INTRODUCTION
There are a number of risk factors- from biological, psychological, and social domains- for non-specific chronic low back pain (cLBP). Many cLBP treatments target risk factors on the assumption that the targeted factor is not just associated with cLBP but is also a cause (i.e, a causal risk factor). In most cases this is a strong assumption, primarily due to the possibility of confounding variables. False assumptions about the causal relationships between risk factors and cLBP likely contribute to the generally marginal results from cLBP treatments. The objective of this study was to identify causal relationships between risk factors and cLBP by accounting for confounding bias.
METHODS
Baseline data from 1,376 participants in BACKHOME, a longitudinal observational e-Cohort of U.S. adults with cLBP that is part of the NIH Back Pain Consortium (BACPAC) Research Program, were analyzed. The primary outcome was the Pain, Enjoyment of Life, and General Activity (PEG) Scale (range: 0-10). Six risk factors were selected based on prior evidence, including Mendelian randomization1-5 and/or mediation analyses,6 supporting a causal association with cLBP: fear avoidance, catastrophizing, smoking, sleep disturbance, depression, and obesity. Confounders were: age, sex, education, financial strain, opioid use, self-efficacy, anxiety, cognitive function, physical function, fatigue, social isolation, and social role. Type A confounders were identified using the ESC-DAG approach,7 a method for building directed acyclic graphs based on causal criteria. Type B confounders were chosen using weaker, change-in-criterion evidence. Separate regression coefficients were estimated for each risk factor: unadjusted, minimally sufficiently adjusted (Type A confounders only); and fully adjusted (Type A and B). A single model with all six risk factors adjusted for common Type A confounders was also estimated.
RESULTS
Participants had the following characteristics: age 54.9 ± 14.4 years, 67.4% female, 60% never smokers, 29.9% overweight, 39.5% obese, PROMIS sleep disturbance t-score 54.8 ± 8.0, PROMIS depression t-score 52.6 ± 10.1, Fear-avoidance Beliefs Questionnaire 11.6 ± 5.9, Patient Catastrophizing Scale 4.5 ± 2.6, PEG 4.4 ± 2.2.
The unadjusted, minimally-adjusted (Type A confounders), and fully-adjusted (Type A and B confounders) mean difference in baseline PEG score for a unit change in each baseline risk factor is in Table 1. In the model with all risk factors (Table 2), the associations of fear avoidance, pain catastrophizing, sleep disturbance, and depression with baseline PEG were weaker than those in the minimally-adjusted models but stronger than those in the fully-adjusted models. In the multiple risk factors model the association with PEG was weaker for obesity and there was no significant association with smoking.
DISCUSSION
Fear avoidance, pain catastrophizing, sleep disturbance, depression, and obesity were associated with PEG after adjustment for a comprehensive set of confounders. Based on this analysis, each of these is a cLBP causal risk factor. Convergence of these findings with the results from Mendelian randomization studies and mediation analyses, which have different designs and assumptions, strengthen the evidence for the observed causal relationships.8 Future analyses will evaluate these relationships with longitudinal data.