Keywords
glioblastoma multiforme - survival - local recurrence
Palavras-chave
glioblastoma multiforme - sobrevida - recorrência local
Introduction
With an annual incidence rate of 3 to 4 cases per 100,000 persons[1], glioblastoma multiforme (GBM) is by far the most common malignant primary tumor
of the brain in adults. The overall incidence of primary malignant brain tumors is
reported to be around 2.74 per 100,000 persons in Iran.[2] Patients with GBM have a short survival term, and frequently present with tumor
recurrence; therefore, an effective management of these patients is crucial. The longest
reported survival terms, despite aggressive therapy, are lower than two years.[3]
[4]
[5]
[6]
[7]
[8]
[9] Aggressive therapies, including surgery, chemotherapy and radiation are not only
costly, but bear additional complications.[10]
[11]
[12] Nearly all patients with GBM have a poor quality of life, and health related quality
of life (HRQoL) is defined as a multidimensional concept covering physical, psychological,
and social domains, as well as symptoms induced by the disease and its treatment.[13] Treating the tumor is intensive and time-consuming, and treatment complications,
as well as tumor recurrences, are common. Effective treatment will improve the patients'
performance status[14], neurocognitive function[15], overall quality of life[16] and overall survival.[4]
[17]
[18]
[19] In addition, the effective treatment will also improve the psychological health
of the patients. Achieving high quality of life in patients with GBM requires the
cooperation of various specialists, and certain loss of quality of life is intrinsic
to cancer patients. However, one should identify and target the factors that will
help the radiotherapists, oncologists, and neurosurgeons improve the overall survival
of the patients without the recurrence of the tumor.
Our study assesses the factors that are associated with prolonged survival, improved
quality of life and reduced tumor recurrence in patients with GBM.
Material and Methods
Patient Selection
A total of 153 patients with supratentorial GBM tumors were admitted to a referral
tertiary academic center between 2005 and 2013 at a hospital in Tehran, Iran. In all
cases, the GBM patients were diagnosed with the pathology, as confirmed by two senior
neuropathologists, and the grading criteria was based on the classification system
of the World Health Organization (WHO).[20]
[21] Patients at any age with a tissue-proven diagnosis of supratentorial GBM (WHO Grade
IV) were included in the study. Patients who had serious concomitant malignant or
chronic diseases, and patients with infratentorial gliomas and prior lower grade gliomas
were excluded from the analysis to create a more uniform patient population.
Apart from the research's objectives, all patients received various management procedures
depending on their preoperative assessment and on necessity indicators. Additionally,
all patients were followed-up after undergoing the treatment.
Recorded Variables
The clinical, operative, and hospital course records of the patients who met the inclusion
and exclusion criteria were retrospectively reviewed. The information was collected
from neurosurgery and radiotherapy clinical notes, including the patients' demographics,
presenting symptoms, neurological function and neurologic signs, as well as the neuroimaging
perioperative course and the adjuvant therapy. The Karnofsky Performance Status (KPS)
scale was used to specify the patients' preoperative functional status.[22] The KPS scores were collected during a physical examination by oncologists who were
blind to the outcomes of the patients at the clinical visit, and prior to surgery.
Preoperative sensory deficit was defined as decreased sensation to any stimulant.
Motor deficit was defined as decreased force, as identified by a clinician during
a physical examination. Language deficit was defined as any combination of receptive
or expressive aphasia. Finally, cognitive deficits were defined as confusion or memory
loss. The magnetic resonance imaging (MRI) characteristics were recorded, including
the specific lobe location and eloquent brain involvement. This assessment was based
on radiographic, not clinical, criteria. Unfortunately, the sizes of the lesions were
not registered in the records. The geometric estimation of the volume of the resected
tumor was based on the comparison of the enhanced tumor margin in the gadolinium-enhanced
T1-weighted sequences of pre-op MRIs with those of post-op MRIs obtained less than
48 hours after tumor resection. The resections were then defined as either gross total
resections (GTRs; > 99% resection) or subtotal resections (STR; 90–99% resection)
by an independent neuroradiologist who was blind to the outcomes of the patients.
The patients who underwent biopsies were not classified as having undergone a resection.
