RESUMO
Development of back pain is multifactorial, and it is not well understood which factors are the main drivers of the disease. We therefore applied a machine-learning approach to an existing large cohort study data set and sought to identify and rank the most important contributors to the presence of back pain amongst the documented parameters of the cohort. Data from 399 participants in the KORA-MRI (Cooperative health research in the region Augsburg-magnetic resonance imaging) (Cooperative Health Research in the Region Augsburg) study was analyzed. The data set included MRI images of the whole body, including the spine, metabolic, sociodemographic, anthropometric, and cardiovascular data. The presence of back pain was one of the documented items in this data set. Applying a machine-learning approach to this preexisting data set, we sought to identify the variables that were most strongly associated with back pain. Mediation analysis was performed to evaluate the underlying mechanisms of the identified associations. We found that depression and anxiety were the 2 most selected predictors for back pain in our model. Additionally, body mass index, spinal canal width and disc generation, medium and heavy physical work as well as cardiovascular factors were among the top 10 most selected predictors. Using mediation analysis, we found that the effects of anxiety and depression on the presence of back pain were mainly direct effects that were not mediated by spinal imaging. In summary, we found that psychological factors were the most important predictors of back pain in our cohort. This supports the notion that back pain should be treated in a personalized multidimensional framework. PERSPECTIVE: This article presents a wholistic approach to the problem of back pain. We found that depression and anxiety were the top predictors of back pain in our cohort. This strengthens the case for a multidimensional treatment approach to back pain, possibly with a special emphasis on psychological factors.