- © 2003 by American Society of Clinical Oncology
Thallium-201 Single-Photon Emission Computed Tomography As an Early Predictor of Outcome in Recurrent Glioma
- Maaike J. Vos,
- Otto S. Hoekstra,
- Frederik Barkhof,
- Johannes Berkhof,
- Jan J. Heimans,
- Cees J. van Groeningen,
- W. Peter Vandertop,
- Ben J. Slotman and
- Tjeerd J. Postma
- From the Departments of Neurology, Nuclear Medicine, Radiology, Clinical Epidemiology and Biostatistics, Medical Oncology, Neurosurgery, and Radiotherapy, VU University Medical Center, Amsterdam, the Netherlands.
- Address reprint requests to Maaike J. Vos, MD, Department of Neurology, VU University Medical Center, PO Box 7057, 1007 MB Amsterdam, the Netherlands; e-mail: mj.vos{at}vumc.nl.
Abstract
Purpose: With limited response rates and potential toxicity of chemotherapeutic treatment in patients with recurrent glioma, reliable response assessment is essential. Currently, the assessment of treatment response in glioma patients is based on the combination of radiologic and clinical findings. However, response monitoring with computed tomography (CT) or magnetic resonance imaging (MRI) is hampered by several pitfalls and is prone to interobserver variability. The aim of this study was to establish the value of thallium-201 single-photon emission computed tomography (201Tl-SPECT) as a predictor of overall survival and response to chemotherapy in recurrent glioma, and to compare the value of 201Tl-SPECT with that of CT and MRI.
Patients and Methods: We studied patients who underwent CT or MRI and 201Tl-SPECT before chemotherapy (n = 57), and patients who also had undergone CT or MRI and 201Tl-SPECT after two courses of chemotherapy (n = 44). The value of the radiologic variables (CT-MRI tumor size, 201Tl-SPECT tumor size, and maximal tumor intensity) at baseline and at follow-up in predicting overall survival, and the percentage of patients alive and progression-free at 6 months (APF6) were examined using Cox regression and logistic regression analysis.
Results: Both at baseline and at follow-up, 201Tl-SPECT maximal tumor intensity was the strongest predictive variable and was inversely related to overall survival and APF6. In particular, progression of maximal tumor intensity after two courses of chemotherapy was a powerful predictor of poor outcome.
Conclusion: 201Tl-SPECT is superior to conventional CT-MRI in the early prediction of overall survival and response to chemotherapy in patients with recurrent glioma.
THE PROGNOSIS of patients with high-grade glioma is poor. Despite primary multimodality treatment, the disease recurs almost invariably, with few long-term survivors. Until now, therapeutic options for patients with recurrent glioma are limited. Second surgery, reirradiation, chemotherapy, immune therapy, and gene therapy are possible treatment modalities for recurrent disease. Chemotherapy may lead to a clinically meaningful response, although response percentages are not high in patients with recurrent malignant astrocytoma.1 The effect of therapy should be evaluated as early as possible to select responding patients for further treatment and to avoid chemotherapy-related toxicity in nonresponding patients. Reliable response assessment is essential, and early predictors of tumor recurrence or progression are needed.
Currently, the assessment of treatment response in glioma patients is based on the combination of radiologic and clinical findings. However, response monitoring with computed tomography (CT) or magnetic resonance imaging (MRI) in glioma patients is often difficult because of several confounding factors,2–,5 and seems to be susceptible to interobserver variability.6 Conventional CT and MRI assess morphologic abnormalities. Changes in these abnormalities not only reflect genuine therapeutic efficacy, but can also represent nonspecific inflammatory treatment reactions, radiation-induced or spontaneous tumor necrosis, or postoperative enhancement along the resection margin, and can also be affected by corticosteroid therapy.3,7–,11 Conventional CTs and MRIs visualize structural abnormalities, of which the pathologic substrate is uncertain, and of which quantitative evaluation is prone to interobserver variability.
