Elevated neutrophil-to-lymphocyte ratio as a prognostic factor in primary central nervous system lymphoma

Abstract

Background. Primary central nervous system lymphoma (PCNSL) is a rare form of non-Hodgkin lymphoma that represents 4%-7% of all brain tumors. Several bodies of literature show that preoperative hematological values correlate with the prognosis of solid malignancies including those of the CNS. This study aimed to study the prognostic value of preoperative neutrophil-to-lymphocyte ratio (NLR) in a retrospective analysis of patients presenting with biopsy proven primary CNS lymphoma.

Methods. Preoperative complete blood counts (CBC) were collected for PCNSL patients who went through surgical resection (n = 55). Clinical information, such as demographic information, imaging characteristics, and clinical follow-up, was acquired. Cox proportional hazard models, ROC curves, and Kaplan-Meier curves were created to evaluate overall survival (OS) and recurrence-free survival (RFS).

Results. Univariate Cox regression analysis showed how the NLR is significantly correlated with OS and RFS (P = .002, P = .006, respectively). In multivariate Cox regression analysis, NLR is prognostically significant for OS and RFS (P = .006, P = .001, respectively). The ROC curve analyzing NLR for OS had an area of 0.627 and had an optimal cutoff value of 6.000. Using the NLR cutoff value of 6 in Kaplan-Meier analysis for RFS, the NLR was significant in predicting unfavorable OS in PCNSL patients (P = .0297).

Conclusions. A higher NLR was associated with poor OS and RFS in PCNSL patients. Further investigation must be conducted to validate the results and determine the optimal cutoff value for NLR in PCNSL patients.

Key Points.

• A higher NLR was associated with poor OS and RFS in primary CNS lymphoma patients.

• An optimal cutoff value for NLR must be further studied in PCNSL patients.

Importance of the Study

Preoperative hematological values, including the NLR, have been found to correlate with the prognosis of solid malignancies. We conducted a retrospective analysis of 55 patients with biopsy proven PCNSL and acquired clinical information, such as demographic information, imaging characteristics, and clinical follow-up, and preoperative CBCs. Using univariate and multivariate analysis, a higher NLR was found to be correlated with both poor OS and RFS. If other studies validate our findings, the results can be used to discover an optimal cut-off value and ensure a cheap, repeatable, and universally available test for PCNSL patients.

Primary central nervous system lymphoma (PCNSL) is a rare form of non-Hodgkin lymphoma (NHL) which may involve the brain, eyes, leptomeninges, and spinal cord. It represents roughly 1% of all NHLs and 4%-7% of all brain tumors.¹ Effective treatment of PCNSL is based upon induction therapy with a combination of high-dose methotrexate backbone, rituximab, and alkylator chemotherapy followed by consolidation with high-dose chemotherapy with autologous stem cell transplant, non-myeloablative chemotherapy, or maintenance chemotherapy. Radiotherapy can be used as consolidation or salvage ­therapy as PCNSL can be very radiosensitive.² While patients can achieve remission, there is also a risk of relapse, estimated at 30%-66%.³

Controversy remains with regard to the role of neurosurgery in the management of PCNSL. Some reports suggest that tumor debulking may improve outcome as it could lead to a decrease in intracranial pressure and improve neurological symptoms.⁴ However, for the majority of patients, the preferred surgical intervention is stereotactic biopsy in order to establish diagnosis.⁵

Several bodies of literature support the idea that preoperative hematological values may correlate with the prognosis of solid malignancies including those of the CNS.^6–8 This includes the neutrophil-to-lymphocyte ratio (NLR), which has been shown to indicative of a systematic inflammatory response correlating with poor prognosis in solid malignancies, such as within the colon, prostate, or bladder.