The date of death was recorded for any patient whose record was available in the hospital
records. Time until death was defined as the time from the initial glioblastoma diagnosis
(with the pathology) until death. Patients whose deaths were unconfirmed were classified
as lost to the follow-up at the time of the last clinic visit. The concepts of stable
disease, local recurrence and progression were defined according to the Response Assessment
in Neuro-oncology (RANO) criteria. Briefly, the RANO criteria are based on the evaluation
of the product of the maximal cross-sectional diameters of an enhancing lesion in
the post-gadolinium enhanced T1-weighted MRI and/or T2-weighted /flair sequences before
and 4 weeks after surgery. Depending on meeting a complex criteria comprised of the
following, (i) postoperative radiographic assessment of tumor size based on the extent
of the preoperative involvement (that is, disappearance, reduction or progression
of all measurable and non-measurable lesions on gadolinium-enhanced T1-weighted images
in addition to stable, regressing, or progressing tumor size in the T2-weighted/flair
images), (ii) clinical status (stable, improved, or deteriorated condition), (iii)
the use of corticosteroids (that is, none, stable/decreased or increased [conditional]
dosage of medication), and (iv) the presence of new lesions (that is, none or present);
the patients with glioblastoma were divided into 4 categories: “complete response,”
“partial response,” “stable disease” or “progressive disease.” For the present manuscript,
the groups of patients with “complete response” and “partial response” on the RANO
criteria were designated as having a “stable disease”, and the group of patients with
“stable diseases” and “progressive diseases” on the RANO criteria were defined as
having “local recurrence.”
Perioperative Treatment
All patients had been visited by neurosurgeons and radiation oncologists before surgery.
The general aim of the neurosurgeons was to achieve GTR of the tumor when possible.
Subtotal resection was achieved primarily when the tumor involved eloquent brain as
confirmed by intraoperative mapping and/or monitoring, and surgical navigation (computed
tomography [CT] and/or MRI wand) was used in all cases. Implant therapy was not performed
in any of the patients. Radiation oncologists treated all the patients with 60 Gy
2-dimensional or 3-dimensional radiotherapy in 30 fractions. The patients were prescribed
6 sessions of adjuvant chemotherapy with 150 mg/m2 over 5/28 days in 6 cycles of the first-line agent temozolomide (TMZ) in addition
to the concurrent chemotherapy with 75 mg/m2/day TMZ 1 hour prior to radiotherapy. A total of 6 cycles of 110mg/m2 lomustine (CCNU) adjuvant chemotherapy was performed as the second-line agent because
of inaccessibility to TMZ due to the cost of it and the lack of insurance coverage.
Although procarbazine, CCNU and vincristine (PCV) remain the salvage chemotherapy
regimen in patients with high-grade gliomas,[23] the alternative agent CCNU was used as an adjuvant chemotherapy regimen in this
group of patients because of the lower complication rates, better tolerability and
comparable survival rate to the use of PCV in our country.[24]
In this study, many patients were denied surgery or chemotherapy options, or both,
because of the inability of the patients or their families to pay for the treatments.
Therefore, apart from the study's objectives, some patients were treated depending
on their preoperative assessment and based on necessity indicators depending on standard
treatment options,[25]
[26] and some patients received incomplete treatments perforce. The decision involved
input from a surgeon, a radiation oncologist and the patients themselves. Recurrent
tumors were usually discovered on follow-up visits via postoperative MRI performed
at 3-month intervals following surgery, or at the time that any symptoms developed.
Statistical Analysis
All analyses were performed using the Statistical Package for the Social Sciences
(SPSS, IBM Corp. Armonk, NY, US) software, version 20. Summary data was presented
as mean ± standard deviation (SD) for parametric data, and nonparametric data, as
median (interquartile range [IQR]). For the intergroup comparison, the Student's t-test was used for parametric data, and the Mann-Whitney U-test was used for nonparametric
data. The percentages were compared using the chi-square test or Fisher's exact test
where appropriate. Survival as a function of time was plotted using the Kaplan-Meier
method. Moreover, log-rank analysis was used to compare the Kaplan-Meier plots. The
factors associated with overall survival were assessed using the Cox proportional
hazard regression models for multivariate associations. For this purpose, all variables
associated with survival in the univariate analysis (p < 0.10) were included.
The factors predicting the outcomes of survival and local recurrence were separately
analyzed using the neural network analysis. For this purpose, two models for neural
network analyses were developed to firstly predict the survival and secondly to predict
local recurrence using selected baseline characteristics of the patients.
The analysis of a neural network uses a learning algorithm to define the nonlinear
mathematical transfer functions to modify the synaptic weights of a network's processing
units in an orderly fashion to obtain the desired outcome prediction (training datasets).
Both the weights and the value of the activation functions can be adjusted during
the training of an artificial neural network. However, this is impractical, as it
would be simpler to only adjust for a single parameter. To surpass this problem, the
bias neuron is generated. The bias neurons in layer 1 are connected to all the neurons
in the following layer, but with none of the neurons present in the previous layer.
The hidden layer contains unobservable network nodes (units). Each hidden unit is
a function of the weighted sum of the inputs. It is similar to the correlation coefficient
in the linear regression model. In all subsequent analyses, values of p < 0.05 were considered statistically significant.
