This prospective cohort study was conducted at The Second Hospital of Hebei Medical University between January 2019 and December 2023. A total of 260 patients with pathologically confirmed lymphoma who had completed standard first-line chemotherapy were enrolled and followed longitudinally. Based on clinical outcomes during follow-up, patients were classified into a relapse group (n = 78) and a non-relapse group (n = 182).

The study protocol was reviewed and approved by the Ethics Committee of The Second Hospital of Hebei Medical University. Written informed consent was obtained from all participants prior to enrollment. All study procedures adhered to the principles of the Declaration of Helsinki.

The sample size was determined based on the anticipated relapse rate and the requirements of multivariable logistic regression. Assuming an expected 2-year relapse rate of approximately 30%, at least 80 relapse events would be needed to satisfy the rule of ≥10 events per variable for stable model estimation. Accordingly, the target sample size was set at 250–300 patients, which was deemed sufficient to ensure adequate statistical power for both logistic and survival analyses.

Patients were eligible for enrollment if they met all of the following criteria: (1) Age ≥18 years; (2) Pathologically confirmed non-Hodgkin lymphoma (NHL), including diffuse large B-cell lymphoma (DLBCL), T-cell lymphoma, or other NHL subtypes; (3) Completion of a standard first-line chemotherapy regimen (e.g., R-CHOP) with documented complete response (CR) or partial response (PR); (4) Ability and willingness to complete at least 24 months of clinical follow-up; (5) Availability of complete serum biomarker data at all predefined time points (T1–T3).

Patients were excluded if they met any of the following conditions: (1) Evidence of active infection, autoimmune disease, or chronic inflammatory conditions during chemotherapy or within 3 months before biomarker assessment; (2) Presence of other concurrent malignancies; (3) Severe hepatic or renal dysfunction, or a recent acute cardiovascular or cerebrovascular event; (4) Inability to comply with follow-up procedures or loss to follow-up.

Serum IL-6 and TNF-α concentrations were quantified using commercially available enzyme-linked immunosorbent assay (ELISA) kits. All ELISA plates were read using a SpectraMax iD5 microplate reader. Assay sensitivity was 0.70 pg./mL for IL-6 and 1.2 pg./mL for TNF-α, with manufacturer-reported detection ranges of 1.0–300 pg./mL and 2.0–500 pg./mL, respectively. For additional confirmation, parallel measurements were performed using chemiluminescence immunoassay kits based on the New Industry Chemiluminescence Analyzer MAGLWNI X8 platform. All calibrators and controls were run according to the standard Roche protocol. Serum β₂-MG levels were determined using a latex-enhanced immunoturbidimetric assay on a Roche Cobas 8000 automated biochemical analyzer. The analytical sensitivity was 0.1 mg/L, with a linear detection range of 0.1–10.0 mg/L. All assays were performed strictly following manufacturers’ instructions. Calibration was performed before each batch, and internal quality controls were included at low, medium, and high concentration levels.

All biomarker analyses were performed in the same certified clinical laboratory to minimize inter-laboratory variability. Both intra-assay and inter-assay coefficients of variation (CVs) were maintained below 10%. Each sample was measured in duplicate, and the mean value was used for analysis. Laboratory personnel conducting the assays were blinded to all clinical information, including relapse status and time points, to prevent measurement bias.

Continuous variables were assessed for normality and summarized as mean ± standard deviation for normally distributed data or as median (interquartile range) when distributional assumptions were not met. Categorical variables were presented as counts and percentages. Missing data were minimal; when the proportion of missing values was less than 5%, mean or median imputation was applied as appropriate. Variables with more than 5% missingness were documented and handled as described in the Appendix.

Baseline differences between the relapse and non-relapse groups were evaluated using the independent-sample t test or Mann–Whitney U test for continuous variables, depending on data distribution, and the χ² test or Fisher’s exact test for categorical variables.

Univariate logistic regression analyses were performed to evaluate the associations between each biomarker (IL-6, TNF-α, and β₂-microglobulin) measured at T1, T2, and T3 and the risk of relapse. Dynamic changes in biomarker levels (ΔT2–T1 and ΔT3–T2) were also examined. Multivariable logistic regression models were subsequently constructed focusing on biomarker levels at T2, given their strongest temporal association with relapse. IL-6, TNF-α, and β₂-microglobulin at T2 were entered simultaneously along with key clinical covariates. Results were reported as odds ratios (ORs) with 95% confidence intervals (CIs). Multicollinearity was assessed using variance inflation factors (VIFs); because dynamic change variables showed substantial collinearity with absolute T2 values, they were included only in univariate analyses.

