Clinical variables serve as prognostic factors in a model for survival from glioblastoma multiforme: an observational study of a cohort of consecutive non-selected patients from a single institution

This study included a consecutive series of 225 patients with newly diagnosed GBM (WHO grade IV) recruited from 2005 to 2010 who were not selected other than having ECOG PS 0–2. There were 80 women and 145 men with a median age of 59.2 years (range, 22.6–75.4 years). Of these, 198 patients presented with a single tumor, while 26 patients had multifocal disease (data missing, n = 1). ECOG PS was 0 (n = 132), 1 (n = 66), or 2 (n = 19) [data missing, n = 8]. Patient demographics are shown in Table 1.

Patients underwent surgery, taking either a tumor biopsy (n = 29) or resulting in partial (n = 104) or complete tumor resection (n = 89) prior to additional therapy (data missing, n = 3). The extent of surgical radicality was based on the impression of the surgeon.

Patients received 6 weeks of concomitant RT/TMZ therapy as primary treatment. They received TMZ 75 mg/m²/day plus RT at a dose of 60 Gy to the planning target volume in 30 fractions with 5 fractions/week delivered by a megavoltage linear accelerator. Cerebral CT was performed with 3-mm slices and fused with baseline MRI for treatment planning. Treatment planning was performed three-dimensionally using Eclipse™ treatment planning system (Varian Medical Systems, Palo Alto, CA) and volumes of interest were defined in agreement with International Commission on Radiation Units & Measurements Reports 50 and 62. The gross tumor volume (GTV) was defined as the contrast-enhanced tumor on post-contrast T1 image and/or the non-enhancing area on the T2 image on the baseline MRI scan. The clinical target volume (CTV) as defined as the GTV + 2 cm margin, except for bony structures. Meningeal structures were considered anatomic barriers to tumor spread, if appropriate clinically. If present, the surgical cavity was included. The internal target volume was identical to the CTV. No variations in size, shape or position of CTV in relation to anatomical reference structures were considered. Planning target volume was defined as the CTV + 0.5 cm margin for patient setup inconsistencies. Tolerance doses for organs at risk were as described by Emami et al. [30].

During this treatment, patients were also given antibiotic prophylaxis with 400 mg sulfamethoxazole/80 mg trimethoprim 3 times/week. In addition, a number of patients received corticosteroid therapy to relieve neurological symptoms: 165 patients (73%) received corticosteroid therapy at the initiation of RT/TMZ therapy.

Four weeks after completion of initial therapy, patients were given up to six courses of adjuvant TMZ therapy, with one course defined as TMZ for 5 days followed by 23 days without therapy. The initial course was given at a dose of 150 mg/m²/day and the remaining courses at a dose of 200 mg/m²/day. The dose was adjusted based on relevant blood tests. The number of adjuvant TMZ therapy courses given is summarized in Table 1.

As therapy for recurrent tumors, patients who maintained ECOG PS 0–2 were initially considered for secondary surgery to remove as much tumor as possible. These patients were thereafter considered for secondary therapy with TMZ 150–200 mg/m²/day if they had already received 6 courses of adjuvant TMZ and thereafter had a recurrence-free period ≥6 months. The courses consisted of 5 days TMZ therapy followed by 23 days without therapy. From 2006, regardless of adjuvant TMZ therapy and extent of recurrence-free period, the patients were additionally considered for second-line therapy with BEV 10 mg/kg every 2 weeks and irinotecan (IRI), as previously described [31]. In total, 74 patients underwent secondary surgery, 12 received second-line therapy with TMZ, and 85 received second-line therapy with BEV/IRI. Characterization of the therapy is detailed in Table 1.

At treatment initiation, a full medical history was determined and patients were examined for baseline physical and neurological status. In addition, ECOG PS [32] was determined, routine laboratory tests (including blood chemistry and urinalysis) were performed, and MRI scans were undertaken to evaluate tumor size and location.

The median duration of observation from the day patients first received therapy to the project cut-off day (22 October 2012) was 60 months (range, 23–92 months). In this period contrast and non-contrast MRI scans were repeated after 2, 5, and 6 courses of adjuvant TMZ. Patients’ neurological and clinical performance, together with corticosteroid treatment, was recorded at these time points. All patients were thereafter followed every 3 months until death or study cut-off date using the same procedures. Safety was determined using NCI-CTCAE, version 3.0, criteria [33].

Evaluations were made on formalin-fixed, paraffin-embedded tissue. Tumor tissue was classified and graded as GBM according to WHO 2007 guidelines. Diagnosis was based on conventional histological and immunohistochemical (IHC) procedures, including staining with hematoxylin and eosin, glial fibrillary acidic protein (GFAP), p53, EGFR, and MGMT. For IHC, sections were pre-treated in a microwave oven with a Tris/ethylene glycol tetra-acetic acid buffer (pH 9.0) and immunostained on a DAKO Cytomation autostainer using murine monoclonal antihuman antibodies against GFAP (Z 0334, 1:6400), p53 (M 7001, 1:800), EGFR (M 7239, 1:200) [all from DAKO, Glostrup, Denmark] and MGMT (MAB16200, 1:200, Millipore, USA). The p53, EGFR, and MGMT IHC reactions were semiquantitatively evaluated according to the number of cells stained: <10%, 10–25%, 26–50%, and >50%. Staining examples are shown in Figure 1. For statistical analysis, expression evaluated as <10% was considered negative, while ≥10% was considered positive.