Results
Preoperative, Perioperative and Postoperative Characteristics of the Patients
Among the 155 patients diagnosed with supratentorial primary GBM, 153 met the eligibility
criteria and were included in the analysis. The pre-, peri- and postoperative characteristics
of these 153 patients (99 men, 64.7% of the total study population) are summarized
in [Table 1]. The mean ± SD age of the patients was 55.69 ± 15.10 years at the time of the diagnosis.
In total, 40 patients (26.1%) were younger than 45 years, 88 patients (57.5%) were
between 45 and 70 years old, and 25 patients (16.3%) were older than 70 years of age.
The median preoperative KPS was 60 (IQR: 50–80, range: 20–100). A total of 52 patients
did not express any neurologic symptoms at their consultations. Among 101 patients
with neurologic signifiers, the major symptoms presented are described in declining
order: seizures in 36 patients (23.5%); motor deficits in 21 patients (13.7%); sensory
and language deficits in 15 patients (9.8%); visual deficits in 9 patients (5.9%);
and cognitive deficits (memory loss/confusion) in 5 patients (3.3%). The median duration
of the symptoms was 2 months prior to the diagnosis of the pathology. A total of 81
tumors (52.9%) were found in the right hemispheres, with the remainder involving the
left hemispheres. Ninety-four tumors (61.4%) involved only 1 brain lobe, while all
other tumors involved 2 brain lobes. Twenty-nine tumors (19.0%) involved the frontal
lobe, 37 tumors (24.2%), the parietal lobe, 18 tumors (11.8%), the temporal lobe,
8 tumors (5.2%), the occipital lobe, 24 tumors (15.7%), the temporoparietal lobe,
16 tumors (10.5%), the parieto-occipital lobe and 21 tumors (13.7%) involved other
areas. A total of 60 patients (39.2%) underwent biopsy, 91 patients (59.5%) underwent
near total resection (NTR) or STR, and only 2 patients (1.3%) underwent GTR. There
were no cases of perioperative mortality. Radiotherapy was performed in all patients
(100%) with a median dose of 60 Gy in 30 fractions. A total of 100 patients (94.8%)
underwent 2-dimensional radiotherapy, whereas 8 patients (5.2%) underwent 3-dimensional
radiotherapy. Of the 153 patients, 78 (51%) underwent only radiotherapy, 57 (37.3%)
underwent adjuvant chemotherapy, and 18 (11.8%) underwent concurrent + adjuvant chemotherapy.
Concurrent + adjuvant chemotherapy was performed using TMZ. Among the 75 patients
who underwent chemotherapy, TMZ was administered to 39 (25.5%), and CCNU was administered
to 36 (23.5%). At the last follow-up, 136 (88.9%) patients had died, 10 patients (6.5%)
were alive, and 7 patients (4.6%) did not make appointments, and had an unknown status.
The median follow-up time for the surviving patients was 14 months (IQR: 10–20 months).
The median overall survival rate of the patients was 14 months (IQR: 9–17 months).
The median survival rates at 3, 6, 9, 12, 18, 24 and finally, 32 months of the patients
in this study were 98.0%, 85.6%, 70.5%, 55.5%, 22.8%, 15.6% and 5.8% respectively.
The patients were divided into certain categories to match case and controls for better
analysis ([Table 2]).
Table 1
pre-, peri- and postoperative characteristics of the patients
|
Study population → N = 153
|
|
|
Characteristics
|
N (percent)
|
|
Age (mean ± SD)
|
55.69 ± 15.10
|
|
Male
|
99 (64.7%)
|
|
Preoperative factors
|
|
KPS > 60
|
48 (31.4%)
|
|
KPS = 60
|
62 (40.5%)
|
|
40 < KPS < 60
|
35 (22.9%)
|
|
KPS < 40
|
8 (5.2%)
|
|
Neurologic sign
|
101 (66.0%)
|
|
Confusion/memory loss
|
5 (3.3%)
|
|
Language deficit
|
15 (9.8%)
|
|
Motor deficit
|
21 (13.7%)
|
|
Sensory deficit
|
15 (9.8%)
|
|
Seizure
|
36 (23.5%)
|
|
Visual deficit
|
9 (5.9%)
|
|
Mass location
|
|
Right hemisphere
|
81 (52.9%)
|
|
Frontal lobe
|
29 (19.0%)
|
|
Parietal lobe
|
37 (24.2%)
|
|
Temporal lobe
|
18 (11.8%)
|
|
Occipital lobe
|
8 (5.2%)
|
|
Temporoparietal lobes
|
24 (15.6%)
|
|
Others
|
37 (24.2%)
|
|
Perioperative factors
|
|
Operative method
|
|
Biopsy
|
60 (39.2%)
|
|
Total and near total resection
|
93 (60.8%)
|
|
Chemoradiation plan
|
|
Radiotherapy
|
78 (51.0%)
|
|
Radiotherapy + adjuvant chemotherapy
|
57 (37.2%)
|
|
Radiotherapy + concurrent chemotherapy + adjuvant chemotherapy
|
18 (11.8%)
|
|
Chemotherapy drugs
|
|
TMZ
|
39 (25.5%)
|
|
CCNU
|
36 (23.5%)
|
|
Radiation method
|
|
2D
|
145 (94.8%)
|
|
3D
|
8 (5.2%)
|
|
Postoperative factors
|
|
Died at last follow-up, n
|
136 (88.9%)
|
|
Follow-up months (range)
|
49 (3–49)
|
|
Median survival (months)
|
14.0
|
|
Mean survival (months)
|
15.34 ± 9.63
|
|
3-month survival rate
|
98.0%
|
|
6-month survival rate
|
85.6%
|
|
9-month survival rate
|
70.5%
|
|
12-month survival rate
|
55.5%
|
|
18-month survival rate
|
22.8%
|
|
24-month survival rate
|
15.6%
|
|
32-month survival rate
|
5.8%
|
|
Recurrence
|
|
Tumor recurrence, n
|
115 (75.2%)
|
|
progression-free survival (median)
|
7.1
|
Abbreviations: 2D, two-dimensional; 3D, three-dimensional; CCNU, lomustine; KPS, Karnofsky
Performance Status; TMZ, temozolomide.
