nielsen KR , Mikkelsen S , Consortium DG , Consortium GO , Jonsdottir I , Bjornsson A , Olafsson IH , Ulfarsson E , Blondal J , Vikingsson A , Brunak S , Ostrowski SR , Ullum H , Thorsteinsdottir U , Stefansson H , Gudbjartsson DF , Thorgeirsson TE , Stefansson K . Rare SLC13A1 variants associate with intervertebral disc disorder highlighting role of sulfate in disc pathology . Nat Commun 2022 ; 13 ( 1 ): 634 . OpenUrl PubMed [8]. ↵ Bortsov AV , Parisien M , Khoury S , Martinsen AE , Lie MU , Heuch I , Hveem K , Zwart JA , Winsvold BS , Diatchenko L . Brain-specific genes contribute to chronic but not to acute back pain . Pain Rep 2022 ; 7 ( 5 ): e1018 . OpenUrl PubMed [9]. ↵ Bouhassira D , Attal N , Alchaar H , Boureau F , Brochet B , Bruxelle J , Cunin G , Fermanian J , Ginies P , Grun-Overdyking A , Jafari-Schluep H , Lanteri-Minet M , Laurent B , Mick G , Serrie A , Valade D , Vicaut E . Comparison of pain syndromes associated with nervous or somatic lesions and development of a new neuropathic pain diagnostic questionnaire (DN4) . Pain 2005 ; 114 ( 1-2 ): 29 – 36 . OpenUrl CrossRef PubMed Web of Science [10]. ↵ Boyle EA , Li YI , Pritchard JK . An expanded view of complex traits: from polygenic to omnigenic . Cell 2017 ; 169 ( 7 ): 1177 – 1186 . OpenUrl CrossRef PubMed [11]. ↵ Brier GW . Verification of forecasts expressed in terms of probability . Mon Wea Rev 1950 ; 78 : 1 – 3 . OpenUrl CrossRef [12]. ↵ Bruyère O , Demoulin M , Brereton C , Humblet F , Flynn D , Hill JC , Maquet D , Van Beveren J , Reginster JY , Crielaard JM , Demoulin C . Translation validation of a new back pain screening questionnaire (the STarT Back Screening Tool) in French . Arch Public Health 2012 ; 70 ( 1 ): 12 . OpenUrl PubMed [13]. ↵ Bulik-Sullivan BK , Loh PR , Finucane HK , Ripke S , Yang J , Schizophrenia Working Group of the Psychiatric Genomics C , Patterson N , Daly MJ , Price AL , Neale BM . LD Score regression distinguishes confounding from polygenicity in genome-wide association studies . Nat Genet 2015 ; 47 ( 3 ): 291 – 295 . OpenUrl CrossRef PubMed [14]. ↵ Choi SW , O’Reilly PF . PRSice-2: Polygenic Risk Score software for biobank-scale data . Gigascience 2019 ; 8 ( 7 ). [15]. ↵ Choiniere M , Ware MA , Page MG , Lacasse A , Lanctot H , Beaudet N , Boulanger A , Bourgault P , Cloutier C , Coupal L , De Koninck Y , Dion D , Dolbec P , Germain L , Martin V , Sarret P , Shir Y , Taillefer MC , Tousignant B , Trepanier A , Truchon R . Development and Implementation of a Registry of Patients Attending Multidisciplinary Pain Treatment Clinics: The Quebec Pain Registry . Pain Res Manag 2017 ; 2017 : 8123812 . OpenUrl PubMed [16]. ↵ Collaborators GBDLBP . Global, regional, and national burden of low back pain, 1990-2020, its attributable risk factors, and projections to 2050: a systematic analysis of the Global Burden of Disease Study 2021 . Lancet Rheumatol 2023 ; 5 ( 6 ): e316 – e329 . OpenUrl CrossRef PubMed [17]. ↵ Danecek P , Bonfield JK , Liddle J , Marshall J , Ohan V , Pollard MO , Whitwham A , Keane T , McCarthy SA , Davies RM . Twelve years of SAMtools and BCFtools . Gigascience 2021 ; 10 ( 2 ): giab008 . OpenUrl CrossRef PubMed [18]. ↵ Das S , Forer L , Schönherr S , Sidore C , Locke AE , Kwong A , Vrieze SI , Chew EY , Levy S , McGue M . Next-generation genotype imputation service and methods . Nature genetics 2016 ; 48 ( 10 ): 1284 – 1287 . OpenUrl CrossRef PubMed [19]. ↵ Derogatis LR , Lipman RS , Covi L . SCL-90: an outpatient psychiatric rating scale--preliminary report . Psychopharmacol Bull 1973 ; 9 ( 1 ): 