Factors Associated With the Development of  Brain Metastases  Analysis of 975 Patients With Early Stage Nonsmall Cell Lung Cancer
Jessica L. Hubbs, MS1; Jessamy A. Boyd, MD2; Donna Hollis, MS3; Junzo P. Chino, MD2; Mert Saynak, MD4; and Chris R. Kelsey, MD2

METHODS: The records of all patients who underwent surgery for T1-T2 N0-N1 NSCLC at Duke University between the years 1995 and 2005 were reviewed.
The cumulative incidence of brain metastases and distant metastases was estimated by using the Kaplan-Meier method. A multivariate analysis assessed factors associated with the development of brain metastases.

RESULTS: Of 975 consecutive patients, 85% were stage I, and 15% were stage II. Adjuvant chemotherapy was given to 7%. The 5-year actuarial risk of developing brain metastases and distant metastases was 10%(95% confidence interval [CI], 8-13)
and 34%(95% CI, 30-39), respectively. Of patients developing brain metastases, the brain was the sole site of failure in 43%. On multivariate analysis, younger age (hazard ratio [HR], 1.03 per year), larger tumor size (HR, 1.26 per cm), lymphovascular space invasion (HR, 1.87), and hilar lymph node involvement (HR, 1.18) were associated with an increased risk of developing brain metastases.

CONCLUSIONS: In this large series of patients treated surgically for
early stage NSCLC, the 5-year actuarial risk of developing brain metastases was 10%. A better understanding of predictive
factors and biological susceptibility is needed to identify the subset of patients with early stage NSCLC who
are at particularly high risk. Cancer 2010;000:000–000. VC 2010 American Cancer Society.
KEYWORDS: nonsmall cell lung cancer, brain metastases, patterns of failure, prophylactic cranial irradiation

Lung cancer is the leading cause of cancer death in the United States,1 and nonsmall cell lung cancer (NSCLC) comprises approximately 85% of lung cancer cases. Approximately 25% of patients present with stage I-II disease,2 typically
managed with upfront surgery, with or without adjuvant chemotherapy.3,4 However, despite having early stage disease, approximately 50% of these patients will relapse.5,6 Brain metastases are 1 of the most common sites of disease recurrence after definitive treatment of lung cancer. The risk of developing brain metastases after treatment for locally advanced (stage III) NSCLC is approximately 30%-50%.7–12
The risk of developing brain metastases in patients with early stage disease is less defined. Furthermore, clinical and pathological factors associated with the development of brain metastases have not been fully elucidated in this population. Withmore effective chemotherapy, intracranial relapse may become increasingly important,13,14 and the ability to identify patients at highest riskmay facilitate interventions such as prophylactic cranial irradiation. We herein report the actuarial incidence of brain metastases in patients with early stage (stage I-II) NSCLC within a
cohort of 975 patients who underwent surgery at Duke University during an 11-year period. Further, we assess clinical and pathological factors independently associated with a higher risk of developing brain metastases.

MATERIALS AND METHODS
After obtaining institutional review board approval, the records of all patients who underwent surgery for T1-2 N0-1 NSCLC at Duke University Medical Center between the years 1995 and 2005 were reviewed. Patients who had received preoperative therapy (chemotherapy and/or radiation therapy), who presented with synchronous primary tumors, or who had a prior history of lung cancer were excluded from this study. Medical records and pertinent radiological imaging studies were reviewed to characterize each patient’s demographic information, to obtain surgical and pathological details, and to score patterns of failure after surgery. Stage was recorded based on the American Joint Committee on Cancer, sixth edition, staging system.15
All patients underwent curative-intent surgery. Adjuvant therapy (chemotherapy and/or radiation therapy) was given at the discretion of the treating physicians. Follow-up imaging was not standardized. In general, imaging of the brain was performed upon development of suspicious symptoms or as part of restaging at the time of disease
recurrence. All failures were reviewed by 2 investigators (C.K. and J.B.).