Study endpoints were time to progression (TTP), OS, OS from recurrence, response at 3 and 6 months, and best response. TTP was defined as the time from the start of RT/TMZ treatment to radiological or clinical progression. OS was defined as the time from start of RT/TMZ treatment until death from any cause, while OS from recurrence was defined as the time from tumor relapse until death from any cause.

Response was evaluated 3 and 6 months after completion of RT. Response evaluation was based on MacDonald criteria [34], considering MRI measurements of contrast-enhancing tumor size and recording of the largest cross-sectional area of the tumor, patient neurological status, and corticosteroid dose. Complete response (CR) was defined as complete disappearance of measurable disease by MRI, partial response (PR) as >50% reduction of MRI contrast enhancing tumor, and progressive disease (PD) as >25% increase in area of contrast enhancement. Patients with CR or PR also had to be taking the same or decreased corticosteroid dose and have stable or improved neurological findings. Patients, by definition, had stable disease (SD) if the criteria for CR, PR, or PD were not met and no clinical progression was observed.

For each patient, responses after 3 and 6 months were compared and a ‘best response’ determined, defined as the maximum achieved response registered for the patient in the observation period.

Factors that were analyzed as potential markers of prognostic significance included: age, gender, ECOG PS, extent of resection, tumor location, tumor size, previous corticosteroid therapy, and tumor EGFR, p53, and MGMT expression. Univariate and multivariate analyses of response data were performed using logistic regression analysis modeling the probability of MacDonald response at 3 and 6 months as well as the best response. Estimates of survival probabilities for OS (primary endpoint) and TTP (secondary endpoint) were performed by the Kaplan-Meier method. Univariate and multivariate analyses of OS and TTP for the chosen explanatory variables were performed using the Cox proportional hazards regression model. Analysis of time-dependent variables was performed using the landmark method as well as the time-dependent Cox regression model.

The final model was chosen using a backwards selection procedure, the entry level was 5%. The analysis was repeated removing the least significant covariate in order to use all available data, in particular the molecular markers were only done for a subset of patients. Model assessment was done using Schoenfeld and martingale residuals. The overall concordance index (C-index) was used as a measure of discrimination [35, 36] and calculated in accordance to previously published guidelines [37]. In addition, a 5 fold cross validation was done to evaluate the model.

As shown in Table 1, best responses to first-line RT/TMZ among evaluable patients were: CR (n = 6; 2.9%); PR (n = 17; 8.1%); SD (n = 93; 44.3%); and PD (n = 94; 44.8%). Data were missing for 15 patients, who were therefore not evaluated. The effects of clinical and molecular variables on best response and response at 3 and 6 months on univariate analysis are summarized in Tables 2 and 3, respectively. The only clinical variable with a significant effect on response was patient age, for which a 10-year increase resulted in a reduction of the best response (P = .045). None of the other clinical factors examined had a statistically significant impact on patient best response or response at 3 and 6 months. EGFR, p53, and MGMT expression were examined as potential molecular markers for response (Table 3). Because of missing data, analyses were only available for subsets of patients: 145 of 199 patients presented EGFR-positive tumors; 105 of 202 patients presented p53-positive tumors; and 65 of 163 patients presented MGMT-positive tumors. There was no significant correlation between EGFR and MGMT expression and best response or response at 3 and 6 months. The odds ratio for response at 3 months among patients with p53-positive tumors was significantly (P = .043) higher as compared to those with p53-negative tumors. Although not significant, this tendency was also seen for the best response (P = .053) but not for response at 6 months (P = .21).

All 225 patients had TTP data, of whom 199 had disease progression. Median TTP was 8.0 months (95% CI, 6.7–9.0 months) with progression free survival of 61% (95% CI, 54–67%) at 6 months and 28% (95% CI, 22–34%) at 12 months (Figure 2). Increased patient age (P = .034), higher ECOG PS score (P = .046), and use of corticosteroid therapy at RT/TMZ initiation (P = .036) had a significant negative impact on TTP (Table 2). None of the other examined clinical or molecular variables had a significant impact on TTP (Tables 2 and 3).