Functional brain imaging techniques, such as single-photon emission computed tomography (SPECT) and positron emission tomography, are claimed to be more specific than structural imaging methods, and their value in the follow-up of glioma patients is the topic of several studies. Thallium-201 (201Tl) SPECT measures metabolically active tissue by cell-specific tracer uptake in malignant cells. Because 201Tl accumulates in viable tumor cells, and hardly accumulates in normal brain tissue or other nontumor tissue (such as radiation necrosis), irrespective of corticosteroid treatment, 201Tl-SPECT is a useful method to differentiate between tumor recurrence and radiation necrosis.12–,17 Recent studies suggest that 201Tl-SPECT might also be valuable in the evaluation of treatment response during chemotherapy, and in the early identification of treatment failure during chemotherapy.18–,21
The aim of this study was to establish the value of 201Tl-SPECT as predictor of overall survival (OS) and response to therapy in patients treated with paclitaxel or the combination of procarbazine, lomustine, and vincristine (PCV) chemotherapy for recurrent glioma, and to compare the value of 201Tl-SPECT with conventional radiologic techniques (CT-MRI). We performed two substudies. In study A, we compared the value of baseline (before chemotherapy) CT-MRI (tumor size) and 201Tl-SPECT (tumor size and maximal tumor intensity), in combination with clinical variables (sex, age, Karnofsky performance score [KPS], and histology), in predicting OS and the percentage of patients alive and progression free at 6 months (APF6). In study B, we compared the value of the change in these radiologic modalities after two courses of chemotherapy, in combination with clinical variables, in predicting OS and the percentage of patients APF6.
PATIENTS AND METHODS
Patients
Patients treated with chemotherapy for recurrent glioma were identified in the brain tumor database of the VU University Medical Center (Amsterdam, the Netherlands). For study A, patients were eligible if they had histologically confirmed glioma, chemotherapeutic treatment with paclitaxel or PCV for recurrent disease, adequate imaging modalities (CT or MRI before and after contrast material administration and 201Tl-SPECT) before chemotherapy, and measurable disease on CT or MRI before chemotherapy. For study B, patients were eligible if they also had undergone CT or MRI and 201Tl-SPECT after two courses of chemotherapy. Patients without proper radiologic evaluation after recent surgery for recurrent disease were excluded.
Imaging
All CT scans were performed using a standard protocol including 3- to 5-mm slices for the posterior fossa and 8- to 10-mm slices for the remainder of the brain before and after contrast material administration. All MRI scans were performed using a standard protocol including transverse T2-weighted images and T1-weighted images before and after contrast material administration.
SPECT was started 30 minutes after intravenous injection of 150 MBq 201Tl-chloride; a dual-head gamma camera (Genesys; ADAC, Milpitas, CA) was used. Projection data were acquired with a 64 × 64 matrix, 60 seconds per projection. Images were reconstructed with a Hanning filter (cutoff frequency 0.56 cycles/cm), without attenuation correction. Maximal tumor intensity was defined as the ratio of the mean tumor counts in the axial slice with the maximal tumor activity and the mean activity in the contralateral supratentorial hemisphere. In an earlier study we found that this background activity varies by 5.5% ± 2.5% (pooled coefficient of variation ± standard deviation).18 Calculation of the volume of scintigraphic hot spots requires a correction for intensity. We therefore used a histogram technique, which automatically provides the number of pixels having intensities above a threshold level.18 Threshold was set at the sum of 50% of the average background level and 50% of the average tumor intensity. This method compensates for the partial volume effect that results from the limited spatial resolution (1.5 cm) of SPECT.
Analysis
All CT and MRI scans were evaluated by one neuroradiologist, who was unaware of the patients’ clinical status and treatment. He evaluated each baseline and follow-up scan by choosing the image slice on which the lesion showed the largest diameter, and measuring tumor size on this slice. Tumor size was defined as the product of the two largest perpendicular transverse enhancing tumor diameters on the postcontrast images. The diameters were measured on the hard copies by reference to the centimeter scale printed on the film. For study A, baseline tumor size data were used. For study B, baseline and follow-up tumor size data were classified in three response categories on the basis of the percentage change from baseline. These response categories were derived from the widely used criteria developed by Macdonald et al.2 Tumor regression was defined as 50% or more reduction in cross-sectional enhancing tumor size. Tumor progression was defined as 25% or more increase in cross-sectional enhancing tumor size. All other situations were considered stable disease (classification I).
