Smoking status and nonlinear associations of cerebrospinal fluid GDF11 and IGF2 with Subjective sleep quality: a cross-sectional study using generalized additive models

Cigarette smoking is consistently associated with poorer sleep, yet the central biochemical correlates of this relationship remain insufficiently characterized in humans. Growth differentiation factor 11 (GDF11) and insulin-like growth factor 2 (IGF2) are candidate biomarkers of vascular and neurotrophic signaling, but their associations with sleep quality in smokers and non-smokers are unclear. In this cross-sectional study, 191 male participants (87 smokers and 104 non-smokers) undergoing anterior cruciate ligament reconstruction were enrolled. Preoperative cerebrospinal fluid (CSF) concentrations of GDF11 and IGF2 were measured by enzyme-linked immunosorbent assay, and sleep quality was assessed using the Pittsburgh Sleep Quality Index (PSQI). Linear regression and overall generalized additive models (GAMs) were fitted first; smoking-stratified GAM and two-piecewise linear regression analyses were then conducted as exploratory secondary analyses. No formal biomarker × smoking interaction test was performed. Smokers had poorer PSQI scores, lower CSF GDF11 levels, and higher CSF IGF2 levels than non-smokers. Based on source-data reanalysis, the unadjusted mean differences (non-smokers minus smokers) were − 1.48 points for PSQI (95% CI, -2.19 to -0.77), 192.62 pg/mL for GDF11 (95% CI, 166.56 to 218.68), and − 1.68 ng/mL for IGF2 (95% CI, -3.10 to -0.26). Adjusted group differences remained significant for PSQI and GDF11 but not for IGF2 (p = 0.059). In separate fully adjusted linear models, GDF11 was not statistically significant, whereas the association for IGF2 was borderline (B = 0.068; 95% CI, -0.001 to 0.137; p = 0.055). Overall GAMs identified nonlinear associations of both biomarkers with PSQI scores. Exploratory smoking-stratified analyses suggested different curve shapes and sample-derived inflection points across groups. CSF GDF11 and IGF2 showed nonlinear associations with subjective sleep quality, and exploratory smoking-stratified analyses suggested possible heterogeneity in curve shape. Because no formal interaction test was performed, these subgroup patterns should not be interpreted as definitive effect modification. The sample-derived inflection points also require validation in independent cohorts. Linear regression showed no significant biomarker-PSQI associations. GAMs identified nonlinear biomarker-sleep associations. Smoking-stratified GAM curves showed different estimated shapes. Exploratory analyses suggested a threshold-like GDF11 pattern in non-smokers. Exploratory analyses suggested a threshold-like IGF2 pattern in smokers.