It has been postulated that a lower NLR correlates with reduced cell-mediated immunity, hence tumor-specific immunity.⁹ In non-small cell lung cancer (NSCLC), preoperative neutrophilia and leukocytosis were associated with worse overall survival (OS) and progression-free survival (PFS) on multivariate analysis, decreased local/regional control, and brain metastasis.⁶ In meningioma, neutrophilia was found to be an independent predictor of recurrence-free survival (RFS) on multivariate analysis and associated with higher rates of tumor recurrence.⁷ In glioblastoma (GBM), the NLR (>4) was correlated with OS on univariate and multivariate analysis when observing age, gender, ECOG ­performance status, location, extent of resection (EoR), and completion of adjuvant chemoradiation protocol.⁸

This study aimed to study the prognostic value of preoperative NLR in a retrospective analysis of patients presenting with biopsy-proven primary CNS lymphoma.

Methods

Patients and Clinical Data

The research protocol for this study was reviewed by the Northwestern Memorial HealthCare Institutional Review Board (IRB) and determined to be exempt from IRB review under regulation 45 CFR 46.101(b). A Waiver of Consent and Authorization was granted by the IRB for the analysis of Protected Health Information.

A retrospective analysis of 55 patients with biopsy-proven PCNSL was conducted. Ethics review was properly conducted for this study by an IRB. The clinical information, such as demographic information, imaging characteristics, and clinical follow-up, was acquired from electronic patient records. Imaging characteristics were defined as either unifocal or multifocal based upon MRI reports. Tumor recurrence was defined as a new enhancing lesion on surveillance MRIs that required further treatment based on clinical documentation.

Laboratory Data

Preoperative CBCs were collected and included hemoglobin (HGB) count, absolute neutrophil count (ANC), and absolute lymphocyte count (ALC). NLR was defined as ANC/ALC.

Statistical Analyses

The 2 primary outcomes of interest were OS and RFS, with OS being defined as the time from the date of surgery to the date of death or April 1, 2022, and with RFS being defined as the time from the date of surgery to the date of recurrence or August 1, 2022. Kaplan-Meier curves were created to analyze how various factors affected survival. Cox proportional hazard models were utilized for both univariate and multivariate analysis to determine the relationship between clinical and pathological factors with time after the surgery. The final multivariate model included NLR, HGB, age at diagnosis, whether the lesion was multifocal on imaging, and Karnofsky Performance Status (KPS) at diagnosis. Briefly, KPS is a validated scale from 0 to 100 assessing functional independence, where higher scores indicate better functional status. Treatment was incorporated via 2 additional covariates: a binary indicator for MTR induction (high-dose methotrexate, temozolomide, and rituximab versus all other induction regimens), and a three-level consolidation treatment variable comparing high-dose cytarabine and autologous stem cell transplantation each against a reference group of no intensive consolidation.

Significance was noted if P was less than .05 for both univariate and multivariate analyses. Hazard ratios (HRs), along with the P value and 95% CIs, were collected, using Python 3.11.4. Spearman coefficient testing was done to determine correlation between independent variables. Finally, ROC curves and their areas were calculated to determine the optimal cutoff values for predicting OS and PFS from NLR.

Results

We evaluated 55 patients with PCNSL who went through surgical resection. As shown by the demographic information and univariate analysis in Table 1, the mean age at diagnosis was 67.67 ± 9.42 years, ranging from 40 to 89 years, with a majority being men (56.4%). All patients presented with CNS Diffuse large B cell lymphoma (DLBCL). 28 patients had unifocal lesions while 27 patients had multifocal disease. Eight patients had tumor progression after chemotherapy. Leukocytosis was found in 6 patients (≈10%) and neutrophilia was found in 10 patients (≈19%). Among patients who received induction chemotherapy, the most common regimen was a combination of high-dose methotrexate, temozolomide, and rituximab (45.5%). Among patients who received consolidation therapy, autologous hemopoietic stem cell transplant (25.5%) and cytarabine (25.5%) were the most common regimens. Full demographic and treatment information is detailed in Table 1.

Table 1.