Table 2
Case control matching
|
Categorization of the patients
|
|
|
Study population → N = 153
|
|
|
Groups:
|
N (percent)
|
|
Age groups (two categories)
|
|
• Age ≥ 70
|
28 (18.3%)
|
|
• Age < 70
|
125 (81.7%)
|
|
Age groups (three categories)
|
|
• Age ≥ 70
|
28 (18.3%)
|
|
• 70 > Age ≥ 45
|
87 (56.9%)
|
|
• Age < 45
|
38 (24.8%)
|
|
Survival
|
|
• More than 14 months
|
77 (50.3%)
|
|
• Less than 14 months
|
76 (49.7%)
|
|
TMZ versus without TMZ
|
|
• Radiotherapy + resection + TMZ
|
28 (18.3%)
|
|
• Radiotherapy + resection
|
42 (27.5%)
|
|
TMZ
|
|
• Using TMZ
|
39 (25.5%)
|
|
• Not using TMZ
|
114 (74.5%)
|
|
CCNU versus without CCNU
|
|
• Radiotherapy + resection + CCNU
|
21 (13.7%)
|
|
• Radiotherapy + resection
|
42 (27.5%)
|
|
CCNU
|
|
• Using CCNU
|
36(23.5%)
|
|
• Not using CCNU
|
117(76.5%)
|
|
TMZ versus CCNU
|
|
• Radiotherapy + resection + TMZ
|
28 (18.3%)
|
|
• Radiotherapy + resection + CCNU
|
21 (13.7%)
|
|
Adjuvant versus concurrent chemotherapy
|
|
• Radiotherapy + resection + adjuvant
|
33 (21.6%)
|
|
• Radiotherapy+ resection + concurrent + adjuvant
|
16 (10.5%)
|
|
Resection versus chemotherapy
|
|
• Radiotherapy + resection
|
42 (27.5%)
|
|
• Radiotherapy + biopsy + chemotherapy
|
26 (17.0%)
|
|
Resection versus without resection
|
|
• Radiotherapy + chemotherapy + resection
|
49 (32.0%)
|
|
• Radiotherapy + chemotherapy + biopsy
|
26 (17.0%)
|
|
Resection
|
|
• Radiotherapy + resection
|
42 (27.5%)
|
|
• Radiotherapy + biopsy
|
34 (22.2%)
|
Abbreviations: CCNU, lomustine; TMZ, temozolomide.
Factors Independently Associated with Survival
Univariate Analysis
We investigated the factors associated with the overall survival and progression-free
survival using the Kaplan-Meier analysis. We found that age (p = 0.005), confusion and/or memory loss (p < 0.001), CCNU (p < 0.001), TMZ (p < 0.001), KPS (p < 0.001), operative method (p < 0.001), TMZ versus CCNU (p = 0.007), 2D versus 3D radiation protocol (p < 0.001), frontal lobe involvement (p = 0.009) and local recurrence (p < 0.001) had various degrees of impacts on both the overall survival and progression-free
survival rates of our patients with glioblastoma multiforme ([Table 3]).