13 – 28 . OpenUrl PubMed [20]. ↵ Deyo RA , Dworkin SF , Amtmann D , Andersson G , Borenstein D , Carragee E , Carrino J , Chou R , Cook K , DeLitto A . Report of the NIH Task Force on research standards for chronic low back pain . Physical therapy 2015 ; 95 ( 2 ): e1 – e18 . OpenUrl Abstract / FREE Full Text [21]. ↵ Deyo RA , Dworkin SF , Amtmann D , Andersson G , Borenstein D , Carragee E , Carrino J , Chou R , Cook K , DeLitto A , Goertz C , Khalsa P , Loeser J , Mackey S , Panagis J , Rainville J , Tosteson T , Turk D , Von Korff M , Weiner DK . Focus article: report of the NIH Task Force on Research Standards for Chronic Low Back Pain . Eur Spine J 2014 ; 23 ( 10 ): 2028 – 2045 . OpenUrl CrossRef PubMed [22]. ↵ Elgaeva EE , Zorkoltseva IV , Nostaeva AV , Verzun DA , Tiys ES , Timoshchuk AN , Kirichenko AV , Svishcheva GR , Freidin MB , Williams FMK , Suri P , Aulchenko YS , Axenovich TI , Tsepilov YA . Decomposing the genetic background of chronic back pain . Hum Mol Genet 2025 . [23]. ↵ Euesden J , Lewis CM , O’Reilly PF . PRSice: Polygenic Risk Score software . Bioinformatics 2015 ; 31 ( 9 ): 1466 – 1468 . OpenUrl CrossRef PubMed [24]. ↵ Forgetta V , Darmond-Zwaig C , Belisle A , Li R , Balion C , Roshandel D , Ragoussis J . The Canadian Longitudinal Study on Aging Genome-wide Genetic Data on 9,900 Participants . 2018 . [25]. ↵ Foster NE , Thomas E , Bishop A , Dunn KM , Main CJ . Distinctiveness of psychological obstacles to recovery in low back pain patients in primary care . PAIN® 2010 ; 148 ( 3 ): 398 – 406 . OpenUrl [26]. ↵ Freidin MB , Tsepilov YA , Palmer M , Karssen LC , Suri P , Aulchenko YS , Williams FMK , Boer CG , Yau MS , Evans DS , Gelemanovic A , Bartz TM , Nethander M , Arbeeva L , Neogi T , Campbell A , Mellstrom D , Ohlsson C , Marshall LM , Orwoll E , Uitterlinden A , Ratter JI , Lauc G , Psaty BM , Karlsson MK , Lane NE , Jarvik G , Polasek O , Hochberg M , Jordan JM , Van Meurs JBJ , Jackson R , Nielson CM , Mitchell BD , Smith BH , Hayward C , Smith NL , Grp CMW . Insight into the genetic architecture of back pain and its risk factors from a study of 509,000 individuals . Pain 2019 ; 160 ( 6 ): 1361 – 1373 . OpenUrl CrossRef PubMed [27]. ↵ Freidin MB , Tsepilov YA , Stanaway IB , Meng W , Hayward C , Smith BH , Khoury S , Parisien M , Bortsov A , Diatchenko L , Borte S , Winsvold BS , Brumpton BM , Zwart JA , Aulchenko YS , Suri P , Williams FMK , Pain HA-I . Sex- and age-specific genetic analysis of chronic back pain . Pain 2021 ; 162 ( 4 ): 1176 – 1187 . OpenUrl PubMed [28]. ↵ Fry A , Littlejohns TJ , Sudlow C , Doherty N , Adamska L , Sprosen T , Collins R , Allen NE . Comparison of Sociodemographic and Health-Related Characteristics of UK Biobank Participants With Those of the General Population . Am J Epidemiol 2017 ; 186 ( 9 ): 1026 – 1034 . OpenUrl CrossRef PubMed [29]. ↵ Fuchsberger C , Abecasis GR , Hinds DA . minimac2: faster genotype imputation. Bioinformatics (Oxford, England) 2015 ; 31 ( 5 ): 782 – 784 . OpenUrl [30]. ↵ Galesloot TE , van Steen K , Kiemeney LA , Janss LL , Vermeulen SH . A comparison of multivariate genome-wide association methods . PLoS One 2014 ; 9 ( 4 ): e95923 . OpenUrl CrossRef PubMed [31]. ↵ Ge T , Chen CY , Ni Y , Feng YCA , Smoller JW . Polygenic prediction via Bayesian regression and continuous shrinkage priors . Nat Commun 2019 ; 10 . [32]. ↵ Genomes Project C , Auton A , Brooks LD , Durbin RM , Garrison EP , Kang HM , Korbel JO , Marchini JL , McCarthy S , McVean GA , Abecasis GR . A global reference for human genetic variation . Nature 2015 ; 