The prognostic value of FET PET at radiotherapy planning in newly diagnosed glioblastoma

This retrospective study is, to date, the largest standardized published series evaluating the prognostic value of FET PET scanning postoperatively in a multivariate setting in glioblastoma patients. We found that BTV, regardless of which of three different activity cut-off thresholds was used, together with PS, MGMT protein and age were strong independent prognostic factors for OS. For poor PFS, increasing BTV and poor performance status were prognostic. The rationale behind the higher thresholds applied in previous studies [4, 11] adhere to the increased non-specific FET uptake in regional reactive astrogliosis following surgical trauma [21‐23], but could also underestimate clinically significant gliomatose infiltration (Fig. 3). The strong correlation between the three BTV thresholds indicates that for a statistical and prognostical evaluation, there is no clear advantage of one threshold over another, probably reflecting the above trade-off between sensitivity and specificity for tumor tissue. A previous smaller FET PET study similarly did not find any prognostic superiority of different thresholds [11]. For practical reasons, we chose TBR > 1.6 B for the multivariate analysis since it is already implemented in the clinic for standard RT planning at our institution, where tumor sensitivity is a priority. Previous studies indicate the use of different thresholds at different stages of tumor evaluation whether in primary diagnosis, RT planning, or in assessment of recurrence or response [15, 24, 25].

TBR_mean and TBR_max were associated with OS and PFS in the univariate analysis, but failed to predict survival in the multivariate model, similar to a previous study [4]. This is not surprising given that FET is not metabolically fixated in tissue and the kinetics differ between tumors [4]. As we acquire images only late in the time-activity curve, these different kinetics will increase the variability in TBR_mean and TBR_max and influence the prognostic value.

Former studies on prognostic factors for glioblastoma patients treated with standard therapy suggests that age, MGMT protein status, use of corticosteroids and performance status are associated with OS [2, 4, 5]. A FET PET study identified postoperative BTV as the most important independent imaging biomarker prognostic for both PSF and OS irrespective of surgical mode [4]. Our study confirms BTV, age, MGMT protein status and performance status. In our model, positivity for MGMT protein was found prognostic of poor OS. In this retrospective study, polymerase chain reaction (PCR) of MGMT promotor methylation status on the DNA level was not performed. The evaluation of MGMT protein in the tumor cells using immunohistochemistry (IHC) was the diagnostic standard in glioblastoma pathology during the observation period. Previously, we have demonstrated a good agreement between MGMT promotor methylation and MGMT protein status in a study of 151 glioblastoma patients from our institution and, therefore, we find MGMT protein a reliable surrogate marker for MGMT promotor methylation, at least in our hands, and that our data and interpretation would not change should we reanalyze the material using the PCR technique [26].

Based on the significant factors in the multivariate model, we formulated a prognostic index using Cox modeling for OS. The prognostic index can be used in clinical practice to estimate survival probability of the individual patient at 6, 12 and 18 months as a function of BTV (Fig. 1, online research figure S2). The model underlines the relative sensitivities of increasing BTV in different patient populations revealing a profound dissociation. Increasing BTV has a marked short-term negative impact on estimated survival probability in the poor prognostic group. However, for the good prognostic group, the impact of BTV is only slight on the short term, and becomes clinically significant on the 18-month estimated survival probability. The described prognostic model can potentially be a valuable tool in the clinical decision process, tailoring the optimal therapeutic strategy for the individual glioblastoma patient. Thus, it could be used to identify unfortunate patients with a life expectancy of less than 6 months, where standard treatment can be considered futile. Conversely, it could identify potential long-time survivors, where preservation of cognitive function and quality of life is given high priority and is reflected in the choice of therapy. It follows that the model could be used to reduce the risk of unnecessary side effects in both patient profiles. It should be noted that the model in its present form should be considered hypothetical and validation in an independent cohort prior to practical use is a prerequisite.

We found that larger BTV was prognostic of poor PFS as in previous studies [4, 11]. PFS is only marginally confounded by second-line therapy, and has the benefit of offering an earlier assessment and higher statistical power at the time of analysis. A prolonged PFS by itself may have a clinically significant beneficial impact on overall function and quality of life for the glioblastoma patient given the infiltrative and destructive nature of the disease. A recent large metaanalysis of 91 unique glioblastoma trials has shown that PFS and OS are strongly correlated, indicating that PFS may be an appropriate surrogate endpoint for OS [27].

One limitation was the use of the Macdonald criteria, but it has been shown that PFS evaluated by both RANO and Macdonald were associated with OS [28].

The prognostic value of postoperative BTV found in this and in other FET PET studies [4, 10, 11] repeatedly underlines the importance of maximal tumor resection [10, 29, 30]. We found a strong correlation between Gd + MRIvol and the BTV (Fig. 2) with a slope significantly lower than unity, indicating a general underestimation of tumor volumes based on Gd + MRIvol compared to BTV. In multivariate analysis, Gd + MRIvol did have independent prognostic value, providing BTV was not included, in which case it was lost. Only BTV was prognostic for PFS, while Gd + MRIvol was not in any combination emphasizing BTV as the more important prognostic factor. Extending the MRI tumor volume by including non-contrast-enhancing changes on T2-weighted sequences will not be practical on MRI 2–3 week postoperatively because of reactive changes, including oedema, ischemia, and gliosis, limiting the value in evaluating the extent of surgical resection and local tumor control.