Table 3
Univariate analysis of the pre- peri- and postoperative characteristics of the patients
using the Kaplan-Meier analysis
|
Group:
|
A
|
B
|
C
|
D
|
|
Time:
Status:
|
Overall survival
Recurrence: No
|
Overall survival
Recurrence: Yes
|
Overall survival
Dead
|
Progression-free survival
Recurrence: Yes
|
|
Age groups (two categories)
|
No[#] (p = 0.532)
|
Yes* (p = 0.027)
|
Yes (p = 0.009)
|
No (p = 0.738)
|
|
Age groups (three categories)
|
No (p = 0.066)
|
No (p = 0.060)
|
Yes (p = 0.005)
|
No (p = 0.731)
|
|
Motor deficit
|
No (p = 0.300)
|
No (p = 0.910)
|
No (p = 0.584)
|
No (p = 0.052)
|
|
Confusion and/or memory loss
|
Yes (p < 0.001)
|
No (p = 0.222)
|
No (p = 0.307)
|
No (p = 0.855)
|
|
Seizure
|
No (p = 0.414)
|
No (p = 0.403)
|
No (p = 0.135)
|
No (p = 0.089)
|
|
CCNU
|
No (p = 0.114)
|
No (p = 0.108)
|
No (p = 0.144)
|
Yes (p < 0.001)
|
|
TMZ
|
No (p = 0.450)
|
Yes (p < 0.001)
|
Yes (P <0.001)
|
Yes (p = 0.003)
|
|
Resection versus biopsy
|
No (p = 0.085)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
|
Resection versus chemotherapy
|
No (p = 0.827)
|
No (p = 0.211)
|
No (p = 0.171)
|
Yes (p < 0.001)
|
|
Resection versus without resection
|
No (p = 0.306)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
Yes (p = 0.030)
|
|
TMZ versus CCNU
|
No (p = 0.249)
|
Yes (p = 0.007)
|
No (p = 0.315)
|
No (p = 0.935)
|
|
CCNU versus without CCNU
|
No (p = 0.237)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
Yes (p = 0.001)
|
|
TMZ versus without TMZ
|
No (p = 0.339)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
Yes (p = 0.003)
|
|
Occipital lobe
|
No (p = 0.053)
|
No (p = 0.067)
|
No (p = 0.685)
|
No (p = 0.468)
|
|
Frontal lobe
|
No (p = 0.051)
|
Yes (p = 0.009)
|
Yes (p = 0.034)
|
No (p = 0.912)
|
|
Operative method
|
No (p = 0.056)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
Yes (p = 0.016)
|
|
KPS
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
|
2D versus 3D radiation
|
Yes (p < 0.001)
|
Yes (p = 0.004)
|
No (p = 0.547)
|
−
|
|
Chemotherapy plan
|
No (p = 0.090)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
Yes (p < 0.001)
|
|
Recurrence
|
−
|
−
|
Yes (p < 0.001)
|
−
|
Abbreviations: 2D, two-dimensional; 3D, three-dimensional; CCNU, lomustine; KPS, Karnofsky
Performance Status; TMZ, temozolomide.
Notes: A: Time defined as overall survival and status defined as no tumor recurrence.
B: Time defined as overall survival and status defined as tumor recurrence.
C: Time defined as overall survival and status defined as death.
D: Time defined as progression-free survival and status defined as tumor recurrence.
*Yes means there is a significant correlation between an obvious factor and survival
or recurrence.
# No means there is not a significant correlation between an obvious factor and survival
or recurrence.
Multivariate Analysis
All variables associated with survival in the univariate analysis (p < 0.10) and clinically important variables were included in the multivariate proportional
hazards regression model. We found that age (hazard ratio [HR] [95% CI (confidence
interval)], 2.939 [1.73–4.99], p < 0.001), operative method (HR [95% CI], 7.416 [3.81–14.42], p < 0.001), TMZ (HR [95% CI], 11.723 [5.46–25.13], p < 0.001), CCNU (HR [95% CI], 8.139 [4.04–16.38], p < 0.001), occipital lobe involvement (HR [95% CI], 3.088 [1.81–5.25], p < 0.001) and KPS (HR [95% CI], 4.831 [3.00–7.77], p < 0.001) had various degrees of impact on both the overall survival and progression-free
survival rates of our patients with glioblastoma multiforme ([Table 4]).
