Keywords
Brain Neoplasms - Survival Analysis - Biomarkers
Palavras-chave
Neoplasias Encefálicas - Análise de Sobrevida - Biomarcadores
INTRODUCTION
Brain metastasis (BM) are the most common malignant intracranial tumors, and their incidence continues to rise over time. Advances in oncologic treatments and supportive care provides patients with longer survival to the point of developing brain metastases in end-stage disease.[1] Treatment of BM ranges from supportive care to multimodal strategies considering surgery, fractioned radiotherapy, whole brain radiotherapy, stereotactic radiosurgery, or chemotherapy. Treatment decisions are based on primary tumor site, metastatic tumor characteristics, and patient performance status. Determining the most suitable treatment option based on the patient prognosis can be challenging.[2]
Cancer is a disease state associated with marked systemic inflammation.[3] Previous studies have demonstrated that elevated levels of inflammatory biomarkers can be associated with cancer progression and recurrence of solid tumors.[4] For example, the neutrophil-to-lymphocyte ratio (NLR), monocyte-to-lymphocyte ratio (MLR), platelet-to-lymphocyte ratio (PLR), and red blood cell distribution width (RDW) have shown promise as predictors of cancer survival in several different malignant diseases.[5]
[6]
[7]
[8]
[9]
The use of validated prognostic tools can greatly aid clinicians in creating patient-centered treatment plans. While the NLR, MLR, PLR, and RDW have demonstrated efficacy as prognostic factors in many malignancy states, few studies have evaluated their performance specifically for patients with BM.[6] This study aims to fill this gap by examining the relationship between these biomarkers and patient outcomes in a prospective cohort.
METHODS
This report adheres to the Standards for the Reporting of Diagnostic Accuracy Studies (STARD) statement. Although the STARD was developed for diagnostic accuracy studies, it also provides a useful framework for reporting other studies evaluating the performance of other clinical tests, including prognostic studies.[10]
Study design and patient selection
We retrospectively reviewed a database of 200 patients recruited prospectively and consecutively for a recent diagnosis of BM, and primarily treated with surgical resection between August 2010 and July 2016 at the Instituto do Câncer do Estado de São Paulo of the Faculdade de Medicina of the Universidade de São Paulo (ICESP-FMUSP).
Patients were included in this study if they met the following criteria: 1) adult patients (≥ 18 years old); 2) at least one BM treated by surgical resection; and 3) did not receive any treatment for BM before surgery. One patient had a limited surgical resection (< 50%) which was considered as an open biopsy. After surgical resection, patients were referred for adjuvant therapy after multidisciplinary evaluation. The patients were prospectively followed up and underwent magnetic resonance imaging (MRI) every 4 months for a maximum follow-up period of 60 months, or until death. Follow-up was lost on 16 patients, for whom the date of the last appointment was considered for analysis.
Data collection and hematologic variables
Patients' epidemiological and clinical information were collected from an electronic database, and their performance status was assessed using the American Society of Anesthesiologists (ASA) and the Karnofsky performance status (KPS) classifications. The graded prognostic assessment (GPA) oncologic prognostic score was used to adjust for risk. The primary outcome of interest was 1-year survival.
Blood samples were collected no more than 30 days before surgery in ethylenediaminetetraacetic acid (EDTA), and the complete blood count (CBC) was processed using an automated hematology analyzer. The NLR, MLR, and PLR were respectively calculated by dividing the absolute neutrophil, monocyte, and platelet counts, respectively, by the absolute lymphocyte count. The RDW was provided by the hematology analyzer.
Statistical analysis
Categorical variables were presented as relative and absolute frequencies. Normally distributed continuous data were presented as mean and standard deviation (SD), or median and quartiles, as appropriate and indicated. The normality assumption was assessed by skewness and kurtosis values, as well as graphical methods for each variable.
