Pattern based classification of endocervical adenocarcinoma: a review

Summary

While endocervical adenocarcinoma is comprised of several histological subtypes, the most common subtype is human papillomavirus-associated usual type, and management of these tumours is primarily determined by FIGO (International Federation of Gynecology and Obstetrics) stage, a clinically based staging system. Early stage cervical cancer is determined by the pathological evaluation of tumour microscopic measurement (depth of invasion), which can be particularly challenging because of the lack of a defined point of origin. Yet important treatment decisions, cone versus radical surgery, are based on this imperfect system, resulting in overtreatment and related morbidities in many patients for whom it may not be necessary. There is a growing consensus, however, for a more conservative approach, one that reduces morbidity and prevents loss of fertility in these (often young) patients. This movement has been supported, in part, by the development of a morphology based risk stratification system which was devised in order to recognise those tumours that, while invasive, could potentially be treated more conservatively. In this review, we provide the reader with the background and rationale for a more conservative approach in treating endocervical adenocarcinoma, summarise the risk stratification system, and review the system’s utility and reproducibility. In addition, we comment on recent updates that attempt to further refine the system. The application of this morphology based classification could help identify a subset of patients with endocervical adenocarcinoma (who would otherwise undergo radical surgery based on FIGO staging alone) that have good clinical outcomes and could be treated more conservatively.

CERVICAL CARCINOMA STAGING AND TREATMENT

Endocervical adenocarcinoma is a heterogeneous disease comprised of various histological subtypes that have different aetiologies and clinical outcomes.^1,2 The most common type is human papillomavirus (HPV)-associated usual endocervical adenocarcinoma, and despite the success of cervical cancer screening, the incidence of this adenocarcinoma appears to be increasing.³ Management of cervical cancer is based primarily on the International Federation of Gynecology and Obstetrics (FIGO) Staging system, which relies on clinical parameters, meaning that clinicians determine tumour stage by the presence or absence of a grossly visible or palpable lesion.^4,5 When a mass is clinically identified, the tumour is stage FIGO IB, independent of the microscopic examination and its morphologic features, including depth of invasion. Since patients beyond stage FIGO IB1 are usually not surgical candidates, pathologists infrequently encounter these types of cases.⁶ In tumours that are not clinically visible, stage is determined by microscopic measurements.⁷ Therefore, in early stage cervical carcinoma, accurate measurements are crucial for patient management. According to the FIGO staging system, tumours measuring ≤7 mm in horizontal extent and invading no greater than 5 mm are stage IA, while those with greater measurements but less than 4 cm in size are stage IB1. The difference in treatment based on microscopic measurements can be dramatic, such that a stage IA1 tumour with no lymphovascular invasion can be treated with cone biopsy alone, yet a tumour that is 1 mm greater in depth of invasion (stage IA2) could result in a radical hysterectomy with lymph node dissection.⁸

There are several problems with the current staging system as it applies to adenocarcinoma, as the staging parameters lack specific guidelines on how to measure invasive adenocarcinoma. Unlike the squamous epithelium which has a defined basement membrane, beyond which invasion is obvious, endocervical glands do not have such landmark. Rather, the endocervical mucosa is a series of invaginations along the canal that form gland-like structures throughout the cervical wall, sometimes reaching deep into the wall or expanding with mucus to form deeply placed Nabothian cysts, making determination of the exact point of origin of invasive adenocarcinoma quite challenging.^9,10 This is especially challenging if adenocarcinoma in situ (AIS) is associated with an invasive carcinoma, since a clear distinction between the two may not be possible in up to 20% of cases.^10–12 Additionally, when the tumour is predominantly exophytic, it can be difficult to know how to measure the depth of invasion (although in most cases, there will be clinically visible and classified as stage IB, independent of pathological measurement). Inconsistent measuring methods can greatly affect therapy since, as mentioned above, a difference of a millimetre can mean the difference between retaining or losing fertility, as well as undergoing a morbid procedure such as lymph node dissection.

