In this study, the 3 main risk classifiers described in the last decade (ESMO-ESGO-ESTRO 2016, ProMisE and ESGO-ESTRO-ESP 2020) were evaluated in a large early-stage EC cohort. The results showed that all of these classifiers differentiate RFS between high and low-risk groups, but there was an overlap between the intermediate and high-intermediate risk groups. Similar findings have been observed in other studies. For example, regarding the 2016 Classifier, 2 retrospectives cohorts reported no differences between the intermediate and high-intermediate group, one of them with overlapping K-M OS curves [21, 22]. In terms of the ProMisE Classifier, there are other publications that also showed no significant differences between the 2 intermediate molecular subtypes, although it performed well on the 2 extreme groups: the POLE group, with a very low incidence of relapses, and the p53abn group, with a high risk of recurrence [23, 24]. In our study, we observed similar results, with only 1 relapse in the POLE group.
The recently published 2020 Classifier has incorporated the molecular variables of the ProMisE classification into the prognostic stratification carried out in the 2016 Classifier, with the aim of improving its accuracy and thus making better therapeutic recommendations. In this new classification, stage I-II POLE mutated tumours are included in the low-risk group, for which adjuvant treatment is not recommended, whereas most of the p53abn tumours (except those without myometrial invasion) have been incorporated into the high-risk group, for which adjuvant chemotherapy is strongly recommended.
In this study, we have provided one of the first evaluations of this new risk classification in a cohort of patients and, to our knowledge, the first comparison of the 3 classifiers focused on early-stage EC. Two recent publications have evaluated the 2020 Classifier in 2 large patient cohorts, including those with advanced disease [25, 26]. Similarly to our results, Ortoft et al described fewer patients allocated to the high risk group using the 2020 Classifier and reported a poorer RFS for this group than that achieved with the 2016 Classifier [25]. These findings suggest that the 2020 Classifier achieves a better redistribution of the 4 risk groups that impact the 5-year survival rates. However, in terms of c-index values, we found only a discrete improvement over the 2016 Classifier. Furthermore, in our experience this classifier is still not good enough to separate the 2 intermediate groups, and following this classification, different adjuvant treatments would be recommended to patients with a similar prognosis (intermediate and high-intermediate groups). In the same way, Imboden et al found significant differences in RFS using the 2020 Classifier, but with an overlap of K-M curves of both intermediate-risk groups [26]. These results reaffirm the unmet need for an accurate stratification system and motivate us to explore the potential of other biomarkers that could improve the current options.
To improve the precision of the 2020 Classifier, we focused on the molecular biomarkers previously explored in EC, with potential prognostic value but not yet included in the main risk classifiers. We first evaluated their association with prognosis in our entire cohort. Among them, only ER showed a significant correlation with RFS, and ER, PR and ECAD with OS. These results are in agreement with previous publications [27, 28]. There are several reports on HER2 amplification, specifically in non-endometrioid histologies and a subset of high-grade endometrioid tumours. We had almost no HER2 overexpression, so no correlations with the prognosis could be established [29]. Loss of ARID1A has been linked to shorter progression-free survival in EC, and loss of PTEN might be a good prognostic factor [30, 31]. Our results are similar in terms of the positive proportion of cases for both biomarkers, but we did not find any statistical significance related to survival.
Among the remaining analysed markers, probably the most intriguing results concern L1CAM, which has frequently been associated with distant recurrence and OS. We have used a previously established cutoff for IHC to achieve the best correlation with prognosis [32]. Our results are similar regarding positivity rates to those published for the PORTEC-1 trial samples, but do not reach significance, probably because of the lower positivity of the marker and the smaller size of our cohort [33]. The other biomarker frequently associated with prognosis is CTNNB1. In our cohort, it showed significance only when intermediate risk groups were merged, and for this reason it was subsequently considered for their inclusion in the risk classifier.
The impact of the CTNNB1 mutation and other biomarkers (like POLE, MMRd, p53, L1CAM, or LVSI) prompted the design of the PORTEC-4 trial. This phase III study, including patients with high-intermediate risk EC, randomises patients between a standard arm with adjuvant vaginal brachytherapy and an experimental arm with an adjuvant radiation therapy tailored by a molecular-integrated risk profile. In this trial, patients with p53wt/NSMP and no mutation in CTNNB1 are considered to be in the same low-risk group as those with the POLE mutation [34]. However, in our study, patients initially classified in the intermediate or high-intermediate groups with no mutation in CTNNB1 have a poorer prognosis than those of the low-risk group (which included patients with the POLE mutation).
The CTNNB1 mutation leads to the over-activation of beta catenin, which results in the aberrant signalling of the Wnt pathway, contributing to tumour progression [35]. The poorer prognosis associated with the CTNNB1 mutation in exon 3 has been shown in other studies, mainly in grade 1–2 endometrioid or NSMP cohorts [36, 37], suggesting that this mutation is more likely to be functional, and not a passenger event [38]. The ESGO-ESTRO-ESP 2020 guidelines mention that the CTNNB1 mutation might be potentially useful in the group of low-grade p53wt/NSMP EC, but they did not include it in the risk stratification proposal. In our study, the CTNNB1 status was significantly associated with RFS in the intermediate and high-intermediate risk groups.
Further, the CTNNB1 mutational analysis over both intermediate groups could reallocate some patients into the high-risk group (those with the CTNNB1 mutation), while the remaining patients would be considered to be within the intermediate risk group. Moreover, by including the CTNNB1 status in the 2020 Classifier, we simplified the 4-group classification into 3 groups. Based on this proposal, adjuvant treatment recommendations could be made for each novel group; for example, patients allocated as intermediate or high-intermediate by the 2020 Classifier with the CTNNB1 mutation can be considered for adjuvant chemotherapy.
The main limitation of our study is related to its retrospective design and the absence of a validation cohort. Second, the study is based on TMA and not on whole tissue sections, which might not completely reflect the heterogeneity of some tumours. On the other hand, as strengths, the large number of patients with a long follow-up, and the high homogeneity of the series should be highlighted, given it encompasses only early stages (FIGO I-II). Furthermore, it is the first study to evaluate and compare the 3 most important risk classifiers in EC, including the recent ESGO-ESTRO-ESP Classification, focused on early-stage disease.