All 225 patients had OS data, of whom 204 (90.7%) died during the observation period. Median OS was 14.3 months (95% CI, 13.0–15.8 months) with an OS rate of 27.1% (95% CI, 21–33%) at 2 years and 13.9% (95% CI, 9.5–19.0%) at 3 years (Figure 2). Median OS from tumor recurrence was 5.9 months (95% CI, 5.0–6.9 months). Increased patient age (P < .0001), higher ECOG PS score (P = .0015), and use of corticosteroid therapy at RT/TMZ initiation (P < .0001) had a significant negative impact on OS (Table 2). Increased patient age (P = .0001), higher ECOG PS score (P = .0007), and use of corticosteroid therapy at RT/TMZ initiation (P < .0001) also showed a significant negative correlation with decreased OS from disease recurrence. None of the other clinical covariates were significantly correlated with OS or OS from disease recurrence. None of the molecular markers (EGFR, p53, and MGMT) were significantly correlated with patient survival (Table 3). There was a non-significant trend for longer OS (P = .10) and OS from disease recurrence (P = .071) among patients with p53-positive tumors as compared to those with p53-negative tumors.

A total of 199 patients presented relapse. Most of these patients underwent reoperation of the tumor (n = 31; 15.6%), received BEV/IRI therapy (n = 42; 21.1%), or had a combination of both modalities (n = 43; 21.6%) for recurrent disease. In addition, 12 patients received second-line TMZ therapy as they had received 6 courses of adjuvant TMZ therapy and did not have disease recurrence for >6 months: due to the limited number of patients receiving this therapeutic option, this treatment was excluded when analyzing the effect of the different second-line treatments on survival. Compared to patients who received no second-line therapy, there was a significant OS increase in those who underwent reoperation (hazard ratio (HR) = 0.39; 95% CI, 0.25–0.60) or received BEV/IRI therapy (HR = 0.23; 95% CI, 0.15–0.34) as single treatments. When comparing OS for patients who received BEV/IRI as single second-line therapy with those who received a combination of reoperation plus second-line BEV/IRI therapy, there was no significant beneficial effect, although there was a tendency for better survival among those who received the combination (HR = 0.87; 95% CI, 0.46– 1.37). In contrast, when the reoperation-BEV/IRI combination was compared to reoperation alone, there was a significant increase in survival (HR = 0.51; 95% CI, 0.31–0.83).

Multivariable analyses of TTP and OS were done including the covariates described in Tables 2 and 3.

Multivariable analysis of the secondary endpoint, TTP, yielded a final model only including corticosteroid therapy (yes vs no, HR = 1.41 (95% CI, 1.02-1.92), p = 0.036). The p-values to include ECOG PS and age in the final model were 0.12 and 0.21 respectively. The p-values to include the remaining covariates were all >0.14.

A final model was selected for the primary endpoint OS, the following covariates were statistically significant ECOG PS (PS 1 vs 0, HR = 1.22 (95% CI, 0.89-1.68), PS 2 vs 0, HR = 2.06 (95% CI, 1.25-3.42), p = 0.015), corticosteroid therapy (yes vs no, HR = 2.06 (95% CI, 1.47-2.87), p < 0.0001) and age (per 10 years, HR = 1.31 (95% CI, 1.11-1.54), p = 0.001). P values to include the excluded covariates in the final model were >0.17 (P53, p = 0.57; MGMT, p = 0.24; EGFR, p = 0.45; tumor size, p = 0.51; operation, p = 0.84; multifocal vs. single lesion, p = 0.17). Significant interactions could not be demonstrated suggesting an additive effect of these covariates. Model assessment was found to adequate. The overall concordance index (C-index) [35‐37] for the final model was 0.82 (95% CI, 0.71–0.92), which can be interpreted as the probability of concordance between predicted and observed survival, thereby demonstrating a substantial discrimination for this model. The results of the five-fold internal cross validation supported the chosen model validating the model in the 5 test sets (C-indices > 0.80).

Based on estimated regression coefficients, patient survival chances at 6, 12, 18, and 24 months after diagnosis were calculated for various levels of each of the three covariates (Table 4). For example, the survival probability for a 40-year-old patient with ECOG PS 0 receiving no corticosteroid therapy was 97%, 86%, 73%, and 64% at 6, 12, 18 and 24 months, respectively, following diagnosis. A much lower survival probability is exemplified for a 80-year-old patient with ECOG PS 2 receiving corticosteroids: 67%, 15%, 2%, and 0% at 6, 12, 18, and 24 months, respectively from diagnosis. It can also be seen that a change in several variables at the same time can have a major negative impact on the survival probability for the individual patient, while a change in only one of the three factors had a relatively minor impact on the survival probability. This is exemplified by a survival probability of 24% at 12 months after diagnosis for a 70-year-old patient with ECOG PS 2 receiving corticosteroid therapy compared to 67% for a 50-year-old patient with ECOG PS 0 receiving corticosteroid therapy, and 82% for a 50-year-old patient with ECOG PS 0 not receiving corticosteroid therapy. It is noteworthy that a 20-year increase of patient age has a negative effect on survival probability that is similar to that seen for an increase in ECOG PS from 0 to 2 or corticosteroid therapy vs. no therapy.