For the outcome of 1-year survival, potential categorical predictors were identified using the Kaplan-Meier method, using the logrank (Mantel-Cox) test for the comparison of the survival functions. Continuous variables were analyzed through a univariate Cox regression. Variables with a univariate p-value less than 0.10 were included in a multivariate Cox regression model, and the results were presented as hazard ratios (HR), with 95% confidence intervals (CI). The change-in-estimate criterion strategy was also employed as a sensitivity analysis while not changing the direction or effect size of the relationship between the hematological parameters and the outcome. The hematological parameter was then included in the regression model as a continuous variable. The proportionality and linearity assumptions were verified using graphical methods and the Schoenfeld and Martingale residuals, respectively. Since age, preoperative KPS, and the number of BM lesions are considered in the GPA score, these variables were not considered for inclusion in the multivariate models.
The cutoff values for the NLR, MLR, PLR, and RDW are not well established for solid cancers, although recent studies have suggested thresholds of NLR < 4[9] and PLR < 150.[10] We built separate multivariate regression models to assess the predictive performance of each of these biomarkers as continuous or categorical variables. Afterwards, we refined the cut-off determination directed at the survival outcome, as proposed by Budczies et al.[11] A Cox proportional hazard model was fitted to the dichotomized hematological and survival parameters. The survival analysis was performed using the functions coxph and survfit from the R package survival.[12] The optimal cutoff was defined as the point with the most significant (logrank test) split. Differences in survival were calculated from the mean survival times in the good and poor prognosis groups. Mean survival times were estimated from the area under the Kaplan-Meier curve using the maximum time that occurs in the data as the uniform time endpoint. To assess the robustness of the cutoff value and avoid overestimation of effect size, the HR, including 95% CI, was plotted against all possible cutoff values. A wider range of cutoff values demonstrating significance would indicate more robust findings.
Significance was tested through two-tailed tests at a significance level of p < 0.05 for all analyses. All analyses were conducted with the R (R Foundation for Statistical Computing, Vienna, Austria), version 3.5.2. The sample size was based on the available data, and no a priori power calculations were performed.
RESULTS
A total of 200 patients (mean age 56.1 ± 12.6 years, 55% female) met the eligibility criteria and were included in the analysis ([Table 1]). The most common presenting malignancies were non-small cell lung cancer (NSCLC) (33.0%), breast cancer (18.0%), and cancers of the gastrointestinal tract (13.5%). A single BM was diagnosed in 52% of patients, three or more lesions in 30.5% during initial diagnosis, and carcinomatous meningitis in 10%. The performance status assessment by different scales is illustrated in [Table 1].
Table 1
Sample characteristics and univariate survival analysis
|
Variables
|
Total (n = 200)
|
Death at 1-year
|
p-value *
|
|
No (n = 69)
|
Yes (n = 131)
|
|
Age (years)
|
56.1 ± 12.6
|
54.5 ± 12.8
|
56.9 ± 12.6
|
0.196
|
|
Female
|
110 (55.0)
|
42 (60.9)
|
68 (51.9)
|
0.138
|
|
Synchronous presentation
|
72 (36.0)
|
19 (27.5)
|
53 (40.5)
|
0.260
|
|
Compartment
|
Infratentorial
|
54 (27.0)
|
20 (29.0)
|
34 (26.0)
|
0.814
|
|
Supratentorial
|
137 (68.5)
|
46 (66.7)