Patient characteristics

Parameter	Overall
Number of patients	55
Avg age at diagnosis	67.67
Sex
Women	24 (43.6%)
Men	31 (56.4%)
Imaging
Unifocal	28 (50.9%)
Multifocal	27 (49.1%)
Chemotherapy regimen
Induction
MTR	25 (45.5%)
RMVP	8 (14.5%)
MR	8 (14.5%)
Rituximab	4 (7.3%)
HDMTX	4 (7.3%)
MT	1 (1.8%)
Cyclophosphamide	1 (1.8%)
None	4 (7.3%)
Consolidation
AutoHSCT	14 (25.5%)
Cytarabine	14 (25.5%)
Maintenance TMZ + BTKI	8 (14.5%)
None	19 (34.5%)
Pathology
CNS DLBCL	55 (100%)
CBC
Avg ANC	5.15
Avg ALC	1.83
Avg HGB	12.84
Avg WBC	7.67
Avg NLR	4.76
Neutrophilia	10 (18.2%)
Leukocytosis	6 (10.1%)
Median KPS at diagnosis (range)	70 (40-100)

Abbreviations: ALC, absolute lymphocyte count; ANC, absolute neutrophil count; AutoHSCT, autologous hematopoietic stem cell transplantation; BTKi, Bruton’s tyrosine kinase inhibitors; CBC, complete blood count (measured in billion cells/Liter); DLBLC, diffuse large B-cell lymphoma; HDMTX, high-dose methotrexate; HGB, hemoglobin; KPS, Karnofsky Performance Status; MR, HDMTX + rituximab; MTR, HDMTX + rituximab + TMZ; MT, HDMTX + TMZ; NLR, neutrophil-to-lymphocyte ratio; RMVP, HDMTX + rituximab + vincristine + procarbazine; TMZ, temozolomide; WBC, white blood cell.

Univariate Cox regression analysis demonstrated that both NLR and KPS at diagnosis were significantly associated with OS (P = .002 and P = .021, respectively), whereas none of the other variables reached statistical significance (Figure 1A). Multivariate Cox regression analysis confirmed that NLR remained an independent prognostic factor for OS (P = .006), after adjusting for KPS at diagnosis, age at diagnosis, HGB, multifocal imaging, and treatment variables. KPS at diagnosis also demonstrated mild significance (P = .047), although no other variables showed independent associations with OS (Figure 1B). Full statistics are shown in Table 2.

Figure 1.

(A) Forest plot of univariate analysis for determinants of overall survival; (B) multivariate analysis for determinants of overall survival; (C) univariate analysis for determinants of recurrence-free survival; (D) multivariate analysis for determinants of recurrence-free survival.

Table 2.

Univariate and multivariate Cox regression analysis of factors associated with overall survival

OS	Univariate			Multivariate
	HR	P	95% CI	HR	P	95% CI
NLR	1.112	.002	1.040-1.190	1.121	.006	1.033-1.216
HGB	0.862	.204	0.686-1.084	0.909	.513	0.684-1.209
Age at diagnosis	0.981	.471	0.930-1.034	0.965	.210	0.913-1.020
Multifocal imaging	1.54	.372	0.596-3.974	1.762	.352	0.535-5.800
KPS at diagnosis	0.96	.021	0.927-0.994	0.960	.047	0.922-1.000
MTR	0.488	.173	0.174-1.369	1.118	.879	0.263-4.752
Cytarabine	0.582	.341	0.191-1.776	0.349	.157	0.081-1.502
AutoHSCT	0.538	.327	0.156-1.859	0.363	.298	0.054-2.447

Abbreviations: AutoHSCT, autologous hematopoietic stem cell transplant; CI, confidence interval; HGB, hemoglobin; HR, hazard ratio; KPS, Karnofsky Performance Status; MTR, high-dose methotrexate + temozolomide + rituximab; OS, overall survival.