All 201Tl-SPECT scans were evaluated by one nuclear medicine physician, who was unaware of the patients’ clinical status and treatment, and who calculated both functional tumor volume and maximal intensity. 201Tl-SPECT tumor volumes were converted into areas (tumor size) assuming spherical geometry. For study A, baseline tumor size and maximal intensity data were used. For study B, baseline and follow-up tumor size and maximal intensity data were classified in three response categories on the basis of the percentage change from baseline.
Regarding the calculation of the change in maximal tumor intensity from baseline, a correction was made for the partial non–tumor-related change in intensity. Because maximal tumor intensity is defined as the ratio of the mean tumor counts in the axial slice with the maximal tumor intensity and the mean activity in the contralateral supratentorial hemisphere, the maximal intensity of normal brain tissue is 1. To calculate only the tumor-related change in intensity, we subtracted 1 from all baseline and follow-up maximal tumor intensity data. Similar to CT-MRI tumor size data, classification of 201Tl-SPECT data in the three response categories was based on the criteria developed by Macdonald et al,2 whereby tumor regression was defined as 50% or more reduction in tumor size or maximal tumor intensity, tumor progression was defined as 25% or more increase in tumor size or maximal tumor intensity, and all other situations were defined as stable disease (classification I). Thereafter, we adjusted for possible distributional differences between responses on CT-MRI and 201Tl-SPECT by matching the number of patients in the three response categories with the category sizes for CT-MRI tumor size (classification II). According to this classification, 201Tl-SPECT tumor size regression was redefined as 31% or more reduction in tumor size, tumor progression was defined as 8% or more increase in tumor size, and all other situations were defined as stable disease. Regarding 201Tl-SPECT maximal tumor intensity, tumor regression was redefined as 29% or more reduction in maximal tumor intensity, tumor progression was defined as 12% or more increase in maximal tumor intensity, and all other situations were defined as stable disease.
Definition of Outcome Variables
OS was defined as the interval (in months) from start of chemotherapy to death. APF6 was defined as the percentage of patients APF6 after start of chemotherapy. Data on OS and APF6 were obtained from patient files.
Statistical Analysis
The values of the radiologic variables (CT-MRI tumor size, 201Tl-SPECT tumor size, and maximal tumor intensity) at baseline (study A) and after two courses of chemotherapy (study B) in combination with clinical variables (sex, age, KPS, and histology) for the prediction of OS and APF6 were examined using Cox regression and logistic regression analysis. The effects of the regression coefficients were evaluated using a Wald test (two-tailed, P < .05).
RESULTS
Patient Characteristics
From January 1994 to April 2001, 67 patients had been treated with PCV or paclitaxel for recurrent glioma in the VU University Medical Center. Fifty-seven of these 67 patients fulfilled the inclusion criteria for study A, of whom 44 were assessable for study B. Patient characteristics and outcome variables of these patient groups (and also of the subgroup of 13 patients not eligible for study B) are shown in Table 1⇓. In study A (n = 57) median patient age was 49 years (range, 25 to 68 years). Tumor histologies included glioblastoma multiforme (GBM; n = 35), anaplastic astrocytoma (n = 9), (anaplastic) oligodendroglioma (n = 11), oligoastrocytoma (n = 1), and ependymoma (n = 1). Because of the heterogeneity of tumor histology in this relatively small patient group, tumor histology was dichotomized (GBM v non-GBM), resulting in 35 GBMs versus 22 non-GBMs. All patients had previously undergone cranial surgery and radiation therapy. Two of 57 patients had been reirradiated before the initiation of chemotherapy. Although these patients were potentially at higher risk of treatment-related toxicity such as radiation necrosis, they were included in the analyses because of positive tumor-to-nontumor ratios on 201Tl-SPECT. Three patients had received previous chemotherapy (one suramin, one temozolomide, and one topotecan), all in the context of a clinical trial. The median time to tumor recurrence (from initial diagnosis) was 10 months, the median time to tumor progression (from start of chemotherapy) was 4 months, the median OS was 9 months, and APF6 was 32%. In the subgroup of patients not assessable for study B (n = 13), no follow-up imaging was performed because of rapid tumor progression in 11 patients and early death during the second course of PCV in two patients. At baseline, this subgroup had worse clinical status (higher median age, lower median KPS, and shorter time to tumor recurrence) and worse outcome variables than the patients evaluated in study B (n = 44; median time to tumor progression, 1 v 5 months; median OS, 3 v 10 months; and APF6, 0% v 41%, respectively).