526 ( 7571 ): 68 - 74 . OpenUrl CrossRef PubMed [33]. ↵ Grau J , Grosse I , Keilwagen J . PRROC: computing and visualizing precision-recall and receiver operating characteristic curves in R . Bioinformatics 2015 ; 31 ( 15 ): 2595 – 2597 . OpenUrl CrossRef PubMed [34]. ↵ Hill JC , Dunn KM , Lewis M , Mullis R , Main CJ , Foster NE , Hay EM . A primary care back pain screening tool: identifying patient subgroups for initial treatment . Arthritis Rheum 2008 ; 59 ( 5 ): 632 – 641 . OpenUrl CrossRef PubMed Web of Science [35]. ↵ Hoy D , Bain C , Williams G , March L , Brooks P , Blyth F , Woolf A , Vos T , Buchbinder R . A systematic review of the global prevalence of low back pain . Arthritis Rheum 2012 ; 64 ( 6 ): 2028 – 2037 . OpenUrl CrossRef PubMed Web of Science [36]. ↵ Hoy D , Brooks P , Blyth F , Buchbinder R . The Epidemiology of low back pain . Best Pract Res Clin Rheumatol 2010 ; 24 ( 6 ): 769 – 781 . OpenUrl CrossRef PubMed [37]. ↵ Hu Y , Lu Q , Liu W , Zhang Y , Li M , Zhao H . Joint modeling of genetically correlated diseases and functional annotations increases accuracy of polygenic risk prediction . Plos Genet 2017 ; 13 ( 6 ): e1006836 . OpenUrl CrossRef PubMed [38]. ↵ Hunter-Zinck H , Shi Y , Li M , Gorman BR , Ji SG , Sun N , Webster T , Liem A , Hsieh P , Devineni P , Karnam P , Gong X , Radhakrishnan L , Schmidt J , Assimes TL , Huang J , Pan C , Humphries D , Brophy M , Moser J , Muralidhar S , Huang GD , Przygodzki R , Concato J , Gaziano JM , Gelernter J , O’Donnell CJ , Hauser ER , Zhao H , O’Leary TJ , Program VAMV , Tsao PS , Pyarajan S. Genotyping Array Design and Data Quality Control in the Million Veteran Program . Am J Hum Genet 2020 ; 106 ( 4 ): 535 – 548 . OpenUrl CrossRef PubMed [39]. ↵ International Schizophrenia C , Purcell SM , Wray NR , Stone JL , Visscher PM , O’Donovan MC , Sullivan PF , Sklar P . Common polygenic variation contributes to risk of schizophrenia and bipolar disorder . Nature 2009 ; 460 ( 7256 ): 748 – 752 . OpenUrl CrossRef PubMed Web of Science [40]. ↵ Janes H , Pepe M . The optimal ratio of cases to controls for estimating the classification accuracy of a biomarker . Biostatistics 2006 ; 7 ( 3 ): 456 – 468 . OpenUrl CrossRef PubMed [41]. ↵ Jantzen R , Payette Y , de Malliard T , Labbe C , Noisel N , Broet P . Five-year absolute risk estimates of colorectal cancer based on CCRAT model and polygenic risk scores: A validation study using the Quebec population-based cohort CARTaGENE . Prev Med Rep 2022 ; 25 : 101678 . [42]. ↵ Julienne H , Lechat P , Guillemot V , Lasry C , Yao C , Araud R , Laville V , Vilhjalmsson B , Menager H , Aschard H . JASS: command line and web interface for the joint analysis of GWAS results . NAR Genom Bioinform 2020 ; 2 ( 1 ): lqaa003 . OpenUrl [43]. ↵ Junqueira DR , Ferreira ML , Refshauge K , Maher CG , Hopper JL , Hancock M , Carvalho MG , Ferreira PH . Heritability and lifestyle factors in chronic low back pain: results of the Australian twin low back pain study (The AUTBACK study) . Eur J Pain 2014 ; 18 ( 10 ): 1410 – 1418 . OpenUrl CrossRef PubMed [44]. ↵ Kurki MI , Karjalainen J , Palta P , Sipila TP , Kristiansson K , Donner KM , Reeve MP , Laivuori H , Aavikko M , Kaunisto MA , Loukola A , Lahtela E , Mattsson H , Laiho P , Della Briotta Parolo P , Lehisto AA , Kanai M , Mars N , Ramo J , Kiiskinen T , Heyne HO , Veerapen K , Rueger S , Lemmela S , Zhou W , Ruotsalainen S , Parn K , Hiekkalinna T , Koskelainen S , Paajanen T , Llorens V , Gracia-Tabuenca J , Siirtola H , Reis K , Elnahas AG , Sun B , Foley CN , Aalto-Setala K , Alasoo K , Arvas M , Auro K , Biswas S , Bizaki-Vallaskangas A , Carpen O , Chen CY , Dada OA , Ding Z , Ehm MG , Eklund K , Farkkila M , Finucane H , Ganna A , Ghazal A , Graham RR , Green EM , Hakanen A , Hautalahti M , Hedman AK , Hiltunen M , Hinttala R , Hovatta I , Hu X , Huertas-Vazquez A , Huilaja L , Hunkapiller J , Jacob H , Jensen JN , Joensuu H , John S , Julkunen V , Jung M , Junttila J , Kaarniranta K , Kahonen M , Kajanne R , Kallio L , Kalviainen R , Kaprio J , FinnGen , Kerimov N , Kettunen J , Kilpelainen E , Kilpi T , Klinger K , Kosma VM , Kuopio T , Kurra V , Laisk T , Laukkanen J , Lawless N , Liu A , Longerich S , Magi R , Makela J , Makitie A , Malarstig A , Mannermaa A , Maranville J , Matakidou A , Meretoja T , Mozaffari SV , Niemi MEK , Niemi M , Niiranen T , CJ OD , Obeidat ME , Okafo G , Ollila HM , Palomaki A , Palotie T , Partanen J , Paul DS , Pelkonen M , Pendergrass RK , Petrovski S , Pitkaranta A , Platt A , Pulford D , Punkka E , Pussinen P , Raghavan N , Rahimov F , Rajpal D , Renaud NA , Riley-Gillis B , Rodosthenous R , Saarentaus E , Salminen A , Salminen E , Salomaa V , Schleutker J , Serpi R , Shen HY , Siegel R , Silander K , Siltanen S , Soini S , Soininen H , Sul JH , Tachmazidou I , Tasanen K , Tienari P , Toppila-Salmi S , Tukiainen T , Tuomi T , Turunen JA , Ulirsch JC , Vaura F , Virolainen P , Waring J , Waterworth D , Yang R , Nelis M , Reigo A , Metspalu A , Milani L , Esko T , Fox C , Havulinna AS , Perola M , Ripatti S , Jalanko A , Laitinen T , Makela TP , Plenge R , McCarthy M , Runz H , Daly MJ , Palotie A . FinnGen provides genetic insights from a well-phenotyped isolated population . Nature 2023 ; 613 ( 7944 ): 508 – 518 . OpenUrl CrossRef PubMed [45]. ↵ Lacasse A , Roy JS , Parent AJ , Noushi N , Odenigbo C , Page G , Beaudet N , Choiniere M , Stone LS , Ware MA , Quebec Pain Research Network’s Steering Committee of the Low Back Pain Strategic I. The Canadian minimum dataset for chronic low back pain research: a cross-cultural adaptation of the National Institutes of Health Task Force Research Standards . CMAJ Open 2017 ; 5 ( 1 ): E237 – E248 . OpenUrl Abstract / FREE Full Text [46]. ↵ Lee JJ , Wedow R , Okbay A , Kong E , Maghzian O , Zacher M , Nguyen-Viet TA , Bowers P , Sidorenko J , Karlsson Linnér R . Gene discovery and polygenic prediction from a genome-wide association study of educational attainment in 1.1 million individuals . Nature genetics 2018 ; 50 ( 8 ): 1112 – 1121 . OpenUrl CrossRef PubMed [47]. ↵ Loh P-R , Danecek P , Palamara PF , Fuchsberger C , A Reshef Y , K Finucane H , Schoenherr S , Forer L , McCarthy S , Abecasis GR . Reference-based phasing using the Haplotype Reference Consortium panel . Nature genetics 2016 ; 48 ( 11 ): 1443 – 1448 . OpenUrl CrossRef PubMed [48]. ↵ Maier RM , Zhu Z , Lee SH , Trzaskowski M , Ruderfer DM , Stahl EA , Ripke S , Wray NR , Yang J , Visscher PM . Improving genetic prediction by leveraging genetic correlations among human diseases and traits . Nat Commun 2018 ; 9 ( 1 ): 989 . OpenUrl CrossRef PubMed [49]. ↵ Martin AR , Kanai M , Kamatani Y , Okada Y , Neale BM , Daly MJ . Clinical use of current polygenic risk scores may exacerbate health disparities . Nature Genetics 2019 ; 51 ( 4 ): 584 – 591 . OpenUrl CrossRef PubMed [50]. ↵ Masse-Alarie H , Angarita-Fonseca A , Lacasse A , Page MG , Tetreault P , Fortin M , Leonard G , Stone LS , Roy JS . Low back pain definitions: effect on patient inclusion and clinical profiles . Pain Rep 2022 ; 7 ( 2 ): e997 . OpenUrl PubMed [51]. ↵ Mavaddat N , Michailidou