Innovative Analysis of Predictors for Overall Survival from Systemic Non-Hodgkin T Cell Lymphoma Using Quantile Regression Analysis
Introduction
Systemic non-Hodgkin T/NK cell lymphoma (NHL-T) represents a rare and heterogeneous group of lymphoid malignancies with aggressive clinical behavior and poor prognosis. Despite advances in understanding its biology, the prognosis for patients with NHL-T remains dismal, particularly in advanced stages or disseminated disease. Traditional prognostic tools, such as the International Prognostic Index (IPI), have been widely used but may not fully capture the variability in clinical outcomes influenced by non-invasive biomarkers. This study aimed to identify key non-invasive clinical predictors of overall survival (OS) in NHL-T patients and explore their differential impacts across the survival distribution using quantile regression analysis, a robust statistical method that provides insights into covariate effects at specific quantiles of the outcome variable.
Methods
Study Design and Patient Population
A retrospective cohort of 183 NHL-T patients diagnosed between January 2006 and December 2015 at Beijing Friendship Hospital, China, was analyzed. Inclusion criteria encompassed histologically confirmed NHL-T according to the WHO 2003 classification. Patients with incomplete clinical data or lost to follow-up were excluded. Demographic, clinical, and laboratory variables were collected, including age, sex, Ann Arbor stage, presence of B symptoms, serum biomarkers (albumin, β2-microglobulin [β2-MG], erythrocyte sedimentation rate [ESR], lactate dehydrogenase [LDH]), hematologic parameters (hemoglobin, platelet count, white blood cell [WBC] count), and IPI scores. Hemophagocytic lymphohistiocytosis (HLH) status was documented due to its association with aggressive disease.
Statistical Analysis
Survival outcomes were assessed using Kaplan-Meier curves and Cox proportional hazards regression. Variables significant in univariate analysis (P < 0.05) were included in multivariable Cox models adjusted for HLH. Quantile regression analysis was performed to evaluate predictor effects at OS distribution quantiles (0.25, 0.50, 0.75, 0.95). This method avoids assumptions of normality and homoscedasticity, enabling detection of varying predictor impacts across survival time percentiles.
Results
Patient Characteristics
The cohort had a median age of 45 years (range: 12–85), with 69.4% males. Advanced Ann Arbor stage (III/IV) was present in 79.8% of patients, and 20.8% had concurrent HLH. The median OS was 5.1 months (interquartile range [IQR]: 1.1–24.5), with a 47.5% mortality rate. Subtypes included NK/T cell lymphoma (37.2%), angioimmunoblastic T cell lymphoma (21.9%), peripheral T cell lymphoma-not otherwise specified (PTCL-NOS; 18.6%), and T cell lymphoblastic lymphoma (T-LBL; 10.9%).
Univariate and Multivariable Cox Regression
Univariate Analysis: Lower OS rates correlated with advanced Ann Arbor stage (43.8% vs. 86.5% for stage I/II; P < 0.001), HLH (10.5% vs. 63.5%; P < 0.001), elevated IPI scores (38.9% for IPI 4–5 vs. 84.6% for IPI 0–1; P < 0.001), and B symptoms (P < 0.001). Protective factors included higher serum albumin (≥34 g/L: 67.0% vs. 37.6%; P < 0.001), ESR ≥23 mm/h (63.6% vs. 44.3%; P = 0.010), and platelet count ≥53×10⁹/L (64.3% vs. 10.0%; P < 0.001).
Multivariable Analysis: After adjusting for HLH, advanced stage with symptoms (HR = 3.16; 95% CI: 1.39–7.20; P = 0.006), thrombocytopenia (HR = 2.57; 95% CI: 1.57–4.19; P < 0.001), and higher IPI scores (HR = 1.29 per unit increase; 95% CI: 1.01–1.66; P = 0.043) were independent risk factors. Protective factors included T-LBL subtype (HR = 0.40; 95% CI: 0.20–0.80; P = 0.010), elevated WBC count (HR = 0.57; 95% CI: 0.34–0.96; P = 0.033), higher albumin (HR = 0.60; 95% CI: 0.37–0.97; P = 0.039), and elevated ESR (HR = 0.53; 95% CI: 0.33–0.87; P = 0.011).
Quantile Regression Analysis
Quantile regression revealed heterogeneous predictor effects across OS quantiles:
- IPI Score: Consistently negative association with OS at 0.25 (β = −1.2, P = 0.024), 0.50 (β = −3.2, P = 0.049), and 0.95 (β = −24.9, P < 0.001) quantiles, indicating stronger detrimental effects at higher survival times.
- ESR: Significant protective effect at median (β = 0.1, P = 0.010) and 0.75 quantiles (β = 0.2, P = 0.020).
- Serum β2-MG: Adverse impact peaked at the 0.75 quantile (β = −8.5, P = 0.001).
- Platelet Count: Protective effect diminished with higher quantiles (0.25: β = 0.06, P = 0.060; 0.95: nonsignificant).
- Clinical Stage with Symptoms: Adverse effects were pronounced at lower quantiles (0.25: β = −3.4, P = 0.007).
Discussion
Key Findings
This study highlights the prognostic utility of non-invasive biomarkers in NHL-T. The IPI score demonstrated robust predictive value across multiple quantiles, reinforcing its role in risk stratification. Elevated ESR and serum albumin emerged as stable protective factors, particularly in intermediate survival quantiles. Quantile regression uncovered dynamic predictor effects, such as the escalating negative impact of IPI and β2-MG with longer survival times, which traditional Cox models might overlook.
Clinical Implications
The identification of ESR as a protective marker contrasts with prior studies linking elevated ESR to poor outcomes in other lymphomas. This discrepancy may reflect NHL-T-specific inflammatory dynamics or HLH-related confounding. The adverse effect of thrombocytopenia aligns with its role in diffuse large B cell lymphoma, suggesting pan-lymphoma relevance. T-LBL’s favorable prognosis underscores subtype-specific biology influencing treatment response.
Methodological Innovation
Quantile regression provided granular insights into predictor effects across survival distributions. For example, HSCT and absence of clinical symptoms showed maximal impact at intermediate quantiles, suggesting tailored utility for average-risk patients. This approach complements Cox models by identifying subgroups where interventions may yield differential benefits.
Limitations
The retrospective design introduces potential selection bias, particularly in a single-center cohort. Heterogeneous treatment regimens (e.g., CHOP vs. asparaginase-based therapy) and limited HSCT cases (n = 14) may confound survival analyses. Despite adjusting for HLH, residual confounding by unmeasured variables (e.g., genetic markers) remains possible.
Conclusion
This study establishes IPI as a robust prognostic marker across survival quantiles and identifies ESR as a stable protective factor in NHL-T. Quantile regression enhances traditional survival analyses by delineating predictor effects at specific outcome levels, offering a framework for personalized risk assessment. Future prospective studies should validate these findings and integrate novel biomarkers to refine prognostic models.
doi.org/10.1097/CM9.0000000000000088
Was this helpful?
0 / 0