|
91 (69.5)
|
|
Both
|
9 (4.5)
|
3 (4.3)
|
6 (4.6)
|
|
Laterality
|
Right
|
62 (31.0)
|
30 (43.5)
|
32 (24.4)
|
0.034
|
|
Left
|
60 (30.0)
|
19 (27.5)
|
41 (31.3)
|
|
Both
|
78 (39.0)
|
20 (29.0)
|
58 (44.3)
|
|
Eloquent brain areas
|
55 (27.5)
|
13 (18.8)
|
42 (32.1)
|
0.005
|
|
Number of BM
|
1 (1–3)
|
1 (1–3)
|
2 (1 -3)
|
0.32
|
|
ASA
|
2 (2–3)
|
2 (2–3)
|
2 (2–3)
|
< 0.001
|
|
KPS
|
Preoperative
|
60 (50–80)
|
70 (60–80)
|
60 (50–70)
|
< 0.001
|
|
Postoperative
|
80 (60–90)
|
90 (80–100)
|
70 (50–90)
|
< 0.001
|
|
GPA
|
1.5 (1–2)
|
2 (1–2.5)
|
1.5 (1–2)
|
0.001
|
|
ECOG
|
2 (1–3)
|
2 (1–3)
|
3 (1–4)
|
< 0.001
|
|
Number of resected BM
|
1 (1–1)
|
1 (1–1)
|
1 (1–1)
|
0.757
|
|
Extent of resection
|
Biopsy
|
1 (0.5)
|
0 (0.0)
|
1 (0.8)
|
0.069
|
|
Partial
|
21 (10.5)
|
4 (5.8)
|
17 (13.0)
|
|
Gross total
|
178 (89.0)
|
65 (94.2)
|
113 (86.3)
|
|
En bloc resection
|
41 (20.5)
|
17 (24.6)
|
24 (18.3)
|
0.471
|
|
Radiotherapy
|
126 (63.0)
|
57 (82.6)
|
69 (52.78)
|
< 0.001
|
|
Whole-brain
|
89 (44.5)
|
33 (47.8)
|
56 (42.7)
|
|
|
RS/FSR (surgical bed)
|
40 (20.0)
|
25 (36.2)
|
15 (11.5)
|
|
|
RS/FSR (concurrent BM)
|
14 (7.0)
|
8 (11.6)
|
6 (4.6)
|
|
Abbreviations: ASA, American Society of Anesthesiologists; BM, brain metastasis; GPA, graded prognostic assessment; KPS, Karnofsky performance status; ECOG, Eastern Cooperative Oncology Group; RS/FRS, Radiosurgery/Fractionated Stereotactic Radiotherapy.
Notes: Categorical variables are presented as n (%). Continuous variables are presented as median and quartiles, except for age (mean ± standard deviation). *Survival analysis.
All patients underwent surgical treatment; 89% underwent gross total resection, and 20.5% underwent en bloc resection. Adjuvant radiotherapy was administered to 63% of the patients. There were no intraoperative deaths; however, 3 patients died in the first postoperative week, and 7.5% died within 4 weeks. The median postoperative KPS score was 80 (quartiles 60–90), which was significantly improved from the preoperative score (p < 0.001). During follow-up, progression of primary disease was the most common cause of death. A total 131 patients died within the first year.
In the univariate survival analysis, GPA, ASA, preoperative hemoglobin, tumor location, laterality, radiotherapy, postoperative KPS, and recurrence were potential predictors. Thus, these variables were included in the adjusted models to assess the predictive performance for each blood-based parameter ([Table 2]). The NLR was significantly associated with 1-year outcome (HR 2.19, 95% CI 1.13–4.25, p = 0.021), while the MLR was not (HR 2.05, 95% CI: 0.98–4.26, p = 0.055). In a multivariate model including all blood-based biomarkers concurrently, we opted to not include the MLR due to a high correlation with the NLR (r = 0.626, p < 0.001). In this model, only the NLR maintained significance (HR 2.66, 95% CI: 1.17–6.01, p = 0.019) ([Figure 1]).
Figure 1 Neutrophil-to-lymphocyte ratio (NLR) histogram. Vertical line: optimal cutoff 3.832.
Table 2
Multivariate survival analysis
|
Blood-based parametera
|
Models with one blood-based parameter at a timeb
|
Model with all parametersc
|
|
HR (95% CI)
|
p-value
|
HR (95% CI)
|
p-value
|
|
RDW-CV
|
1.66 (0.08–33.55)
|
0.741
|
0.15 (0.0–4.97)
|
0.285
|
|
NLR
|
2.19 (1.13–4.25)
|
0.021
|
2.66 (1.17–6.01)
|
0.019
|
|
MLR
|
2.05 (0.98–4.26)
|
0.055
|
−
|
−
|
|
PLR
|
1.68 (0.76–3.7)
|
0.198
|
0.7 (0.26–1.92)
|
0.492
|
Abbreviations: CI, confidence interval; HR, hazard ratios; MLR, monocyte-to-lymphocyte; NLR, neutrophil-to-lymphocyte; PLR, platelet-to-lymphocyte ratio; RDW-CV, red-cell distribution width coefficient of variation.
