Identifying genetic predictors of outcomes for Veterans with chronic low back pain and lumbosacral spinal disorders

Low back pain (LBP) is the #1 contributor to disability globally and the 4th most prevalent reason for new VA disability compensation. The societal burden of LBP is largely attributed to 2 distinct subgroups of patients: (1) those who use healthcare resources for chronic (persistent or recurrent) LBP; and (2) those undergoing

surgical treatments for specific spine-related conditions associated with LBP and/or neuropathic symptoms/signs, such as lumbosacral radicular syndrome (LSRS) and symptomatic lumbar spinal stenosis (SLSS). Personalized approaches to improve the efficiency of care and treatment outcomes for these subgroups of Veterans have the potential to reduce the burden of LBP for the Veteran population. Stratified

care for LBP based on prognosis showed early promise when linked to clinical decisions regarding physical therapy. More robust effects from stratified care may come through improving the feasibility and prognostic ability of risk stratification or linking risk stratification to clinical decisions regarding treatments with large

magnitude effects in subgroups of patients with LBP (e.g., decompression surgery for LSRS). The proposed research will apply these two approaches to improving stratified care for LBP, which will develop and validate powerful prediction models using clinical electronic health record (EHR) and genomic data.

This research will two parts to achieve each of the two study aims. Part I will involve genome-wide association study (GWAS) meta-analyses to predict outcomes for LBP-associated conditions, including participants from the Million Veteran Program (MVP), the Electronic Medical Records and Genomics Network phase 3

(eMERGE3) network, and the UK Biobank, as well as summary data from other genomic biobanks. Part II will involve the development and validation of multivariable prognostic models for LBP-related outcomes. First, multivariable prognostic models will be developed using a cross-validation approach in 80% of the MVP

sample, using only clinical data (visits, diagnoses, pharmacy, vital signs, etc.) from the VA EHR; only genomic data (genome-wide PRSs); and both clinical and genomic data. Next, the best-performing multivariable models developed in each aim will be validated in an independent 20% sample of MVP participants, the eMERGE network phase 3, and UK Biobank. Aim 1. Develop and validate prognostic

models for the risk of chronic LBP with healthcare use (CLBP-HU) in Veterans. These models will identify Veterans with LBP of substantial impact sufficient to warrant healthcare use, who should be prioritized for rehabilitative pain treatments. GWAS of CLBP-HU will be conducted. Validated variants will be characterized

and their potential biological roles examined. Multivariable models for predicting CLBP-HU will then be developed and compared with each other. These models will be informed by (a) EHR-defined clinical data, (b) genomic data (genome-wide PRSs), and (c) both clinical and genomic data. Hypothesis: prognostic models for

predicting CLBP-HU will have acceptable discrimination (area under the receiver operating characteristic curve [AUC] ≥ 0.75). The best-performing models will then be validated in other samples. Aim 2. Develop and validate prognostic models for the risk of failure of non-operative treatment (surgical decompression)

in two LBP subgroups: (1) Veterans with LSRS and (2) Veterans with SLSS. The same approach will be followed as used for GWAS and model development in Aim 1. Models developed in Aim 2 will identify Veterans at high risk for progression to decompression surgery, for whom prolonged rehabilitation (e.g., physical

therapy) is unlikely to be successful. Hypothesis: prognostic models for predicting decompression surgery using genomic data only will have acceptable discrimination (AUC ≥ 0.75).