K , Dennis J , Lush M , Fachal L , Lee A , Tyrer JP , Chen TH , Wang Q , Bolla MK , Yang X , Adank MA , Ahearn T , Aittomaki K , Allen J , Andrulis IL , Anton-Culver H , Antonenkova NN , Arndt V , Aronson KJ , Auer PL , Auvinen P , Barrdahl M , Beane Freeman LE , Beckmann MW , Behrens S , Benitez J , Bermisheva M , Bernstein L , Blomqvist C , Bogdanova NV , Bojesen SE , Bonanni B , Borresen-Dale AL , Brauch H , Bremer M , Brenner H , Brentnall A , Brock IW , Brooks-Wilson A , Brucker SY , Bruning T , Burwinkel B , Campa D , Carter BD , Castelao JE , Chanock SJ , Chlebowski R , Christiansen H , Clarke CL , Collee JM , Cordina-Duverger E , Cornelissen S , Couch FJ , Cox A , Cross SS , Czene K , Daly MB , Devilee P , Dork T , Dos-Santos-Silva I , Dumont M , Durcan L , Dwek M , Eccles DM , Ekici AB , Eliassen AH , Ellberg C , Engel C , Eriksson M , Evans DG , Fasching PA , Figueroa J , Fletcher O , Flyger H , Forsti A , Fritschi L , Gabrielson M , Gago-Dominguez M , Gapstur SM , Garcia-Saenz JA , Gaudet MM , Georgoulias V , Giles GG , Gilyazova IR , Glendon G , Goldberg MS , Goldgar DE , Gonzalez-Neira A , Grenaker Alnaes GI , Grip M , Gronwald J , Grundy A , Guenel P , Haeberle L , Hahnen E , Haiman CA , Hakansson N , Hamann U , Hankinson SE , Harkness EF , Hart SN , He W , Hein A , Heyworth J , Hillemanns P , Hollestelle A , Hooning MJ , Hoover RN , Hopper JL , Howell A , Huang G , Humphreys K , Hunter DJ , Jakimovska M , Jakubowska A , Janni W , John EM , Johnson N , Jones ME , Jukkola-Vuorinen A , Jung A , Kaaks R , Kaczmarek K , Kataja V , Keeman R , Kerin MJ , Khusnutdinova E , Kiiski JI , Knight JA , Ko YD , Kosma VM , Koutros S , Kristensen VN , Kruger U , Kuhl T , Lambrechts D , Le Marchand L , Lee E , Lejbkowicz F , Lilyquist J , Lindblom A , Lindstrom S , Lissowska J , Lo WY , Loibl S , Long J , Lubinski J , Lux MP , MacInnis RJ , Maishman T , Makalic E , Maleva Kostovska I , Mannermaa A , Manoukian S , Margolin S , Martens JWM , Martinez ME , Mavroudis D , McLean C , Meindl A , Menon U , Middha P , Miller N , Moreno F , Mulligan AM , Mulot C , Munoz-Garzon VM , Neuhausen SL , Nevanlinna H , Neven P , Newman WG , Nielsen SF , Nordestgaard BG , Norman A , Offit K , Olson JE , Olsson H , Orr N , Pankratz VS , Park-Simon TW , Perez JIA , Perez-Barrios C , Peterlongo P , Peto J , Pinchev M , Plaseska-Karanfilska D , Polley EC , Prentice R , Presneau N , Prokofyeva D , Purrington K , Pylkas K , Rack B , Radice P , Rau-Murthy R , Rennert G , Rennert HS , Rhenius V , Robson M , Romero A , Ruddy KJ , Ruebner M , Saloustros E , Sandler DP , Sawyer EJ , Schmidt DF , Schmutzler RK , Schneeweiss A , Schoemaker MJ , Schumacher F , Schurmann P , Schwentner L , Scott C , Scott RJ , Seynaeve C , Shah M , Sherman ME , Shrubsole MJ , Shu XO , Slager S , Smeets A , Sohn C , Soucy P , Southey MC , Spinelli JJ , Stegmaier C , Stone J , Swerdlow AJ , Tamimi RM , Tapper WJ , Taylor JA , Terry MB , Thone K , Tollenaar R , Tomlinson I , Truong T , Tzardi M , Ulmer HU , Untch M , Vachon CM , van Veen EM , Vijai J , Weinberg CR , Wendt C , Whittemore AS , Wildiers H , Willett W , Winqvist R , Wolk A , Yang XR , Yannoukakos D , Zhang Y , Zheng W , Ziogas A , Investigators A , kConFab AI, Collaborators N , Dunning AM , Thompson DJ , Chenevix-Trench G , Chang-Claude J , Schmidt MK , Hall P , Milne RL , Pharoah PDP , Antoniou AC , Chatterjee N , Kraft P , Garcia-Closas M , Simard J , Easton DF . Polygenic Risk Scores for Prediction of Breast Cancer and Breast Cancer Subtypes . Am J Hum Genet 2019 ; 104 ( 1 ): 21 – 34 . OpenUrl