Table 4
Multivariate analysis of the factors associated with overall survival and local recurrence
using the Cox regression models
|
Group:
|
A
|
B
|
C
|
|
Time:
Status:
|
Overall survival
Recurrence
|
Overall survival
Death
|
Progression-free survival
Recurrence
|
|
Hazard Ratio (94% CI)
p-value
|
Hazard Ratio (95% CI)
p-value
|
Hazard Ratio (95% CI)
p-value
|
|
Age groups (two categories)
|
2.939 (1.73–4.99)
p < 0.001
|
3.081 (1.89–5.01)
p < 0.001
|
−
|
|
TMZ
|
11.723 (5.46–25.13)
p < 0.001
|
4.906 (2.51–9.56)
p < 0.001
|
1.394 (0.75–2.58)
p = 0.292
|
|
CCNU
|
8.139 (4.04–16.38)
p < 0.001
|
4.155 (2.19–7.86)
p < 0.001
|
2.047 (1.27–3.29)
p = 0.003
|
|
Operative method
|
7.416 (3.81–14.42)
p < 0.001
|
3.880 (2.00–7.50)
p < 0.001
|
0.493 (0.30–0.80)
p = 0.005
|
|
KPS
|
4.831 (3.00–7.77)
p < 0.001
|
6.078 (3.85–9.57)
p < 0.001
|
7.292 (4.77–11.12)
p < 0.001
|
|
Occipital lobe
|
3.088 (1.81–5.25)
p < 0.001
|
1.599 (0.95–2.69)
p = 0.077
|
−
|
|
Resection versus chemotherapy
|
−
|
−
|
0.171 (0.08–0.33)
p < 0.001
|
Abbreviations: 95% CI, 95% conficence interval; CCNU, lomustine; KPS, Karnofsky Performance
Status; TMZ, temozolomide.
Notes: A: Time defined as overall survival and status defined as tumor recurrence.
B: Time defined as overall survival and status defined as death.
C: Time defined as progression-free survival and status defined as tumor recurrence.
Neural Network Analysis
The variables with the greatest impact on the survival rate of the included patients
were considered for the neural network analysis (input variables for the outcome of
survival: age, occipital lobe involvement, KPS, operative method and the use of CCNU
and TMZ). We found the importance of the variables to predict survival in the following
declining order: KPS = 30.6%, operative method = 20.4%, TMZ = 17.0%, CCNU = 15.0%,
age = 13.5%, and occipital lobe involvement = 3.6% ([Fig. 1]). In this model, four hidden layers and one bias neuron were germane to the calculation.
Fig. 1 Result of neural network analysis for predicting survival. The input variables are
those that had an impact on survival on the multivariate analysis from Cox regression
model analysis. We found the importance of the variables to predict survival as follows:
KPS = 30.6%, operative method = 20.4%, TMZ = 17.0%, CCNU = 15.0%, age = 13.5%, and
occipital lobe involvement = 3.6%. In this model, four hidden layers were included
in the calculation.
Factors Independently Associated with Local Recurrence
Univariate Analysis
Out of the 153 patients, 115 (75.2%) had one local recurrence. We analyzed the factors
associated with local recurrence using the Kaplan-Meier analysis that defined time
as progression-free survival, and status as occurrence of local recurrence. We found
that CCNU (p < 0.001), TMZ (p = 0.003), chemotherapy versus resection (p < 0.001), operative method (p = 0.016) and KPS (p < 0.001) were each associated with local recurrence.
Multivariate Analysis
We identified the factors associated with local recurrence using the Cox regression
model analysis that defined time as progression-free survival and status as occurrence
of local recurrence. All variables associated with survival in the univariate analysis
(p < 0.10), as well as the clinically important variables, were included in the multivariate
proportional hazards regression model. We found that the operative method (HR [95%
CI], 0.493 [0.30–0.80], p = 0.005), CCNU (HR [95% CI], 2.047 [1.27–3.29], p = 0.003), resection versus chemotherapy (HR [95% CI], 0.171 [0.08–0.33], p < 0.001) and KPS (HR [95% CI], 7.29 [4.77–11.12], p < 0.001) as significant risk factors for local recurrence ([Table 4]). Interestingly, TMZ (HR [95% CI], 1.394 [0.75–2.58], p = 0.292) was not a significant predictor of local recurrence.
Neural Network Analysis
Significant variables from the multivariate model of local recurrence were included
as input variables in the neural network analysis (input variables: KPS, operative
method and the use of CCNU and TMZ). Subsequently, we found the importance of the
variables to predict local recurrence in the following decreasing order: KPS = 41.5%,
operative method = 21.8%, TMZ = 21.5%, and CCNU = 15.2% ([Fig. 2]). In this model, four hidden layers and one bias neuron were generated.
Fig. 2 Result of neural network analysis for predicting local recurrence. The input variables
are those that had an impact on local recurrence on the multivariate analysis. We
found the importance of the variables to predict local recurrence as follows: KPS = 41.5%,
operative method = 21.8%, TMZ = 21.5%, and CCNU = 15.2%. In this model, four hidden
layers were included in the calculation.