Notes:
aAll parameters were log-transformed for distribution normalization, except for hemoglobin. All of them were inserted on the models as continuous variables; bAll models adjusted for GPA (age, preoperative KPS, number of CNS metastases, and presence of extracranial metastases), ASA, preoperative hemoglobin, eloquence of the affected area, laterality, radiotherapy, postoperative KPS, and recurrence; cModel adjusted for GPA (age, preoperative KPS, number of CNS metastases, and presence of extracranial metastases), ASA, preoperative hemoglobin, eloquence of the affected area, laterality, radiotherapy, postoperative KPS, and recurrence, besides the parameters on the table (hemoglobin, RDW-CV, NLR, and PLR). Finally, the MLR was not included due to high correlation with NLR.
[Figure 2] presents the analyses aimed at determining the optimal NLR cutoff value. The HR and differences in mean survival times (months) are each plotted against each possible preoperative NLR cutoff. A total 132 of 171 possible cutoffs (77.2%) were significant—the larger the range of significant cutoff values, the more robust the finding. In our series, the optimal NLR cutoff value, based on HR and differences in mean survival, was 3.832. We compared the survival curves and effect sizes according to preoperative NLR cutoff values for solid tumors, as suggested by Templeton et al.,[9]
[15] and our suggested optimal cutoff value. Both presented significant results in survival analysis; however, our value showed a higher HR = 2.07 (1.31–3.25; p = 0.0014) ([Figure 3]).
Figure 2 (A) Hazard ratios (HR) and 95% confidence intervals (CI) according to preoperative NLR cutoff; (B) Difference in mean survival (months) and 95% CI according to preoperative NLR cutoff. A total 132 of 171 possible cutoffs (77.2%) were significant. Vertical line: optimal cutoff 3.832.
Figure 3 Survival curves and effect sizes (HR and CI) according to preoperative NLR cutoff (A: cutoff 3.832; B: cutoff 4.0).
DISCUSSION
The results of this study demonstrated a significant association between preoperative NLR levels and patient survival at 1-year after surgery for BM. Of all the hematologic inflammatory markers we analyzed, only the preoperative NLR was found to be an independent predictor of survival in our adjusted models. More specifically, shorter survival time after surgical resection of BM can be expected for patients with a preoperative NLR > 3.83.
These findings align with previous works testing the clinical utility of similar cutoff values for inflammatory biomarkers. The NLR has been used as a predictor of late recurrence, treatment response and poor prognosis in different solid cancers.[5]
[8] Templenton et al.[9]
[13] conducted a systematic review of 100 studies (n = 40,559 patients) that included different solid tumors and suggested a NLR cutoff of 4.0 for overall survival (OS). While the biological processes underlying these findings are not fully understood, there is some evidence of a tumor-promoting effect of neutrophils during metastasis progression by an increasing number of metastatic initiating cancer cells. Notably, Wculek and Malanchi[14] showed that neutrophil depletion in lungs reduced local metastasis on a breast cancer experimental model.