CrossRef PubMed [52]. ↵ Mbatchou J , Barnard L , Backman J , Marcketta A , Kosmicki JA , Ziyatdinov A , Benner C , O’Dushlaine C , Barber M , Boutkov B , Habegger L , Ferreira M , Baras A , Reid J , Abecasis G , Maxwell E , Marchini J . Computationally efficient whole-genome regression for quantitative and binary traits . Nat Genet 2021 ; 53 ( 7 ): 1097 – 1103 . OpenUrl CrossRef PubMed [53]. ↵ Mitchell BL , Thorp JG , Wu Y , Campos AI , Nyholt DR , Gordon SD , Whiteman DC , Olsen CM , Hickie IB , Martin NG , Medland SE , Wray NR , Byrne EM . Polygenic Risk Scores Derived From Varying Definitions of Depression and Risk of Depression . JAMA Psychiatry 2021 ; 78 ( 10 ): 1152 – 1160 . OpenUrl PubMed [54]. ↵ Morita Y , Kamatani Y , Ito H , Ikegawa S , Kawaguchi T , Kawaguchi S , Takahashi M , Terao C , Ito S , Nishitani K . Improved genetic prediction of the risk of knee osteoarthritis using the risk factor-based polygenic score . Arthritis Research & Therapy 2023 ; 25 ( 1 ): 103 . OpenUrl PubMed [55]. ↵ Ni GY , Zeng J , Revez JA , Wang Y , Zheng ZL , Ge T , Restuadi R , Kiewa J , Nyholt DR , Coleman JRI , Smoller JW , Yang J , Visscher PM , Wray NR , Consortium PG . A Comparison of Ten Polygenic Score Methods for Psychiatric Disorders Applied Across Multiple Cohorts . Biol Psychiat 2021 ; 90 ( 9 ): 611 – 620 . OpenUrl CrossRef PubMed [56]. ↵ Osagie RO , Tufa I , Angarita-Fonseca A , Page MG , Lacasse A , Stone LS , Rainville P , Roy M , Tetreault P , Fortin M , Leonard G , Masse-Alarie H , Roy JS , Grant AV , Meloto CB , Quebec Back Pain C. Impact of different acute low back pain definitions on the predictors and on the risk of transition to chronic low back pain: a prospective longitudinal cohort study . Pain 2025 . [57]. ↵ Patel AP , Wang M , Ruan Y , Koyama S , Clarke SL , Yang X , Tcheandjieu C , Agrawal S , Fahed AC , Ellinor PT , Tsao PS , Sun YV , Cho K , Wilson PWF , Assimes TL , van Heel DA , Butterworth AS , Aragam KG , Natarajan P , Khera AV , Genes, Health Research T, the Million Veteran P. A multi-ancestry polygenic risk score improves risk prediction for coronary artery disease . Nature Medicine 2023 ; 29 ( 7 ): 1793 – 1803 . OpenUrl CrossRef PubMed [58]. ↵ Plomin R , von Stumm S . Polygenic scores: prediction versus explanation . Molecular Psychiatry 2022 ; 27 ( 1 ): 49 – 52 . OpenUrl CrossRef PubMed [59]. ↵ Polderman TJ , Benyamin B , de Leeuw CA , Sullivan PF , van Bochoven A , Visscher PM , Posthuma D . Meta-analysis of the heritability of human traits based on fifty years of twin studies . Nat Genet 2015 ; 47 ( 7 ): 702 – 709 . OpenUrl CrossRef PubMed [60]. ↵ Porter HF , O’Reilly PF . Multivariate simulation framework reveals performance of multi-trait GWAS methods . Sci Rep 2017 ; 7 : 38837 . [61]. ↵ Prive F , Arbel J , Vilhjalmsson BJ . LDpred2: better, faster, stronger . Bioinformatics 2020 ; 36 ( 22-23 ): 5424 – 5431 . OpenUrl [62]. ↵ Privé F , Vilhjálmsson BJ , Aschard H , Blum MG . Making the most of clumping and thresholding for polygenic scores . The American journal of human genetics 2019 ; 105 ( 6 ): 1213 – 1221 . OpenUrl CrossRef PubMed [63]. ↵ Purcell S , Neale B , Todd-Brown K , Thomas L , Ferreira MA , Bender D , Maller J , Sklar P , de Bakker PI , Daly MJ , Sham PC . PLINK: a tool set for whole-genome association and population-based linkage analyses . Am J Hum Genet 2007 ; 81 ( 3 ): 559 – 575 . OpenUrl CrossRef PubMed [64]. ↵ Raina PS , Wolfson C , Kirkland SA , Griffith LE , Oremus M , Patterson C , Tuokko H , Penning M , Balion CM , Hogan D , Wister A , Payette H , Shannon H , Brazil