Patient Characteristics
Descriptive Statistics
Table 2⇓ shows descriptive data on all CT-MRI and 201Tl-SPECT tumor measurements. For study A, 57 CT-MRI (56% CT and 44% MRI) and 57 201Tl-SPECT scans were evaluated. For study B, 88 CT-MRI (59% CT and 41% MRI) and 88 201Tl-SPECT scans were evaluated. The two patients without tumor activity on 201Tl-SPECT before chemotherapy were included in the analyses; recurrent tumor was confirmed histopathologically in one patient and strongly suggestive in the other patient (on the basis of clinical deterioration accompanied by progressive enhancement on MRI).
Descriptive Data on All Tumor Measurements
Study A: Baseline Analysis
The effects of the different radiologic and clinical variables at baseline on the outcome variables were examined in multivariate analyses (results not tabulated). Tumor size on CT-MRI had little prognostic value and was not significantly related to OS and APF6 (P = .10 and P = .31, respectively). Tumor size on 201Tl-SPECT correlated negatively with OS (P = .02; the larger the tumor, the shorter the OS), whereas maximal tumor intensity on 201Tl-SPECT correlated negatively with both OS and APF6 (P < .01 and P < .03, respectively; the higher maximal tumor intensity, the shorter the OS and the smaller the APF6).
Of all clinical variables at baseline, KPS had the strongest prognostic value for OS, in combination with any of the radiologic variables (P < .01) In addition, histology had the strongest prognostic value for APF6, in combination with any of the radiologic variables (P = .03 in combination with CT-MRI tumor size, P = .04 in combination with 201Tl-SPECT tumor size, and P = .05 in combination with 201Tl-SPECT maximal tumor intensity).
Study B: Follow-Up Analysis
Regarding the follow-up after two courses of chemotherapy, we evaluated the degree of agreement on response classification of the different radiologic variables (201Tl-SPECT data on the basis of classification II). In 25 of 44 patients (57%) the response classification on the basis of CT-MRI tumor size corresponded with the response classification on the basis of 201Tl-SPECT tumor size. For the combination of CT-MRI tumor size and 201Tl-SPECT maximal tumor intensity, agreement on response classification was demonstrated in 22 of 44 patients (50%); for 201Tl-SPECT tumor size and 201Tl-SPECT maximal tumor intensity, agreement on response classification was demonstrated in 28 of 44 patients (64%).
