This study shows that the “Endometrial Intraepithelial Neoplasia (EIN)” system is superior to the WHO hyperplasia scheme in discriminating lesions with highest risk for conversion to cancer. Furthemore, a large group of women who initially have been given the diagnosis of “hyperplasia” but are later found not to have EIN, have near negligible risk of cancer development. Prospective discovery of carcinoma may be attributed to one of two different mechanisms. Those cancers seen soon after the index biopsy could be considered concurrent instances present but missed at the time of the index biopsy. In contrast, carcinomas seen at longer follow-up intervals are much more likely to represent time dependent precancer to cancer progression events. Both short and long term cancer outcomes are well predicted by an EIN diagnosis (D-Score<1).
A key feature of premalignant lesions is elevated risk for developing cancer if left untreated. This perspective is an essential and valid standard by which clinical utility of an existing precancer diagnostic scheme should be measured, but is inefficient for discovery of new diagnostic criteria. Prediction of low frequency clinical outcomes requires very large sample sizes in the learning or “criteria discovery” case set, and predictive performance degrades substantially upon extrapolation to new (test or “challenge”) cases if the initial classification criteria are incompletely or poorly defined. In this regard, objective morphometry has the advantage of high reproducibility and the ability to decompose component variables independently. Alternatively, molecular genetic assessment of seminal precancer characteristics such as monoclonal growth, acquisition of mutations which offset lesions from the normal background, and documentation of lineage continuity with subsequent carcinoma may be informative in definitive identification of individual examples of precancers15, 16
. In the case of EIN, these molecular and morphometric approaches have independently identified the same class of premalignant lesions14
, effectively cross-validating each as complementary discovery platforms. The current study goes one step further in documenting a level of clinical predictive value which exceeds that of current practice.
A major problem with the WHO94 scheme is disappointing reproducibility17, 18
. Because each of its four subcategories predicts only 1% to a maximum of 20% of cases that actually develop future cancer, the prospect of this scenario and the subsequent risk of overtreatment is at least 80% (100-20%) to 99% (100 - 1%) of patients. This current study documents that the EIN classification scheme more accurately predicts the development of future cancer than WHO94. Both classification systems are predictive of endometrioid endometrial cancer, the serous type having a different pathway exclusive of EIN19
Of immediate practical significance, the EIN system more accurately identifies cases with benign changes, some of which are regarded as high risk (atypical hyperplasia) under WHO94. The question is why. At the heart of both schemes are diagnostic criteria. WHO94, as initially described, emphasized architectural abnormality with sub-stratification by presence or absence of cytologic (principally nuclear changes) atypia. This evolved over time to where cytology became the principal feature for discrimination of cancer risk20
. WHO atypical cytology is defined in stereotypical terms as loss of nuclear polarity, nuclei with prominent nucleoli and altered chromatin7
. In the EIN system gland architecture is the most important of the 3 independent D-Score variables4
, with cytology being less predictive as an independent variable. Morphometric EIN nuclear pleomorphism is exemplified by the variable of shortest nuclear diameter, a feature found more predictive of cancer association than other nuclear characteristics including textural chromatin features21
. Subjective review of EIN cases identified by morphometry has reaffirmed the importance of cytologic change, but the nature and appearance of these alterations are not absolute amongst all cases. Rather, cytologic changes are best recognized by comparison of lesional to background cytology in the same patient. Presently, neither the WHO94 nor the EIN scheme find nuclear-cytoplasmic ratios significant22
and to date, neither identify textural changes exclusive to the cytoplasm as reliable prognostically.
The second key feature the WHO94 scheme evaluates is architectural change, essentially in the form of glandular outpouchings (“budding”), infoldings, and any other abnormal glandular configuration (“back-to-back positioning”). This feature, while crisply described in textbooks, is subjective in practice and open to considerable interpretation and variation. The EIN system also scores architectural change, but morphometrically with an easily quantifiable proxy in the linear length of basement membrane per gland (OSD=outer glandular density)23
. Any deviation from a perfect circle shows as an increased total length of the basement membrane. The OSD is the 2nd
most important component of the D-score10
The most important contributor to the D-score, the VPS (volume percentage stroma), does not formally exist as a WHO94 criterion. Nonetheless, pathologists often intuitively incorporate this feature (usually in the inverse as “glandular crowding”) when diagnosing complex atypical hyperplasia because the glandular component is so dominant. The importance of the current findings is that the VPS quantifies and recognizes the stromal-glandular ratio, which pathologists use on a daily basis. A major reason for the lack of fixed WHO94 to EIN concordance is that some EIN variables such as lesion size and objective assessment of glandular crowding (“volume percentage stroma”) do not appear as part of the WHO94 scheme.
