We read with great interest the study by Man et al on modifiable risk factors for low back pain (LBP).1 The authors combine GBD population attributable estimates, CHARLS restricted cubic spline analysis, and two-sample Mendelian randomization (MR) to examine the contributions of high BMI, smoking, and occupational factors. While this multi-evidence approach is valuable, we raise several methodological points regarding causal inference, subgroup patterns, and confounding.

First, the MR result supports a causal effect of genetically predicted higher BMI on LBP (OR 1.396), but MR is inherently linear2 and cannot capture the U-shaped pattern seen in CHARLS, particularly the elevated risk below 23.9 kg/m2. This leaves a gap: the left arm of the U-curve may reflect reverse causation (chronic pain reducing activity and body weight), confounding by sarcopenia, or a genuine biological effect. Without non-linear causal verification,3 readers may mistakenly interpret low BMI itself as a risk factor.

Second, the CHARLS analysis did not stratify the U-shaped BMI-LBP association by sex or age. The authors note that women and older adults bear a disproportionate burden, yet it is unclear whether the U-shape is driven by post-menopausal women, in whom estrogen loss accelerates sarcopenia, or holds consistently across middle-aged men, who may show a more linear pattern. If the curve varies by subgroup, the universal claim that low BMI elevates LBP risk becomes harder to sustain, and prevention messages would need tailoring.

Third, the GBD attributable-fraction estimates assume the three risk factors act independently. The comparative risk assessment calculates PAFs for each factor in isolation4 when factors are correlated, the sum of individual PAFs can exceed the joint attributable fraction, and the independent contribution of any single factor may be smaller than its isolated estimate suggests. High BMI and occupational ergonomic exposure may co-occur in the same populations, so ignoring these interdependencies could distort the prioritization of interventions, especially the recommendation to target weight management in low- and middle-SDI regions.

Finally, the CHARLS model omitted physical activity. Low activity is a common cause of both higher BMI and LBP through deconditioning. The authors adjusted for age, sex, education, smoking, alcohol, hypertension, and diabetes, but not activity level. This leaves open the possibility that part of the BMI-LBP association is confounded by sedentary behavior rather than adiposity itself, a point that could be tested using activity data already collected in CHARLS.

We welcome this multi-evidence contribution and offer four suggestions: (1) apply non-linear MR3 to existing GWAS summary statistics to test for a causal non-linear BMI-LBP relationship; (2) repeat RCS analysis within sex, age, and urban-rural strata to clarify effect modification; (3) use sequential PAF or mediation-based methods within the GBD framework to partition independent and overlapping contributions of BMI and occupational risk; (4) incorporate physical activity variables from CHARLS in a sensitivity analysis to test whether the RCS curve changes after adjustment. Where feasible, future observational work should adopt longitudinal designs to better separate reverse causation from genuine causal effects at low BMI levels.

We welcome the authors’ Response. Clarifying these points would add precision to an already valuable study.

The authors declare no conflicts of interest in this communication.

1. Man X, Yun X, Zhang L. Comprehensive analysis of modifiable global risk factors for low back pain. J Pain Res. 2026;19:613910. doi:10.2147/JPR.S613910

2. Burgess S, Davey Smith G, Davies NM, et al. Guidelines for performing Mendelian randomization investigations: update for summer 2023. Wellcome Open Res. 2023;4:186. doi:10.12688/wellcomeopenres.15555.3

3. Karlsson T, Hadizadeh F, Rask-Andersen M, Johansson Å, Ek WE. Body mass index and the risk of rheumatic disease: linear and nonlinear mendelian randomization analyses. Arthritis Rheumatol. 2023;75(11):2027–2. doi:10.1002/art.42613

4. Mansournia MA, Altman DG. Population attributable fraction. BMJ. 2018;360:k757. doi:10.1136/bmj.k757