The other hematologic biomarkers tested in our study did not perform as well as the NLR. The prognostic role of the PLR is hypothesized on increased host inflammatory response with higher secretion of thrombopoietic cytokines, such as IL-6. Platelet recruitment and activation is involved in the process of tumor growth and angiogenesis. There seems to exist a correlation between thrombocytosis and shorter survival time in different solid tumors, in addition to recent findings of higher PLR scores for metastatic patients compared with locoregional disease.[8]
[10]
[15]
[16]
Several biological events related to the cancer activity might implicate in high RDW levels. Higher levels of inflammatory cytokines are associated with increased metabolic activity, increased cellular proliferation, and, consequently, higher RDW rates. Furthermore, the RDW is directly related to the individual nutritional status.[7]
[17] Generalized hypovitaminosis (iron, folate, vitamin B12) in patients with advanced and uncontrolled cancers is an alternative hypothesis that could explain the role of the RDW as a prognosis predictor.[5]
[17] Consistent results of PLR and RDW as outcome predictors are related to advanced-stage cancers or uncontrolled systemic disease. Although brain metastasis is a hallmark of cancer disease progression, the majority of patients in this study had a good performance status preoperatively, possibly suggesting there was adequate control of the disease before brain metastasis detection. Therefore, we hypothesize that neither the PLR nor RDW were sensitive enough for significant results in this clinical setting.
The incidence of BM from solid cancers has been paradoxically rising due to improvements in diagnostic methods, increased availability of neuroimaging, and advances in systemic treatment.[1]
[7] The BM represents the end stage of a cancer disease; prognosis should be carefully considered for decision-making on additional treatments at this point. The decision on whether to treat a BM is challenging in neuro-oncology, as it represents not only a tumor with mass effect, but also a systemic uncontrolled disease in progression that must be considered for individualized decision. Currently, the decision to treat is based on patients' performance status, the number and size of tumors, and treatment responsiveness of the primary cancer.[1]
[18] New biomarkers are under investigation to support decision-making in this setting. Several blood-based biomarkers have been proposed for gliomas with promising results.[10]
[13]
[19]
[20]
[21]
[22] Our results promote the use of preoperative NLR in decision-making, and its contribution for prognostic predictive models for brain metastasis.
Other studies have analyzed the hematologic prognostic markers based on recent insights on cancer-immune system interactions. The NLR, PLR, and RDW were tested in several types of metastatic tumors, as well as in different brain tumors. Starzer et al.[23] addressed the role of systemic inflammation on cancer progression. In their retrospective analysis with more than one thousand patients, lower NLR, PLR, and MLR were associated with longer OS in patients with BM. Mitsuya et al.[6] used the cutoff of NLR > 5 with promising results in a retrospective analysis with 105 BM patients. Cacho-Diaz et al.[24] applied the NLR cutoff > 4.5, with significant correlation with mortality in those patients as well. Both studies have included only patients with uncontrolled primary cancer and high systemic inflammation. Some authors hypothesized that the involvement of the brain, as a critical and sensitive organ, is the main trigger of the systemic inflammation process that might influence the OS. Marini et al.[25] studied predictive factors on OS in patients with glioblastoma—a primary brain cancer with very low risk of distant metastasis. In this study, the NLR > 4 cutoff was significantly related to worse OS in multivariate analysis, similar to other studies with BM.
Our study has several limitations that should be considered when interpreting the results. First, the study included patients with BM from different primary tumors; this was not included as a covariate in our models, given the limited sample size of the cohort. Second, our analyses did not consider the varying doses of corticosteroids that patients received, although this may have had effects on hematologic parameters. Third, we only included patients without any prior treatment for BM (radiotherapy/radiosurgery, immunotherapy etc.). While this was done to reduce treatment confounding effects, we recognize it limits the external validity of the study and favors the proof of concept. However, our study has a representative sample of patients with cancer and BM, and proposes an accessible test for treatment decision making in patients with BMs.
In conclusion, our analyses demonstrated that the preoperative NLR is associated with 1-year survival outcomes after surgery for BM, and may be an important predictor of survival, irrespective of primary cancer site. These findings further support the trend of considering serum inflammatory markers in oncologic treatment decisions. Given the complexity of this disease and its burden on patients, clinicians are tasked with carefully formulating robust treatment plans. Although additional validation studies are warranted, we believe the low cost and accessibility of accessing the NLR distribution rates favors its utility as a prognostic aide when making decisions for the management of BM. Future studies should continue to explore and validate the use of other biomarkers as prognostic factors in metastatic diseases.