The Kaplan-Meier product-limit method16,17 was used to estimate the 5-year probability of developing distant metastases in general and brain metastases in particular, with 95% confidence intervals. Time to recurrence was measured from date of surgery to date of recurrence. A univariate Cox proportional hazards model16,17 was used to assess the strength of association between time to development of brain metastases and clinical and pathological risk factors (covariates) and to estimate hazard ratios and confidence intervals using the Wald chi-square statistic. The
assumption of proportional hazards was assessed by adding each covariate by log-time interaction to the model and assessing the significance of the product term by using the partial likelihood ratio test. In the multivariate Cox proportional
hazards regression model, all variables with P values <.10 in the univariate analysis were included. A stepwise, variable, selection approach was used, with entrance and
exit significance levels of P ¼ .05. Kaplan-Meier survival plots were used to assess subgroup survival similarities. All statistical tests were 2-tailed, and P < .05 was consideredsignificant. SAS 9.1 software (SAS, Cary, NC) was used for all statistical analyses.
RESULTS
The characteristics and surgical and pathological details from all 975 patients, as well as the subset that developed brain metastases, are in Table 1. Median follow-up for all
patients was 33 months (range, 1-149 months). A local and/or distant recurrence was identified in 250 patients. First sites of failure were local only (25%), local and distant (29%), and distant only (46%). Of the 207 patients who failed at distant sites, 60 (29%) developed brain metastases. The median time from surgery to development of brain metastases was 10 months (range, 1-55 months) in patients who developed brain metastases. Of the 60 patients who developed brain metastases, 26 (43%) were isolated to the brain. Most brain metastases were identified at the time of initial relapse (n ¼ 46;77%). The remainder were identified after a previous extracranial failure (n ¼ 14; 23%). The 5-year cumulative risk of developing brain metastases was 10%(95% CI,8%-13%; Fig. 1), whereas the 5-year actuarial rate of any distant recurrence was 34%(95% CI, 30%-39%). Several clinical and pathological factors were found
to be associated with the development of brain metastases on both univariate and multivariate analyses (Table 2). On multivariate analysis, younger age (HR,1.03 per year;P ¼ .01), increasing size (HR, 1.26 per cm; P < .01), lymphovascular space invasion (HR, 1.87; P ¼ .03), and hilar lymph-node involvement (HR, 1.18; P ¼ .04), were independently associated with an increased risk of developing brain metastases (Table 2). Staging computed tomography (CT) or magnetic resonance imaging (MRI) of the brain is not standard for all patients with early stage NSCLC, especially those with stage I disease. Because the median time from surgery to diagnosis of brain metastases was 10 months, with a range of 1 month to 55 months, we examined this further. Overall, 23% of patients had either a CT, or less commonly an MRI, at the time of initial diagnosis. Brain imaging was performed more frequently in patients with
stage II (N1) disease compared with stage I (N0) disease (34% vs 21%; P < .001). Among patients who subsequently developed brain metastases, 37% had negative
brain imaging at diagnosis compared with 23% of patients who did not develop brain metastases (P ¼ .02), reflecting the higher disease stage of patients who developed
brain metastases.
DISCUSSION
This is 1 of the largest series evaluating the development of brain metastases in patients with resected, early stage NSCLC (Table 3). Several of our observations are notable. First, the overall incidence of developing brain metastases was relatively low.With a median follow-up of 33 months, the crude rate was 6%. The 5-year actuarial rate was 10%. This is similar to other studies which report crude brain metastases rates ranging from 2% to 16% for stage I disease12,18–26 and 3% to 19 % for stage II disease. 24,26–30 Second, although the overall rate was low, brain metastases developed in 29% patients who failed at distant sites. Thus, future efforts to reduce the risk of distant recurrence will need to account for failures in the central
nervous system. Third, we identified several factors that were associated with a higher risk of developing brain metastases, including younger age, increasing size, lymphovascular space invasion, and hilar lymph-node involvement. However, although each of these factors were independently associated with developing brain metastases, the hazard ratios were relatively small, ranging from 1.03 to 1.87, suggesting that standard clinical and pathological factors alone may not be sufficient to identify patients prospectively at highest risk who may be candidates for interventions such as prophylactic cranial irradiation. Clinical and Pathologic Risk Factors Although several clinical and pathologic factors have been associated with the development of brain metastases in locally advanced disease,31–35 few studies have evaluated risk factors in early stage disease (Table 3). On multivariate analysis, we identified 4 factors that were independently associated with a higher risk of developing brain metastases. These included younger age, increasing size, lymphovascular space invasion, and hilar lymph-node involvement. In our series, younger age was associated with an increased incidence of developing brain metastases, albeit
with a small hazard ratio of 1.03 per year. Whereas several other series found a similar finding,7,31,36 age in other series was not prognostic.33,34,37 Tumor size is an important prognostic factor in NSCLC and a primary basis for the TNM staging system. 15 Bajard et al showed that clinical T classification was significant for development of brain metastases in a cohort of 305 NSCLC patients (168 with stage I-II disease). 36 Similarly, in a study of 264 NSCLC patients at all stages (31% with stage I-II disease), Mujumoor and colleagues showed that the probability of developing metastases to the brain was significantly increased with increasing clinical tumor size.38 In contrast, other studies have shown that for resected stage IIIA patients, pathologic T classification was not significantly associated with incidence of brain metasteses.33,34.  Another important prognostic factor for NSCLC is nodal status. In our series, hilar nodal involvement was found to be significantly associated with the development of brain metastases. Mujumoor et al also found an increased risk of developing brain metastases with increasing lymph-node status.38 Similarly, results from the Lung Cancer Study Group found that increasing T and N classifications
were associated with a higher risk of developing isolated brain metasetases.22 Several other studies have confirmed this finding.31,37,39 Although the majority of
patients in our series underwent sampling of hilar and mediastinal lymph nodes at the time of surgery, some patients in the earlier years of the current study did not.