The D-Score, using objective measurements, is highly reproducible, as shown by high interobserver correlation coefficients (R≥0.91) among 17 different well-trained technicians performing these measurements over the 22 year period4, 10, 14
. In addition to excellent reproducibility of measurements, the correlations between the presence and absence of disease and later outcome were also high. The data from this study has also been important to further refine the decision thresholds for the D-Score. Previous studies were inconclusive concerning prognostic significance of a D-Score between 0 and +1 due to relatively few cases, a problem resolved in the current greatly expanded study. Hyperplasias with scores between 0 and +1 behaved more like those with D-Score <0, and are thus incorporated into what is considered as EIN.
A molecular-genetic quantitative foundation, partially based on PTEN tumor suppressor gene inactivation15, 24
underlies the EIN classification, distinguishing it from the WHO94, which was developed based on work from the 1980s devoid of molecular input. EIN lesions are monoclonal outgrowths of transformed cells, which begin as a discrete focus and over time spread to involve larger fractions of the endometrial compartment14, 15, 24
. The concept of a focal origin is highly relevant when interpreting the endometrial samples and emphasizes the need to select representative areas of the diagnostic lesion. Large scale architectural features of EIN lesions, particularly the crowding of glands to a point where the gland area exceeds that of stroma (VPS<50%) is of practical significance at low magnification when selecting those individual regions worthy of high resolution scrutiny.
Technical adequacy of specimens and adherence to exclusion of mimics is a prerequisite for accurate EIN diagnosis. The prognostic significance of architectural features underlines the requirement for sufficiently large tissue fragments to evaluate these parameters. Our conclusions apply to those fragments in which the diagnostic focus is at least 1mm in all dimensions. Some cases excluded due to excessive fragmentation contained lesions that were suboptimal because of artifactual distortion. Other cases were also excluded since the epithelium was in strips, devoid of underlying stroma, precluding both morphometric analysis and diagnostic evaluation. All limitations are relevant whether the pathologists wish to apply the WHO94 or EIN schemes. An idiosyncracy of the morphometric implementation of the EIN system is its inability to distinguish EIN from gland-rich areas such as commonly found in mid-to-late secretory endometrium, but these are easily recognized by a well-trained pathologist.
One practical advantage of the D-score is that it can be applied to standard H&E-stained tissue sections, provided the lesion is of sufficient size. Because standard H&E sections are used, there are no additional consumable costs for expensive chemicals. Moreover, equipment for computerized morphometric D-score analysis is commercially available worldwide from different manufacturers. The total costs for a non-automated morphometry unit alone, without microscope is around US$14,000; the hardware elements including a microscope with an automated mechanical scanning stage would be around US$30,000-50,000, with direct costs averaging US$25 per case. The initial capital outlay for a highly automated interactive morphometry workstation is compensated by a lower running cost (approximately 15-30 minutes technician time and 3 minutes pathologist time per case). These expenses clearly fall within the operational budget of the host pathology department. In the Netherlands4
and Norway (unpublished results), use of the D-Score to triage women into treatment groups has standardized the therapeutic decision making process and reduced the frequency of over- and under treatment of endometrial hyperplasias, many of which would have been inconsistently or incorrectly diagnosed under WHO94. The benefit to patients, and savings in unnecessary surgery, more than counterbalances increased pathology costs. It has led to a decline in diagnosis of precancers on the one hand, but detection of early cancers that went undetected by WHO94 on the other. Although cost-effective overall, this is definitely a program that requires increased fiscal support of pathology departments in anticipation of gains to be seen elsewhere. Once the morphometrical equipment has been installed and is in use, a large number of non-endometrial25
and endometrial cancer26
morphometrical applications immediately becomes possible at minimal added cost and effort.
Additional costs are related to training and the actual time of performance for technologist and the pathologists’ time. While highly automated interactive morphometry systems are used in the routine patient care setting in many European institutions, until now such systems are rarely used in the US. Recent work has shown that subjective EIN diagnosis by pathologists using criteria designed to mimic stratification across the D-score threshold of 1 can be used, but these are not as reproducible as the formal D-score27
In summary, this long-term follow-up study of 477 women with hyperplasia suggests that the EIN classification scheme more accurately predicts the development of future cancer than the WHO94 scheme. The results of morphometric analysis are highly reproducible, as shown by its routine use in multiple laboratories in different countries. Under WHO94, many patients would have been erroneously classified as having premalignant disease, but within the EIN scheme, these women are more correctly classified as having a non-neoplastic, benign, lesion.