Reliable imaging biomarkers are necessary to efficiently conduct clinical trials that compare the efficacy of new therapies. Nearly 75% of oncologic clinical trials rely on surrogate imaging endpoints rather than patient survival,11
with most clinical trials for malignant glioma treatment using modified Macdonald criteria to determine treatment response.8
A major limitation of the Macdonald criteria, which were developed 2 decades ago, is the reliance on changes in size of the enhancing tumor. PsP may be indistinguishable from EP using these response criteria. Recognizing this difficulty in diagnosis, many phase II trials for recurrent malignant gliomas exclude patients with worsening enhancing lesions within 3 months after RT.4,5,12
Although advanced imaging techniques such as MR perfusion, MR spectroscopy, and PET may have increased accuracy and sensitivity, these technologies are not as well-studied or ubiquitously available as conventional MRI. As a result, even the most recent attempts to establish new criteria by the Response Assessment by Neuro-Oncology13
group rely on conventional MRI characteristics (contrast enhancement and T2/FLAIR signal abnormality). Determining conventional MRI signs that would best determine PsP would assist the management of these patients and potentially impact clinical decision-making processes.
In this study, we report the largest cohort of patients systematically reviewed for MRI findings that could differentiate PsP from EP. We found that subependymal enhancement predicts the development of EP rather than PsP in worsening enhancing lesions that occur soon after completion of combination chemoradiation therapy. Although subependymal enhancement has high specificity for EP (93.3%), its low sensitivity (38.1%) and low negative predictive value (41.8%) suggest that it may have only limited utility in the majority of patients with suspected PsP.
Subependymal spread of tumor is a known pattern of glioma failure, although it is less common than local progression.14,15
Infiltration of the margins of the ventricles may occur by direct spread of tumor cells in the subependymal space or by deposits transferred by the CSF.16
One study of 51 multifocal gliomas (of which 31 were glioblastoma) reported subependymal spread to be the second most common route for disseminated disease at 24%.17
Less frequent rates of subependymal or spinal spread have been described by other groups,14
ranging from 0% to 14%.
Conventional MRI signs to distinguish radiation necrosis from tumor recurrence have been investigated in patients with malignant gliomas. One study of 27 patients did not find individual signs to be useful predictors for tumor recurrence, although combining 2 signs with involvement of the corpus callosum and multiple enhancing lesions was useful (p
= 0.02), as were combining 3 signs with involvement of the corpus callosum, multiple enhancing lesions, and crossing of the midline (p
= 0.04) or subependymal spread (p
The lack of significance for individual signs such as subependymal spread (p
in contrast to the findings in our study (p
= 0.001), may reflect the smaller number of patients or differing central distributions of lesions in that series. In addition, those patients all had new enhancing lesions that occurred more than 6 months after proton beam RT, which were more consistent with radiation necrosis than with PsP. The 2 entities are similar but not synonymous, with PsP showing earlier onset after completion of RT at 1–3 months that reflects an intermediate stage between subacute radiation reaction and later radiation necrosis at 6–18 months or more.2
Treatment-related necrosis can also occur in the periventricular region and mimic subependymal spread of tumor.19
This is thought to reflect the relatively poor vascularity of the periventricular region, which is supplied by long medullary arteries without collateral supply that are vulnerable to radiation-induced vasculopathy.19
The low incidence of subependymal enhancement in PsP (6.7%) found in our study, however, suggests that this complication may be less frequent than direct spread to the subependymal region by centrally located tumors. Treatment-related necrosis becomes more common with increasing total doses, high fraction doses, hyperfractionation, and concurrent chemotherapy.2,20
The majority of our patients were treated with standard RT plans, with relatively equal small proportions of the EP and PsP groups instead receiving abbreviated RT plans that are acceptable alternatives.9
Although we did not detect a correlation with the RT dose, further examination of the potential relationship between RT dose and fields with subependymal enhancement may be useful.
Survival has been reported as longer in patients with methylated MGMT promoter status who receive temozolomide.6,21
A study attempting to correlate MGMT status and PsP found that methylated MGMT promoter status is a strong predictor of PsP, occurring in 21/23 (91.3%) of methylated vs 11/27 (40.7%) of unmethylated patients (p
Less than a quarter (23.7%) of our patients had known MGMT promoter status, with methylated MGMT promoter detected in 5 patients with EP, including 2 patients with pathologic confirmation of their EP status, and none of the patients with PsP. The low number of patients precludes further analysis, although this finding does highlight the unreliability of using MGMT status to predict PsP or EP in an individual patient. We are prospectively collecting molecular and genetic data in contemporary patients, and plan a separate project to specifically examine the potential relationship between MRI and MGMT status.
One potential limitation of our study relates to the lack of a widely accepted definition of PsP. Specific clinical, imaging, and pathologic criteria for the diagnosis of PsP were established for this study, including patients who did not require additional treatment for a minimum of 6 months. Although this may have underestimated the true incidence of PsP, the primary intent of the study was to determine clinically useful conventional MRI signs that could guide treatment decisions by confidently identifying the patients who would not have required a change in treatment. We recognize that this definition also potentially biased the survival analysis, since the condition of a 6-month interval was not imposed upon the EP group.
Another limitation is that most but not all patients received standard RT and TMZ chemotherapy as established by Stupp et al.1
Since PsP or early treatment-related necrosis has been described with other RT and chemotherapy regimens,19,22
the heterogeneity of TMZ vs non-TMZ treatment and differing adjuvant TMZ schedules should have little effect on the analysis. In addition, the proportion of patients determined to have PsP (32.3%) is similar to previously published rates.2–5
The majority (75.2%) of patients in this study received adjuvant TMZ according to the standard 5/28 day cycle. We did not detect a correlation between PsP and TMZ schedule (p
= 0.27). It is possible that a more effective adjuvant dosing schedule could cause EP to remit and mimic PsP, although differential rates of PsP have not been described with dose-dense vs metronomic TMZ treatment.10
Conventional MRI signs have limited utility in the diagnosis of PsP in patients with recently treated glioblastomas and worsening enhancing lesions. The distinction is important for making treatment decisions, patient counseling, and establishing prognosis. We did not find a sign with a high negative predictive value for PsP, which would have provided the most useful information for treating clinicians. When present, direct subependymal spread of the enhancing lesion is a useful MRI marker in identifying EP rather than PsP. Additional research into advanced imaging modalities or biomarkers such as MRI perfusion, diffusion tensor, spectroscopy, and PET/CT is necessary.