Whereas hilar lymph node sampling was associated with a higher risk of developing brain metastases on univariate analysis, this was not significant on multivariate analysis. Presumably, patients who underwent wedge resection alone had small peripheral tumors at relatively low risk of harboring occult hilar disease or developing brain metastases. Several studies have reported that lymphovascular invasion may have prognostic value in NSCLC,40–43 but only a few have reported that lymphovascular space invasion is a risk factor for developing brain metastases.44,45 A study by Tsuchiya et al of patients with stage IA NSCLC noted a trend for higher rates of brain metastases in patients with lymphovascular invasion during follow-up
compared with patients without lymphovascular space invasion (8 of 23 vs 2 of 22; P ¼ .07).45 Histology was not a statistically significant factor in our series (P ¼ .14). Nonsquamous histology, adenocarcinoma in particular, has been associated with an increased risk of developing brain metastases in some,33,34,38,39,46 but not all,35–37 studies. In general, the hazard ratios that we observed were small. Thus, identifying patients at highest risk of developing
brain metastases on the basis of standard clinical
and pathological factors may not be reliable. More robust
methods are needed to identify which patients are at highest
risk of developing brain metastases.
Biological Predictors
Investigators at Duke have performed immunohistochemical
analyses of pathological specimens from patients
who did and did not develop brain metastases. Patients
with early stage NSCLC who developed isolated brain
metastases had a significantly higher expression of molecular
markers, such as p53 and urokinase plasminogen activator,
and a lower expression of E-cadherin by
immunohistochemical analysis.47 A similar study from
the Brigham and Women’s Hospital (Boston, Mass) also
demonstrated that immunohistochemical markers,
including Ki-67, caspase-3, VEGF-C, and E-cadherin,
may be able to predict patients at higher risk of developing
brain metastases.48
Recently, genomic signatures of malignancies are
being exploited to better understand prognosis and to
identify patients who may benefit from further intervention.
49 Recent studies have shown that gene-expression
profiles can distinguish lung cancer patients at particularly
high risk of disease recurrence.50,51 These genomic signatures
have not, however, been evaluated for their ability to
predict patterns of failure.
Grinberg-Rashi et al analyzed 12 candidate genes
hypothesized to be associated with the development of
brain metastases.52 By using 142 frozen tissue samples,
expression levels of 3 genes (CDH2, KIFC1, and FALZ)
were found to be independently associated with the development
of brain metastases in NSCLC. A model was
developed that stratified patients into low-, intermediate-,
and high-risk cohorts on the basis of the 3-gene expression
model. The investigators observed that 37% of patients
with early NSCLC in the high risk cohort developed brain
metastasis within the first 2 years after diagnosis compared
with a 10% risk for patients in the low-risk and intermediate-
risk groups (P < .02). The investigators were not able
to validate their model with a similar dataset, but did
show that CDH2 levels, assessed with immunohistochemistry,
were concordant with their initial findings.
If patients at very high risk of developing brain metastases
could be reliably identified, this would facilitate
future studies of prophylactic cranial irradiation (PCI) or
other interventions, such as close surveillance.
Prophylactic Cranial Irradiation
PCI has been shown to improve survival in small-cell lung
cancer53 and is currently considered standard for limited
stage small-cell lung cancer. Emerging data show a benefit
to patients with extensive disease.54 The role of PCI in
NSCLC remains controversial and is not currently considered
standard. The studies that have been completed to
date have included only patients with stage III disease.