Discussion
The Karnofsky Performance Status (KPS) scale was the most important factor associated
with decreasing survival in this study ([Fig. 3]). We categorized the patients in four groups for KPS. The first group was composed
of patients with KPS > 60. The second group comprised patients with KPS = 60. The
third group included patients with 40 ≤ KPS < 60. The fourth group featured patients
with KPS < 40. It is interesting that survival decreased equiponderant with the decreasing
KPS scores. Among all four groups, there is a statistically significant correlation
(p = 0.000) between the KPS scores and decreased survival. Many studies have verified
that a lower KPS score has a correlation with decreasing survival in GBM patients.[27]
[28]
[29]
[30]
[31] Abdullah Kalil et al conducted a study on factors associated with increased survival
after surgical resection on GBM patients of more than 80 years of age in which they
found a statistically significant correlation between the KPS and overall survival.[30] In another study, Chaichana et al considered preoperative factors associated with
decreased survival for older patients who underwent resection of a GBM, and found
that one of the preoperative factors that was independently associated with decreased
survival was a KPS score of less than 80.[31] Chaichana et al, in another study, evaluated functional outcomes over time for patients
with glioblastoma, and found that a preoperative KPS score of ≥ 90 is associated
with a prolonged functional outcome. Their findings may help guide treatment strategies
aimed at improving the quality of life of patients with glioblastoma.[32] The KPS was not statistically important in correlations with local recurrence in
this study. Therefore, it seems that the KPS has a greater impact on quantity of life
than on quality of life.
Age was another important factor associated with decreased survival in our study.
We assessed the age effect on survival in two different ways. Initially, we found
that the cut-off point for age in this study was 70 years. Patients with more than
70 years of age had significantly lower survival rates. Since the presence of other
comorbidities in old age is more common, we assessed the correlation between age groups
(age ≤ 45, 45 < age ≤ 70, age > 70) and survival. We found that there is a correlation
between age and survival. Age and preoperative neurological function are the two factors
most consistently associated with survival in several studies;[1]
[31]
[32]
[33] however, we could not find any correlation between age and local recurrence.
Chemotherapy, in the present study, was shown to decrease local recurrence and improve
survival. Chemotherapy plans (radiotherapy alone versus radiotherapy + adjuvant chemotherapy
versus radiotherapy + concurrent chemotherapy) cause a demonstrable statistically
significant decrease in local recurrence (p < 0.001). Also, patients who received CCNU and TMZ had significant lower local recurrence
rates and higher overall survival rates versus patients to whom CCNU or TMZ was not
administered ([Fig. 3A] and [B]). We compare two groups of patients: those who underwent radiotherapy + chemotherapy + biopsy
versus the group of patients who underwent radiotherapy + resection. The interesting
and important thing here is that chemotherapy was significantly more effective than
resection in decreasing the local recurrence rate. Moreover, chemotherapy was effective
on prolonging the overall survival. However, when we compared the efficacy of the
TMZ versus the CCNU, for the first time, we found that patients who used TMZ had a
higher overall survival than patients who used CCNU (p = 0.007). Johnson et al assessed the glioblastoma survival in the United States before
and during the TMZ era.[34] They found that amongst patients treated with surgery and a radiation-containing
regimen, the median survival rate was of 12.0 months during the period without TMZ
against 14.2 months in the TMZ era. The survival of patients with newly diagnosed
glioblastomas improved from one period to the other, likely due to the use of TMZ.
In a recent experimental study, Harvey et al assessed the anticancer properties of
CCNU in glioblastoma cell lines, and found that the combination of docosahexaenoic
acid (DHA) and CCNU strongly induced Uppsala 87 malignant glioma (U87-MG0 apoptosis
and necrosis as indicated by flow cytometric analysis.[35] They suggested a potential role for a combination therapy of CCNU and DHA for the
treatment of glioblastomas. Other studies recommended using CCNU for recurrent GBMs.[36]
[37]
[38]
Fig. 3 (A) Kaplan-Meier curve suggesting the TMZ effect on overall survival (p < 0.001). (B) Kaplan-Meier curves suggesting the CCNU effect on overall survival (p < 0.001). (C) Kaplan-Meier curves suggesting the resection effect on overall survival (p < 0.001). (D) Kaplan-Meier curves for the overall survival of four groups of patients. The first
group of patients had a KPS score > 60, the second group had a KPS score = 60, the
third group had scores 40 ≤ KPS < 60, and the last group had a KPS score < 40. There
was a statistically significant difference (p < 0.001) among the four groups.
Among our patients, we found that if local recurrence did not occur, the patients
experienced a higher overall survival time. This suggests the necessity of effective
treatments to prevent local recurrence, leading to increasing survival rates.
