Data-driven serum cytokine and chemokine phenotypes are associated with MRI-defined disc regression and persistent radicular pain in lumbar disc herniation.

Background: Inflammation after lumbar disc herniation (LDH) may contribute to radicular pain while also participating in resorption of herniated disc material. Whether circulating immune profiles are associated with these endpoint-specific recovery patterns remains uncertain. Methods: This single-center observational cohort included patients with symptomatic LDH treated between 1 January 2023 and 30 November 2025. Stored baseline serum IL-6, IL-8/CXCL8, CCL2/MCP-1, and MMP-9 were log-transformed, standardized, and entered into outcome-blinded Gaussian mixture modeling. MRI-defined regression (>=50% volume reduction during nonoperative follow-up) and persistent radicular pain (recorded follow-up leg-pain VAS >=3) were analyzed in endpoint-specific cohorts. Selection weighting, early-surgery-as-failure, classification-certainty, storage-adjusted, continuous-outcome, and penalized-model sensitivity analyses were performed. Results: Among 146 patients with complete four-marker data, the BIC-preferred EII three-class solution identified low immune activation (n=46), cytokine-dominant (n=32), and chemokine/MMP-dominant (n=68) phenotypes (Delta BIC 7.29 versus the next-best model). In the paired-MRI cohort (N = 117), regression occurred in 52.6%, 40.9%, and 71.9%, respectively. The chemokine/MMP-dominant phenotype was associated with MRI regression in the primary adjusted model (OR 2.99, 95% CI 1.11-8.08; P = 0.031; global P = 0.019), and adjusted continuous volume reduction was 9.88 percentage points greater than in the low-activation group (95% CI 0.08-19.68; P = 0.048). Several sensitivity analyses retained the direction but widened uncertainty, including nonlinear-age, early-surgery, and posterior-certainty analyses. In the persistent-pain cohort (N = 123), the categorical phenotype association was heterogeneous but class-specific estimates were imprecise; the continuous IL-6/IL-8 cytokine axis showed the clearest association with persistent pain (OR per 1-SD 2.38, 95% CI 1.34-4.21; P = 0.003). Conclusion: Baseline serum immune profiles were associated with endpoint-specific structural and pain outcomes in LDH. The chemokine/MMP-dominant phenotype showed the clearest association with MRI-defined regression, whereas persistent pain was more consistently related to a continuous cytokine axis. These observational findings are hypothesis-generating and require prospective multicenter and tissue-level validation before clinical use.