In an early study from M. D. Anderson Cancer Center
(Houston, Tex), 97 patients with locally advanced
NSCLC were randomized to either PCI (3 Gy \u000210 fractions)
or observation.55 The incidence of brain metastases
was markedly decreased with PCI (4% vs 27%; P < .01),
although there was no difference in survival secondary to
Original Article
6 Cancer Month 00, 2010
extracranial progression. In this study, PCI significantly
reduced the incidence of brain metastases only in those
with squamous-cell histology.
A subsequent Radiation Therapy Oncology Group
(RTOG) study randomized 187 patients with locally
advanced or resected stage II-III NSCLC to PCI or observation.
56 Only patients with adenocarcinoma or large cell
histology were eligible. The incidence of brain metastases
was not statistically different with PCI (9% vs 19%, P ¼
.10), and there was no difference in overall survival. In an
unplanned subgroup analysis among patients with
resected stage II-III NSCLC, PCI reduced the incidence
of brain metastases from 25%(3 of 12) to 0%(0 of 14)(P
¼ .06). It was hypothesized that the lack of difference
between the 2 treatment arms was due to inadequate local
control and lack of systemic therapies.
In a more recent German study, operable stage IIIA
NSCLC patients were randomized to surgery and postoperative
radiotherapy (Arm A) versus preoperative chemotherapy
followed by surgery (Arm B).57 Patients in Arm B
were scheduled to receive PCI (2 Gy \u000215), but patients in
Arm A were not. Thus, PCI was not a randomized variable.
However, it was found that patients who received
PCI had a significantly lower incidence of brain metastases
as first site of failure (8% with vs 35% without PCI, P
¼ .04) with minimal effects on subsequent neurocognitive
function. Because first-line therapy differed in the 2
arms, namely with the use of platinum-based chemotherapy
in Arm B, the differences in brain metastases-free survival
may have been affected by factors other than PCI.
The recently closed RTOG 0214 study examined
PCI versus observation in 340 stage III NSCLC
patients.62 Although PCI decreased the risk of developing
brain metastases at 1 year (7.7% vs 18%, P ¼ .004), there
was no significant difference in overall survival (75.6% vs
76.9%, P ¼ .86) nor in disease-free survival (56.4% vs
51.2, P ¼ .11).
Further work is necessary to identify patients with
lung cancer (all stages) who are at very high risk of developing
brain metastases, who may be treated with PCI.
Study Limitations
We acknowledge the limitations of our retrospective analysis.
First, the median time to the development of brain
metastases was 10 months. This is consistent with other
studies in which the median time was 7.5-12.5
months.34,35,48,63 However, not all patients had CT or
MRI imaging of the brain at the time of diagnosis, and it
is possible that some patients had asymptomatic intracranial
disease at the time of surgery. Interestingly, patients
who subsequently developed brain metastases were
actually more likely to have had a negative staging CT or
MRI of the brain compared with patients who did not develop
brain metastases, reflecting the finding that brain
imaging was performed more commonly in patients with
N1 disease. Most clinical guidelines currently recommend
brain imaging only for patients classified as having N disease
or with concerning symptoms.
Furthermore, follow-up brain imaging was not
standardized and typically only performed in symptomatic
patients or those who had recurrences elsewhere.
This would, if anything, underestimate the true risk of
developing brain metastases. Finally, not all patients
underwent biopsy or resection of their brain metastases to
pathologically confirm the diagnosis. However, the clinical
and radiographic picture was consistent with brain metastases
from lung cancer in all patients. The primary
strength of our study was the large number of patients analyzed,
which facilitated identifying multiple clinical and
pathologic risk factors for developing brain metastases in
early stage NSCLC.
Conclusions
In this large series of patients with resected early stage
NSCLC, the 5-year actuarial risk of developing brain metastases
was 10%. Although several clinical and pathological
factors were shown to be independently associated
with an increased risk of central nervous system recurrence,
a better understanding of biological susceptibility is
needed to prospectively identify subgroups at highest risk.
The therapeutic ratio of PCI in NSCLC would be
increased if such a population could be identified. Furthermore,
with improvements in local and systemic disease
control, decreasing the risk of developing brain
metastases, in both early and locally advanced disease, is
expected to become increasingly important.