With the knowledge of the significant interobserver variation, the question of how this affects the conduct and interpretation of prospective randomized clinical trials is highly relevant and all too often ignored. One of the first attempts to explore the role of pathology review within a prospective clinical trial on high grade glioma was done within an RTOG study [28
]. This study noted a high degree of concordance in locally diagnosed GBM cases (96%), but in only 66% of astrocytomas with anaplastic foci (AAF). Locally diagnosed AAF that were reclassified as GBM had a GBM like survival, whereas locally diagnosed GBM cases that were reclassified as AAF had an in-between survival. This is noteworthy, as apparently the initial local diagnosis of a GBM did make a difference in outcome. It suggests that the initial diagnosis of AAF picked out a subgroup that had a better outcome, despite the central review diagnosis of GBM. The authors showed that the misclassification they observed would seriously affect the power of a clinical study on AAF, assuming that the investigational treatment does improve outcome in AAF but not in GBM. Some confirmation of that assumption comes from an EORTC study on AA [14
]. That study failed to show an improvement of adjuvant chemotherapy, but similar to the RTOG study a high level of histological disconcordance was noted. At central pathology review, over half (53%) of the locally diagnosed AA cases could not be confirmed. Of note, a second reviewer disagreed frequently with the first reviewer but a sensitivity analysis of the study based on confirmed AA by either of the two central pathologists showed an improved outcome after adjuvant chemotherapy compared to treatment with radiotherapy alone (EORTC, data on file). Similarly, a large randomized study on grade II glioma confirmed the presence of a low grade glioma in 74% of patients of whom material was available for review [33
]. Since today’s trials are focused on specific tumor types and grades, differences between pathologists have to be addressed in the design and interpretation of these studies in particular on grade II and grade III tumors.
Two recent trials have studied the impact of (neo)adjuvant PCV chemotherapy in anaplastic oligodendroglioma (AOD) and AOA. The rationale of both studies was the observed high response rates to PCV chemotherapy of recurrent anaplastic oligodendroglial tumors, which was at the time of study start still understood as related to oligodendroglial morphology [4
]. This was subsequently associated with combined 1p/19q loss [6
]. For both trials, both pure and mixed anaplastic oligodendroglial tumors were eligible, and they required either two or three anaplastic features to be present as part of the inclusion criteria [5
]. Mixed tumors were allowed in both trials provided 25% of oligodendroglial elements were present. One trial—conducted by the RTOG in North America—required central confirmation of the diagnosis prior to study inclusion, the other—European—EORTC trial had central pathology review after the inclusion. Both trials showed that (neo)adjuvant PCV improves progression-free survival but not overall survival. In both trials, results of pathology review of included patients have been published.
In the EORTC trial on 368 patients, the diagnosis of a grade III oligodendroglial tumor was confirmed in 257 patients out of 348 with material available for review (74%) [15
]. The EORTC trial has been used to study the interobserver variation of the pathology diagnosis. Using a review panel, 114 cases were classified by 9 independent pathologists [17
]. Review diagnoses ranged from low grade astrocytoma to GBM. The panel reached a consensus on 52% of AOD and in only 8% of the AOA, and survival was clearly associated with the diagnosis at review. Molecular analysis in the dataset of the EORTC study demonstrated combined 1p/19q loss in only 26% of the cases with sufficient material available for analysis, and in considerable percentages of patients molecular lesions usually associated with GBM were found (e.g. EGFR amplification in 51 out of 233 samples) [15
]. Clearly, this trial suffered from a more heterogeneous patient population than intended, with many patients harboring glioblastoma-like tumors and only a minority of the patients having the chemotherapy sensitivity-associated 1p/19q co-deletion.
A different route was taken in the RTOG trial 94-02. Here, pathology review was conducted prior to study entry, although unfortunately it is unclear how many patients were rejected from the trial because of discordant pathology diagnoses at central review. The percentage of patients with tumor with 1p/19q loss in this study was considerably higher as compared to the EORTC study (93 of 201 cases; 46%). In this study, a second pathology review study has been conducted after the first central review to confirm the patient’s eligibility resulting in an enriched for oligodendroglioma patient population. For that study, all available samples were reviewed by two new and independent reviewers and in case of disagreement by a third reviewer [11
]. The authors scored the samples for the presence or absence of “classical for oligodendroglioma (CFO), including cellular monomorphism, round/regular nuclei, presence of nodules, microcalcifications, microcysts, and chickenwire vasculature”. CFO tumors were highly associated with 1p/19q loss (present in 80%; as opposed to only 13% of non-CFO). The authors concluded that central pathology review is an important component to establish uniformity of entry criteria in future trials. Moreover, in CFO, a trend toward increased survival after neoadjuvant PCV was present. Nonetheless, the κ
score between the two expert pathologists was only 0.55, indicative of a moderate amount of interobserver agreement. One of the experts did not classify 25 as CFO out of 115 cases that were considered CFO at the end of the review process. Moreover, despite the central pathology review at study entry confirming the oligodendroglial nature of the tumor (and thus all cases were diagnosed as oligodendroglioma by two pathologists), the reviewers that took part in the second review still diagnosed some tumors as AA or GBM.
A German study on anaplastic glioma (regardless of lineage) compared initial treatment with chemotherapy versus initial treatment with radiotherapy [38
]. Similar to the RTOG study, central confirmation of the histological diagnosis was required prior to study entry. The investigators noted a high concordance between the local and the central diagnosis (κ
= 0.7). Remarkably, they also noted a similar survival in AOA and AOD, in which sense this trial is unique: virtually all studies on this topic have shown a better survival in AOD as compared to AOA [11
]. This suggests that despite the good concordance the results may differ from those obtained in other countries. Some indications what may have played a role here are only touched upon by the authors in the discussion part of the manuscript. Here, it is mentioned that a rather restrictive central histologic AOA classification was used: cases of astrocytic tumors with just minute or ambiguous oligodendroglial differentiation features did not qualify for the diagnosis of AOA. It thus appears that the criteria for mixed oligoastrocytomas were narrowed. In fact, older studies have already shown the interobserver variation in the diagnosis of mixed oligoastrocytoma [18
]. If the border between AOA and AA is shifted toward a more oligodendroglial phenotype, stage migration (also known as the Will Rogers phenomenon) occurs and it is no longer a surprise the survival of AOA becomes similar to that of AOD [10
]. Indeed, the blunt statement in the abstract that AOD and AOA share the same better prognosis than AA is in fact illustrative of the different set of criteria used and indicative of the role of interobserver variation and of stage migration.