Table 3⇓ shows the results of multivariate regression analyses in which we examined the effects of the different radiologic variables at follow-up on the outcome variables. Clinical variables were also included; however, they were not tabulated. Regarding 201Tl-SPECT tumor size and maximal tumor intensity, separate analyses were performed for response classifications I and II. The analysis including CT-MRI tumor size data demonstrated that after two courses of chemotherapy, patients with tumor progression had a shorter median OS (P = .05) and less chance of being APF6 (P = .01) than patients with stable disease. The analyses including 201Tl-SPECT tumor size data demonstrated that after two courses of chemotherapy, patients with tumor regression had a longer median OS (P = .02 for classifications I and II) and a better chance of being APF6 (P < .17 for classification I and P = .03 for classification II) than patients with stable disease. Patients with tumor progression after two courses of chemotherapy did not have a substantially different prognosis on OS and APF6 than patients with stable disease. The analyses including 201Tl-SPECT maximal tumor intensity data demonstrated that after two courses of chemotherapy, patients with tumor regression had a longer median OS (18 months, P = .05 for classifications I and II) than patients with stable disease (10 months). In addition, patients with tumor progression had a shorter median OS (7 months, P < .01 for classification I; 8 months, P < .01 for classification II) than patients with stable disease, with corresponding odds ratios (ORs) of 8.7 and 4.4 for classifications I and II, respectively. In addition, for both 201Tl-SPECT tumor size and maximal tumor intensity, for classification I, all patients with tumor regression and none of the patients with tumor progression after two courses of chemotherapy were APF6, resulting in ORs of 0 and infinity, respectively.
Multivariate Regression Analyses of OS and APF6 on the Radiologic Variables at Follow-Up (n = 44)
Of all clinical variables, KPS had the strongest prognostic value for OS in combination with any of the radiologic variables (P < .01), whereas histology had the strongest prognostic value for APF6 in combination with CT-MRI tumor size and 201Tl-SPECT tumor size (for both, P = .03). However, in addition to 201Tl-SPECT maximal tumor intensity, none of the baseline clinical variables had significant prognostic value for APF6 (P = .08 to .73).
Kaplan-Meier estimates of the survival curves, classified in the three response categories (according to classification II for 201Tl-SPECT data) are shown in Figure 1⇓. The Kaplan-Meier curves are in line with the outcomes of the Cox regression and indicate that 201Tl-SPECT maximal tumor intensity (P < .01; Fig 1C⇓) leads to response categories that differ more strongly with regard to OS than the response categories constructed with CT-MRI tumor size (P = .53; Fig 1A⇓) and 201Tl-SPECT tumor size (P = .12; Fig 1B⇓).
Kaplan-Meier survival curves of 44 patients (study B), classified in three response categories (tumor regression [TR], stable disease [SD], and tumor progression [TP]), on the basis of (A) computed tomography or magnetic resonance imaging tumor size, (B) thallium-201 single-photon emission computed tomography (201Tl-SPECT) tumor size, and (C) 201Tl-SPECT maximal tumor intensity. Differences among the survival curves were assessed by the log-rank test.
For each combination of radiologic variables, we examined the effect of either variable on OS and APF6 (Table 4⇓). In these analyses, 201Tl-SPECT data were categorized according to classification II, and clinical variables were also included; however, these data were not tabulated. The analysis with CT-MRI and 201Tl-SPECT tumor size data in one multivariate model demonstrated a significant effect of tumor regression on OS (P = .03) for 201Tl-SPECT, although no effect of either tumor regression or progression on OS for CT-MRI was demonstrated. Multivariate analysis with CT-MRI tumor size and 201Tl-SPECT maximal tumor intensity demonstrated significant effects of tumor regression and progression on OS (P = .04 and P < .01, respectively) and of tumor progression on APF6 (P = .04) for 201Tl-SPECT. Again, there was no effect of either tumor regression or progression on OS and APF6 for CT-MRI. This indicates that CT-MRI tumor size has no prognostic value in addition to 201Tl-SPECT tumor size and 201Tl-SPECT maximal tumor intensity. Lastly, we compared 201Tl-SPECT tumor size and maximal tumor intensity in one multivariate model and found significant effects of tumor progression on OS and APF6 (P < .01 and P = .03, respectively) only for 201Tl-SPECT maximal tumor intensity, with corresponding ORs of 5.9 and 20.0, respectively.
Multivariate Regression Analyses of OS and APF6 on Any Combination of Radiologic Variables (n = 44)
DISCUSSION
Our results partly correspond to the results of other studies that investigated the value of 201Tl-SPECT in the evaluation of treatment response during chemotherapy.18–,21 However, in our study the number of patients was substantially larger, and both 201Tl-SPECT tumor size and maximal tumor intensity were evaluated as response variables.