K . The Canadian longitudinal study on aging (CLSA) . Can J Aging 2009 ; 28 ( 3 ): 221 – 229 . OpenUrl PubMed [65]. ↵ Robin X , Turck N , Hainard A , Tiberti N , Lisacek F , Sanchez JC , Muller M . pROC: an open-source package for R and S+ to analyze and compare ROC curves . BMC Bioinformatics 2011 ; 12 : 77 . [66]. ↵ Robinson MR , Wray NR , Visscher PM . Explaining additional genetic variation in complex traits . Trends in Genetics 2014 ; 30 ( 4 ): 124 – 132 . OpenUrl CrossRef PubMed [67]. ↵ Stanaway IB , Suri P , Afari N , Dochtermann D , Gerstenberger A , Pyarajan S , Roseen EJ , Program MV , Gasperi M . Multi-ancestry meta-analysis of genome-wide association studies discovers 67 new loci associated with chronic back pain . Nat Commun 2025 ; 16 ( 1 ): 1525 . OpenUrl PubMed [68]. ↵ Suri P , Naeini MK , Heagerty PJ , Freidin MB , Smith IG , Elgaeva EE , Compte R , Tsepilov YA , Williams FMK . The association of lumbar intervertebral disc degeneration with low back pain is modified by underlying genetic propensity to pain . Spine J 2024 . [69]. ↵ Suri P , Palmer MR , Tsepilov YA , Freidin MB , Boer CG , Yau MS , Evans DS , Gelemanovic A , Bartz TM , Nethander M , Arbeeva L , Karssen L , Neogi T , Campbell A , Mellstrom D , Ohlsson C , Marshall LM , Orwoll E , Uitterlinden A , Rotter JI , Lauc G , Psaty BM , Karlsson MK , Lane NE , Jarvik GP , Polasek O , Hochberg M , Jordan JM , Van Meurs JBJ , Jackson R , Nielson CM , Mitchell BD , Smith BH , Hayward C , Smith NL , Aulchenko YS , Williams FMK . Genome-wide meta-analysis of 158,000 individuals of European ancestry identifies three loci associated with chronic back pain . Plos Genet 2018 ; 14 ( 9 ): e1007601 . OpenUrl CrossRef PubMed [70]. ↵ Swets JA . Measuring the accuracy of diagnostic systems . Science 1988 ; 240 ( 4857 ): 1285 – 1293 . OpenUrl Abstract / FREE Full Text [71]. ↵ Taliun D , Harris DN , Kessler MD , Carlson J , Szpiech ZA , Torres R , Taliun SAG , Corvelo A , Gogarten SM , Kang HM . Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program . Nature 2021 ; 590 ( 7845 ): 290 – 299 . OpenUrl CrossRef PubMed [72]. ↵ Tom JA , Reeder J , Forrest WF , Graham RR , Hunkapiller J , Behrens TW , Bhangale TR . Identifying and mitigating batch effects in whole genome sequencing data . BMC bioinformatics 2017 ; 18 ( 1 ): 351 . OpenUrl PubMed [73]. ↵ Tsepilov YA , Elgaeva EE , Nostaeva AV , Compte R , Kuznetsov IA , Karssen LC , Freidin MB , Suri P , Williams FMK , Aulchenko YS . Development and Replication of a Genome-Wide Polygenic Risk Score for Chronic Back Pain . J Pers Med 2023 ; 13 ( 6 ). [74]. ↵ Turley P , Walters RK , Maghzian O , Okbay A , Lee JJ , Fontana MA , Nguyen-Viet TA , Wedow R , Zacher M , Furlotte NA , and Me Research T, Social Science Genetic Association C , Magnusson P , Oskarsson S , Johannesson M , Visscher PM , Laibson D , Cesarini D , Neale BM , Benjamin DJ . Multi-trait analysis of genome-wide association summary statistics using MTAG . Nat Genet 2018 ; 50 ( 2 ): 229 – 237 . OpenUrl CrossRef PubMed [75]. ↵ Von Korff M , Dworkin SF , Le Resche L , Kruger A . An epidemiologic comparison of pain complaints . Pain 1988 ; 32 ( 2 ): 173 – 183 . OpenUrl CrossRef PubMed Web of Science [76]. ↵ Wand H , Lambert SA , Tamburro C , Iacocca MA , O’Sullivan JW , Sillari C , Kullo IJ , Rowley R , Dron JS , Brockman D , Venner E , McCarthy MI , Antoniou AC , Easton DF , Hegele RA , Khera AV , Chatterjee N , Kooperberg C , Edwards K , Vlessis K , Kinnear K , Danesh JN , Parkinson H , Ramos EM , Roberts MC , Ormond KE , Khoury MJ , Janssens A , Goddard KAB , Kraft P , MacArthur JAL , Inouye M , Wojcik GL . Improving reporting standards for polygenic scores in risk prediction studies . Nature 2021 ; 591 ( 7849 ): 211 - 219 . OpenUrl CrossRef PubMed [77]. ↵ Ying S , Heung T , Morrow BE , Thiruvahindrapuram B , Yuen RKC , Bassett AS . Influence of Polygenic Risk on Height and BMI in Adults With a 22q11.2 Microdeletion . J Endocr Soc 2025 ; 9 ( 9 ): bvaf115 . OpenUrl [78]. ↵ Zhang QQ , Prive F , Vilhjalmsson B , Speed D . Improved genetic prediction of complex traits from individual-level data or summary statistics . Nat Commun 2021 ; 12 ( 1 ). [79]. ↵ Zorkoltseva IV , Elgaeva EE , Belonogova NM , Kirichenko AV , Svishcheva GR , Freidin MB , Williams FMK , Suri P , Tsepilov YA , Axenovich TI . Multi-Trait Exome-Wide Association Study of Back Pain-Related Phenotypes . Genes (Basel) 2023 ; 14 ( 10 ). View the discussion thread. Back to top Previous Next Posted August 29, 2025. Download PDF Supplementary Material Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following A multi-trait approach improves polygenic risk scores for chronic back pain across population-based and clinically ascertained samples Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share A multi-trait approach improves polygenic risk scores for chronic back pain across population-based and clinically ascertained samples Rachael O. Osagie , Goodarz Koli Farhood , Marc Parisien , Amandeep Kaur , Hsuan Megan Tsao , Benjamin Kaufman , Justin Pelletier , Claude Bhérer , Audrey V. Grant , Carolina B. Meloto medRxiv 2025.08.27.25334588; doi: https://doi.org/10.1101/2025.08.27.25334588 Share This Article: Copy Citation Tools A multi-trait approach improves polygenic risk scores for chronic back pain across population-based and clinically ascertained samples Rachael O. Osagie , Goodarz Koli Farhood , Marc Parisien , Amandeep Kaur , Hsuan Megan Tsao , Benjamin Kaufman , Justin Pelletier , Claude Bhérer , Audrey V. Grant , Carolina B. Meloto medRxiv 2025.08.27.25334588; doi: https://doi.org/10.1101/2025.08.27.25334588 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Genetic and Genomic Medicine Subject Areas All Articles Addiction Medicine (568) Allergy and Immunology (863) Anesthesia (300) Cardiovascular Medicine (4435) Dentistry and Oral Medicine (444) Dermatology (382) Emergency Medicine (608) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1509) Epidemiology (15228) Forensic Medicine (30) Gastroenterology (1124) Genetic and Genomic Medicine (6599) Geriatric Medicine (668) Health Economics (997) Health Informatics (4536) Health Policy (1368) Health Systems and Quality Improvement (1613) Hematology (540) HIV/AIDS (1264) Infectious Diseases (except HIV/AIDS) (15916) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (146) Nephrology (667) Neurology (6599) Nursing (346) Nutrition (998) Obstetrics and Gynecology (1144) Occupational and Environmental Health (957) Oncology (3332) Ophthalmology (974) Orthopedics (369) Otolaryngology (420) Pain Medicine (436) Palliative Medicine (130) Pathology (663) Pediatrics (1693) Pharmacology and Therapeutics (691) Primary Care Research (711) Psychiatry and Clinical Psychology (5447) Public and Global Health (9231) Radiology and Imaging (2198) Rehabilitation Medicine and Physical Therapy (1370) Respiratory Medicine (1196) Rheumatology (593) Sexual and Reproductive Health (712) Sports Medicine (530) Surgery (712) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a006040a8a98fbb0',t:'MTc3OTU1OTgwOQ=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();