Similarly, for RFS, univariate analysis identified NLR (P = .006), KPS at diagnosis (P = .018), and multifocal imaging (P = .043) as significant prognostic factors (Figure 1C). Multivariate Cox regression again demonstrated that NLR was an independent predictor of RFS (P = .001), with multifocal imaging and KPS also remaining statistically significant predictors of RFS (P = .027 and P = .044, respectively) (Figure 1D). Overall, these results suggest that elevated NLR and poor functional status at diagnosis are associated with worse outcomes in PCNSL patients, whereas other preoperative hematologic parameters did not show independent prognostic value. Full statistics are shown on Table 3. Pairwise Spearman correlations among covariates included in the multivariate models confirmed the absence of substantial multicollinearity, with all correlation coefficients below the conventional threshold for concern (all |r| < 0.6).

Table 3.

Univariate and multivariate Cox regression analysis of factors associated with recurrence-free survival

RFS	Univariate			Multivariate
	HR	P	95% CI	HR	P	95% CI
NLR	1.092	.006	1.026-1.162	1.127	.001	1.048-1.211
HGB	0.877	.177	0.724-1.061	0.879	.251	0.705-1.096
Age at diagnosis	0.979	.346	0.937-1.023	0.956	.066	0.912-1.003
Multifocal imaging	2.311	.043	1.027-5.201	2.915	.027	1.127-7.539
KPS at diagnosis	0.964	.018	0.935-0.994	0.964	.044	0.930-0.999
MTR	0.806	.609	0.353-1.841	2.321	.153	0.730-7.374
Cytarabine	0.658	.372	0.262-1.649	0.29	.051	0.084-1.005
AutoHSCT	0.709	.491	0.267-1.884	0.435	.240	0.109-1.741

Abbreviations: AutoHSCT, autologous hematopoietic stem cell transplant; CI, confidence interval; HGB, hemoglobin; HR, hazard ratio; KPS, Karnofsky Performance Status; MTR, high-dose methotrexate + temozolomide + rituximab; OS, overall survival.

Once NLR was proven to be prognostically significant for OS and RFS through univariate and multivariate analyses, receiver operating characteristic (ROC) curve analysis was conducted to determine the optimal cutoff value for stratifying patients in subsequent survival analyses.

The ROC curve quantifies the ability of the NLR continuous variable to predict the binary event outcome (eg, death vs no event for OS, death or recurrence vs no event for RFS) at any given time point. The optimal cutoff was identified by maximizing Youden’s Index, which represents the NLR value that best balances the sensitivity and specificity of the event prediction.

The ROC curve analyzing NLR for OS had an area of 0.627 but did not reach statistical significance (P = .132, 95% CI: 0.442-0.806). The optimal cutoff value was 6.000 (Figure 2A). The ROC curve analyzing NLR for RFS had an area of 0.645 and nearly reached statistically significance (P = .067, 95% CI: 0.490-0.802). The optimal cutoff value was 3.294 (Figure 2B).

Figure 2.

(A) Receiver operating characteristics curve analysis of NLR for overall survival; and (B) NLR for recurrence-free survival.

Kaplan-Meier analysis was then conducted to create survival curves for OS and RFS. Using the cutoff value of 6.000 in Kaplan-Meier analysis for OS, it was found that the NLR was significant in predicting unfavorable OS in PCNSL patients through log-rank test (P = .0297) (Figure 3A). Univariate Cox proportional hazards regression showed that the high NLR group had an increased risk of death, with a HR of 2.71, which was statistically significant (HR = 2.71, 95% CI: 1.06-6.885, P = .037). The estimated median OS time for the low NLR group was not reached over the course of the study, indicating that over 50% of low NLR patients survived. In contrast, the high NLR group had a median OS time of 34.17 months.

Figure 3.

(A) Kaplan-Meier analysis of the association of NLR dichotomized at 6.000 with overall survival; and (B) NLR dichotomized at 3.294 with recurrence-free survival.