Regarding the prognostic value of clinical variables, the results of our study are largely in concordance with several other brain tumor studies.1,22–,25 In our study, KPS was strongly related to OS, whereas histology (on the basis of the dichotomy GBM v non-GBM) was strongly related to APF6 only in the baseline analysis. In the follow-up analysis, histology gained statistical significance for APF6 only in combination with CT-MRI tumor size and 201Tl-SPECT tumor size. In addition to 201Tl-SPECT maximal tumor intensity, the strength of histology as a predictor of APF6 decreased and lost statistical significance. In line with the proven chemosensitivity of oligodendrogliomas,26,27 we also evaluated the prognostic value of histology on the basis of the dichotomy oligodendroglial versus nonoligodendroglial tumors on outcome. In the follow-up analysis, 10 of 44 patients had an oligodendroglial tumor, and this subset of patients was found to exhibit a favorable response to chemotherapy (median OS, 19 months; APF6, 90%). In univariate analyses, histology on the basis of this dichotomy was moderately related to OS (P = .09) and strongly related to APF6 (P < .01), with oligodendroglial tumor patients having a better prognosis. However, in multivariate analysis with 201Tl-SPECT maximal tumor intensity, the effect of histology on OS disappeared (P = .43), and the effect of histology on APF6 decreased and lost statistical significance (P = .09). In summary, histology on the basis of the dichotomy GBM versus non-GBM, and oligodendroglial versus nonoligodendroglial tumor, was related to outcome, although these effects were mediated by the addition of 201Tl-SPECT maximal tumor intensity. These findings emphasize the strength of 201Tl-SPECT as a predictor of outcome in the early follow-up during chemotherapy.
With respect to the prognostic value of tumor size on CT-MRI, our results do not completely correspond with other studies. A recent study by Simon et al28 demonstrated tumor size to be a significant predictor of survival in patients with recurrent GBM treated with brachytherapy. Tumor size on MRI has also been described to be related to the response to chemotherapy (although not necessarily to OS) in another study of patients with recurrent malignant glioma.29 In our study, tumor size on CT or MRI before chemotherapy was not significantly related to OS or to APF6. Conversely, tumor size on 201Tl-SPECT before chemotherapy did correlate with OS. Regarding the powerful prognostic value of 201Tl-SPECT in our study, it might be hypothesized that findings of functional imaging such as SPECT are more related to the biologic behavior of a (recurrent) glioma than are findings by conventional CT-MRI. This raises the question whether conventional CT or MRI has additional value in the assessment of response to chemotherapy in recurrent glioma patients, if SPECT is available.
A limitation of our study is that the collection of imaging data was not performed prospectively. Therefore, several CT and MRI scans did not fulfill the inclusion criteria (eg, only postcontrast images were available for certain patients) and had to be excluded from analysis. Furthermore, although our data suggest that 201Tl-SPECT is superior to conventional radiologic techniques, we did not compare the two modalities in terms of cost effectiveness. Finally, in clinical practice 201Tl-SPECT is not widely available, which precludes the standard use of this modality in the evaluation of response to chemotherapy and other therapies.
In conclusion, in this study 201Tl-SPECT was found to be superior to conventional CT or MRI in the prediction of OS and response to chemotherapy in patients with recurrent glioma. In particular, progression of 201Tl-SPECT maximal tumor intensity in the early follow-up during chemotherapy was a powerful predictor of poor outcome. Currently, we are performing additional analyses to determine the best cutoff values for 201Tl-SPECT in our patient population. On the basis of our results so far, we conclude that 201Tl-SPECT is a valuable tool in the prediction of survival and response to chemotherapy in patients with recurrent glioma.
AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST
The authors indicated no potential conflicts of interest.
- Received December 30, 2002.
- Accepted April 23, 2003.