The role of resection in prolonging survival in our patients appeared in the univariate
and multivariate analyses ([Fig. 3C]). The operative method had a statistically important role in increasing survival
and decreasing local recurrence. Additionally, we compared patients in two groups:
radiotherapy + resection versus radiotherapy + biopsy. Patients who underwent radiotherapy + resection
had higher survival and lower recurrence rates than the biopsy group. This observation
clearly defined the important role of resection in prolonging survival among patients
with poor prognoses, especially those with advanced ages. Chaichana et al assessed
the factors associated with survival for 100 patients with glioblastomas with KPS
scores ≤ 60.[39] They found that the factors associated with improved survival were age < 65 years,
tumor size > 2 cm, radical tumor resection, and TMZ. Chaichana et al, in another study,
assessed the effect of multiple resections on prolonging survival in 578 patients
with GBM.[19] In their study 354, 168, 41, and 15 patients underwent 1, 2, 3, or 4 resections
respectively. The median survival rate for patients who underwent 1, 2, 3, and 4 resections
was of 6.8, 15.5, 22.4, and 26.6 months respectively, and that was statistically significant.
Finally, they concluded that patients with recurrent glioblastomas can have improved
survival rates with repeated resections.
Poor neurologic status before surgery was another factor associated with decreased
survival and increasing local recurrence in our series. We found that confusion and/or
memory loss will decrease survival, while motor deficit will probably increase local
recurrence. In a different study, various neurologic signs have shown to decrease
survival and increase local recurrence rates.[31]
[40]
[41]
[42] We designed a neural network analysis to predict the factors associated with decreasing
survival, which we also found in the multivariate analysis, and the factors associated
with local recurrence. The KPS was the most important factor to predict survival and
local recurrence. We found the importance of each factor in predicting survival and
local recurrence; however, future studies with larger sample sizes are recommended.
Strengths and Limitations
We believe that this study provides several useful insights to identify the factors
associated with survival and local recurrence in patients with GBM. Firstly, the importance
of quantity and quality of life in GBM is equal, and maybe quality of life is preferred,
because of the overall short-term survival of the patients. There are many important
factors associated with survival and local recurrence. The factors that are reversible
are most important because they are the most effective at changing the fate of the
patients.
This study confirms the associations of age, confusion and/or memory loss, CCNU, TMZ,
KPS, operative method, TMZ versus CCNU, 2D versus 3D radiation and frontal lobe involvement
in survival. It also confirms the association of CCNU, TMZ, chemotherapy versus resection,
operative method and KPS with local recurrence. This study also confirmed that CCNU,
TMZ, operative method and KPS are the factors associated with both survival and local
recurrence.
Secondly, studies applying preoperative risk factors in a manner that provides useful
prognostic information have yet to be established, both for survival and local recurrence.
Lastly, this study provides a potentially useful guide that may prognosticate which
GBM patients may benefit from chemotherapy as opposed to radiotherapy and resection.
This means that the aggressive treatment is accompanied by higher survival and lower
local recurrence rates.
This study, however, has some limitations. Firstly, the sample size is not large.
A significantly larger sample size with exact sub-groups will allow a better analysis,
especially for achieving neural network analysis. Secondly, we could not procure some
necessary data from the records, perhaps most importantly the size of each tumor.
Other MRI was missed in this study. Thirdly, some patients did not receive the full
treatment, such as undergoing surgery and/or chemotherapy, because the treatments
were cost-prohibitive. This study also does not account for the potential implication
of molecular markers and genotypes, which may be associated with survival. Recent
studies on GBM patients defined that O6-methylguanine–DNA methyltransferase (MGMT)
promoter methylation leads to prolonged survival after TMZ and radiation therapy compared
with patients without this molecular marker.[43] Additionally, Sanson et al indicated that isocitrate dehydrogenase 1 (IDH1) codon
132 mutation is closely linked to the genomic profile of the tumor, and constitutes
an important prognostic marker in grade 2 to 4 gliomas.[44] These molecular markers, and perhaps other markers associated with survival, were
not analyzed in this study. Additionally, this study was unable to evaluate the other
prognostic factors associated with survival, such as marital status[45] and presence of a caregiver,[46] which have been found in other studies, because these were not consistently recorded
in our patient records. Finally, this study is naturally limited because of its retrospective
design, and, as a result, it is not appropriate to infer direct causal relationships.
Furthermore, we performed multivariate and neural network analyses, and controlled
for potential confounding variables. Given these statistical controls and a relatively
precise outcome measure, we believe that our findings offer useful insights for the
treatment of patients with primary GBM. Prospective studies with huge sample sizes
are needed to provide better data to guide clinical decision making.
Conclusion
Almost all of the patients with GBM will benefit from aggressive therapy, including
radiotherapy, chemotherapy and resection. We cannot guarantee the patients' survival
or guarantee non-recurrence, but it is certain that patients with GBM should be managed
with an effective therapy to reach two goals: higher survival and zero recurrence
rates. These two goals will guarantee better quality and quantity of life for these
patients. In this study CCNU, TMZ, operative method and KPS appear as factors associated
with both increasing survival and decreasing local recurrence rates. A prospective
study with a global partnership and a larger sample size is recommended for the future.