Receiver operating characteristic (ROC) curves were generated to evaluate the discriminative performance of individual biomarkers at T2. A combined model incorporating IL-6, TNF-α, and β₂-microglobulin was also developed. The area under the ROC curve (AUC), sensitivity, specificity, and optimal cutoff values determined by the Youden index were calculated. Comparisons between AUCs were performed using DeLong’s test.

Relapse-free survival (RFS) was analyzed using the Kaplan–Meier method after stratifying patients into high- and low-biomarker groups based on median T2 values. Differences in survival distributions were assessed using the log-rank test. Multivariable Cox proportional hazards regression models were then fitted to examine the associations between T2 biomarker levels and RFS, adjusting for clinically relevant covariates including Ann Arbor stage, International Prognostic Index (IPI) score, and autologous stem cell transplantation status. Hazard ratios (HRs) with 95% CIs were reported, and the proportional hazards assumption was evaluated using Schoenfeld residuals.

To assess potential overfitting and model stability, internal validation was performed using repeated five-fold cross-validation. Model discrimination was summarized using the mean cross-validated AUC. Given the single-center nature of the cohort, orthogonal validation using an independent external dataset was not feasible; therefore, model performance is reported as internally validated discrimination rather than externally validated prediction.

All statistical analyses were conducted using R software (version 4.x; R Foundation for Statistical Computing) and SPSS (version 26.0; IBM Corp., Armonk, NY). ROC, Kaplan–Meier, and Cox regression analyses were implemented using the pROC, survival, and rms packages in R. All tests were two-sided, and a p value < 0.05 was considered statistically significant.

A total of 260 patients were included in the final analysis, of whom 78 (30.0%) experienced relapse during follow-up. The baseline characteristics of the study population are summarized in Table 1. The cohort comprised patients with clearly defined histopathological subtypes, including diffuse large B-cell lymphoma (DLBCL), T-cell lymphomas, follicular lymphoma, marginal zone lymphoma, and mantle cell lymphoma.

Compared with the non-relapse group, patients who experienced relapse had a significantly higher proportion of advanced Ann Arbor stage III–IV disease (76.92% vs. 49.45%, p = 0.001), high International Prognostic Index (IPI) scores (57.69% vs. 32.97%, p = 0.001), and B symptoms (51.28% vs. 27.47%, p = 0.001). Biochemical indicators also differed significantly between groups. The relapse group exhibited higher baseline lactate dehydrogenase (LDH) levels (310.50 ± 95.30 vs. 250.40 ± 80.20 U/L, p = 0.002) and β₂-microglobulin levels (3.90 ± 1.20 vs. 3.10 ± 1.00 mg/L, p = 0.003), whereas lymphocyte counts were significantly lower (1.25 ± 0.45 vs. 1.50 ± 0.50 × 10⁹/L, p = 0.010).

The relapse group was slightly younger than the non-relapse group (53.20 ± 12.10 vs. 56.80 ± 11.50 years), with a marginal but statistically significant difference (p = 0.045). No significant differences were observed in sex distribution, body mass index, chemotherapy regimen, number of chemotherapy cycles, or autologous stem cell transplantation.

Serum levels of IL-6, TNF-α, and β₂-MG at each assessment time point are presented in Table 2 and Supplementary Figure S1. At the end of chemotherapy (T1), all three biomarkers were significantly elevated in the relapse group compared with the non-relapse group, including IL-6 (12.73 ± 4.18 vs. 9.47 ± 3.31 pg./mL, p < 0.001), TNF-α (19.63 ± 6.33 vs. 15.27 ± 5.14 pg./mL, p < 0.001), and β₂-MG (3.87 ± 1.17 vs. 3.12 ± 0.97 mg/L, p = 0.002). These differences became more pronounced at the 3-month follow-up (T2), with relapse patients showing further increases in IL-6 (15.84 ± 4.96 vs. 10.36 ± 3.44 pg./mL), TNF-α (23.92 ± 7.18 vs. 16.51 ± 5.48 pg./mL), and β₂-MG (4.54 ± 1.28 vs. 3.27 ± 1.05 mg/L), all with p < 0.001. By the 6-month time point (T3), the divergence between groups was most marked. IL-6 continued to rise to 18.59 ± 5.47 pg./mL in the relapse group compared with 11.12 ± 3.87 pg./mL in the non-relapse group (p < 0.001). Similarly, TNF-α (27.84 ± 7.82 vs. 17.46 ± 6.12 pg./mL, p < 0.001) and β₂-MG (5.16 ± 1.41 vs. 3.34 ± 1.08 mg/L, p < 0.001) remained significantly higher in patients who later experienced relapse.