Using the cutoff value of 3.294 in Kaplan-Meier analysis for RFS, NLR was not significantly able to predict unfavorable RFS in PCNSL patients through log-rank test (P = .164) (Figure 3B). However, the estimated median RFS time for the low NLR group was not reached over the course of the study, indicating that over 50% of low NLR patients exhibited RFS. In contrast, the median RFS time for the high NLR group was 34.17 months. Univariate Cox proportional ­hazards regression was performed to quantify the risk. The high NLR group had an increased risk of recurrence or death, with a HR of 1.74, although not statistically significant (HR = 1.74, 95% CI: 0.79-3.84, P = .169).

Discussion

Our study illustrated how a higher NLR is correlated with both poor OS and RFS. The clinical use of NLR could vastly aid both patients and doctors if validated by other cohorts, so studies like ours are necessary to ensure an inexpensive, repeatable, and universally available test for PCNSL patients. However, there is still much we do not know about NLR, such as a universal optimal cut-off value for PCNSL patients.

The results of the study align with past findings. A meta-analysis that combined 9 studies consisting of 2297 patients with DLBCL found that NLR is correlated with worse OS and PFS and mentioned no cutoff value.¹⁰ An updated meta-analysis that combined 11 studies consisting of 2515 patients with DLBCL achieved the same conclusion, noting an optimal cutoff value of 3.¹¹ Another study completed with 290 DLBCL patients yielded similar results, with the multivariate analysis for OS and disease free survival displaying a correlation between high derived NLR and negative clinical outcome; this study noted the optimal cutoff derived NLR value was 4, although derived-NLR is a distinct metric computed from WBC and neutrophil counts rather than ANC/ALC directly.¹² Studies in classic Hodkin lymphoma and peripheral T-cell lymphoma have also shown that higher NLR is correlated with lower OS/RFS; however, their cutoffs were derived in different lymphoma subtypes as our study, limiting direct clinical comparison.^13,14

Among DLBCL-specific studies using standard NLR, one study determined the optimal NLR cutoff was 6¹⁵ while the other determined it was 2.32.¹⁶ Our OS cutoff of 6.0 is consistent with Nakaya et al.¹⁵ supporting the prognostic relevance of this threshold in DLBCL-lineage disease. The variability in reported cutoffs across the broader literature is likely attributable to differences in patient populations and treatment eras.

Notably, PCNSL is also biologically distinct entity from systemic DLBCL, formally classified by the WHO as a large B-cell lymphoma of immune-privileged sites, with a unique CNS tumor microenvironment driven by the blood-brain barrier, microglial infiltration, and enriched MYD88/CD79B mutational burden not typical of nodal DLBCL.^{1,  17,18} Thus, the optimal NLR threshold in our study may differ from that established in systemic DLBCL cohorts, underscoring the need for validation in dedicated PCNSL series.

One of the limitations of our study include the number of patients. Our research was constrained to 55 patients, a relatively small-sample size. Future studies in larger and multicenter cohorts will need to be conducted to validate our results. Future analysis will also incorporate other prognostic variables, including molecule profile and imaging characteristics, as more patients are collected. Additionally, a significant interest in new sequencing technologies is growing as they can be utilized to assesses circulating tumor DNA, potentially obviating the need for surgical diagnosis, which can be further analyzed in future research.

In conclusion, our study illustrated how a higher NLR is associated with both poor OS and RFS, proving that it can be a useful marker in predicting prognosis in PCNSL patients.

Conflict of Interest Statement

None of the authors have any conflicts of interest.

Author Contributions

Original writing—draft: R.M., R.T., R.S. Original writing—review/editing: RM, RT, RS, MO, K.D., O.H.K. Data curation, statistical analysis: R.M., R.T., R.S., M.O., K.D., O.H.K. Conceptualization: M.O., K.D., O.H.K.

Funding

This work was supported by the Northwestern Medicine Pre-Med Internship.

Ethics Statement

The research protocol for this study was reviewed by the Northwestern Memorial HealthCare Institutional Review Board (IRB) and determined to be exempt from IRB review under regulation 45 CFR 46.101(b). A Waiver of Consent and Authorization was granted by the IRB for the analysis of Protected Health Information.

Data Availability

Anonymized data can be shared upon request.