As shown in Table 3, IL-6, TNF-α, and β₂-MG measured at the 3-month follow-up (T2) were significantly associated with relapse in univariate analyses. Each 1-pg/mL increase in IL-6 at T2 was associated with a 12% higher odds of relapse (OR = 1.12, 95% CI: 1.08–1.17, p < 0.001), while TNF-α showed a smaller but significant effect (OR = 1.06, 95% CI: 1.03–1.09, p < 0.001). β₂-MG at T2 demonstrated the strongest association, with an OR of 1.55 (95% CI: 1.28–1.88, p < 0.001). Interval changes from T1 to T2 were also predictive; for example, ΔIL-6 (T2–T1) had an OR of 1.21 (95% CI: 1.11–1.32, p < 0.001).

In the multivariate model (Table 3), IL-6 and β₂-MG at T2 remained independent predictors after adjustment for clinical covariates. IL-6 at T2 retained a significant effect with an adjusted OR of 1.08 (95% CI: 1.03–1.14, p = 0.001), and β₂-MG showed an adjusted OR of 1.38 (95% CI: 1.11–1.72, p = 0.003). In contrast, TNF-α at T2 was no longer significant (adjusted OR = 1.03, p = 0.118). Advanced Ann Arbor stage (III–IV) and high IPI score also remained independent risk factors with adjusted ORs of 1.71 (95% CI: 1.00–2.93, p = 0.049) and 2.14 (95% CI: 1.25–3.67, p = 0.023), respectively.

Of the 260 patients included in the study, 150 (57.7%) were diagnosed with diffuse large B-cell lymphoma (DLBCL), representing the most prevalent histological subtype in the cohort. Within the DLBCL subgroup, serum levels of IL-6, TNF-α, andβ₂-MG measured at the 3-month post-treatment time point (T2) were significantly higher in patients who subsequently experienced relapse compared with those who remained relapse-free (Supplementary Table S1). After multivariable adjustment using the same covariate set as in the primary analysis—including age, Ann Arbor stage, International Prognostic Index (IPI) score, and autologous stem cell transplantation—IL-6 at T2 (adjusted OR = 1.07, 95% CI: 1.02–1.13) and β₂-MG at T2 (adjusted OR = 1.33, 95% CI: 1.05–1.69) remained independently associated with relapse risk, whereas TNF-α at T2 did not retain statistical significance (Supplementary Table S2).

As shown in Figure 1 and Table 4, all three biomarkers measured at the 3-month follow-up (T2)—IL-6, TNF-α, and β₂-MG—demonstrated significant diagnostic ability for predicting relapse. β₂-MG yielded the highest performance among the individual markers (AUC = 0.85, 95% CI: 0.79–0.90, p < 0.001), followed by IL-6 (AUC = 0.82, 95% CI: 0.76–0.88) and TNF-α (AUC = 0.79, 95% CI: 0.72–0.85). Sensitivity and specificity for single biomarkers ranged from 70.5 to 77.0% and 75.8 to 80.2%, respectively. The combined biomarker model demonstrated markedly superior discriminative power, achieving an AUC of 0.91 (95% CI: 0.87–0.95, p < 0.001), with improved sensitivity (84.6%) and specificity (86.3%).

Similar discriminative patterns were observed when the analyses were restricted to patients with DLBCL, with the combined biomarker model consistently demonstrating superior predictive performance compared with individual biomarkers (Supplementary Table S3).

In repeated five-fold cross-validation, the combined biomarker model demonstrated stable discriminative performance, with a mean cross-validated AUC of 0.88, which was comparable to the apparent AUC estimated in the full cohort (Supplementary Table S4).

In addition to AUC, classification performance was evaluated using confusion matrices and recall-based metrics. At the optimal cutoff, the combined biomarker model demonstrated balanced sensitivity and specificity for relapse detection, indicating that discriminative performance was not driven solely by correct classification of the majority non-relapse group (Supplementary Table S5).

Kaplan–Meier analyses demonstrated clear separation of relapse-free survival between high and low biomarker groups (Supplementary Figures S2A–C). Patients with high IL-6 levels at the 3-month assessment (T2) exhibited significantly shorter relapse-free survival compared with those with low IL-6 (log-rank p < 0.01). A similar pattern was observed for TNF-α and β₂-MG, with high-level groups showing a more rapid decline in survival probability over time. Among the three biomarkers, β₂-MG showed the widest curve separation, indicating the strongest prognostic impact.

In the multivariable Cox proportional hazards model (Table 5), high IL-6 (adjusted HR = 1.78, 95% CI: 1.20–2.64), high TNF-α (HR = 1.52, 95% CI: 1.03–2.26), and high β₂-MG (HR = 2.06, 95% CI: 1.36–3.11) remained independent predictors of earlier relapse after adjustment for Ann Arbor stage, IPI score, autologous transplantation, and age.

In subgroup analyses limited to DLBCL, higher levels of IL-6 and β₂-MG at T2 were also associated with significantly shorter relapse-free survival, as demonstrated by Kaplan–Meier analyses (Supplementary Figure S3).