RATIONALE FOR A MORE CONSERVATIVE TREATMENT APPROACH AND NEED FOR A NEW RISK STRATIFICATION SYSTEM IN ENDOCERVICAL ADENOCARCINOMA

Conservative management of invasive tumours has become common practice in oncology in an effort to preserve organ function and decrease morbidity without affecting patient outcomes. The uterus, though it does not serve a life sustaining role, does have the unique function of sustaining pregnancy; with many more women delaying child-bearing into their 30s and 40s, the diagnosis of an invasive cervical cancer may not allow that option, with radical hysterectomy as standard treatment in many cases. Not only does the patient lose the ability to become pregnant, but complications related to radical surgery and lymph node dissection have been well reported and are not inconsequential. These include voiding difficulties, urinary tract infection, incontinence, pyrexia, haemorrhage, nerve-site injury and lymphoedema.¹³ Conservative management is becoming more accepted, with the increasing use of sentinel lymph node mapping in gynecological tumours^14–17 and trachelectomies and cone excisions being included in the treatment algorithms of the latest National Comprehensive Cancer Network Clinical Practice Guidelines in Oncology (NCCN Guidelines).⁸ However, these conservative practices are not widespread and most gynecological oncological surgeons will still perform full lymph node dissection for early stage (IA2–IB1) cervical carcinoma. Furthermore, the need for a more specific and tailored treatment algorithm is becoming apparent in this era of personalised medicine. Not much progress has been made in developing ‘targeted therapy’ in cervical cancer; in the head and neck region where the other significant proportion of HPV associated tumours occur, the presence of HPV is of critical importance in determining the prognosis and the treatment plan, since data have now shown that HPV related squamous cell carcinomas respond better to radiation therapy than HPV unrelated tumours.^18–20 In the cervix, the vast majority of cancers are HPV associated and, therefore, there is not the same opportunity to stratify patients based on HPV status. Hence, stratification based on other factors such as pattern of invasion may provide more information in determining therapy and prognosis.

PATTERN BASED CLASSIFICATION SYSTEM

A recently described morphology based classification system for endocervical adenocarcinoma has proposed a novel method of stratifying patients into risk categories based on the pattern of stromal invasion.^21–23 This new system has the potential to greatly reduce the number of unnecessary radical procedures in patients with early stage disease. The initial impetus to study the value of pattern classification in endocervical adenocarcinoma stemmed from the observation that radical surgeries were being performed on patients whose tumours showed histological features that pathologists typically associate with indolent behaviour. These included tumours without destructive stromal invasion or evidence of lymphovascular invasion. These types of cases were often the ones being sent for expert consultation to determine (1) the presence of invasion and/or (2) the depth of invasion. As mentioned previously, the lack of a defined basement membrane in endocervical glands presents unique challenges in the histological evaluation of true stromal invasion and the distinction between in situ and invasive disease.

Diaz De Vivar et al. initially described these three patterns of invasion in endocervical carcinomas in 2013²¹ with subsequent expansion of their morphological details by Rutgers et al. in 2016 (Table 1).²⁴ It is important to note that the system only applies to usual type HPV-associated endocervical adenocarcinomas but not to any of the other variants (e.g., gastric-type, clear cell, mesonephric, etc.) and that the entire tumour must be examined histologically (if obvious destructive stromal invasion is absent) in order to classify the tumour into the appropriate group. Pattern A endocervical adenocarcinomas are characterised by well-formed glands frequently forming groups with relatively well preserved lobular architecture without destructive stromal invasion, single cells or detached clusters of tumour cells (Fig. 1a,b). There should be no solid growth but complex intraglandular proliferations are acceptable (cribriforming or papillae). Lymphovascular invasion should be absent as otherwise would disqualify the tumour from pattern A (all pattern A tumours initially studied lacked lymphovascular invasion). Pattern B tumours show localised (limited/early) destructive invasion arising in a background of pattern A glands (Fig 1c,d). Individual cells or clusters of tumour cells are seen in desmoplastic or inflamed stroma, and these foci can be single, multiple or linear at the base of the tumour, but should not exceed 5 mm (total). Pattern C tumours show diffuse destructive invasion that usually elicits a desmoplastic/inflammatory response (Fig. 1e). The glands can be angulated, or have a canalicular/labyrinthine appearance, and incomplete/fragmented glands (as seen in MELF pattern of endometrioid carcinomas) are frequent, sometimes associated with mucin lakes (Fig. 1f). Solid growth can also be seen. Lymphovascular invasion can be present in either pattern B or C and should be documented separately. The geographic relationship of the tumour to large calibre vessels does not affect pattern classification; e.g., pattern A glands involving deep stroma, usually in the vicinity of medium calibre vessels, are still pattern A independent of the depth of invasion.

Table 1.

New risk stratification system for invasive endocervical adenocarcinomas based on pattern of invasion

Pattern	Risk stratification system
Pattern A	• Well-demarcated glands with rounded contours, frequently forming groups
	• No destructive stromal invasion
	• No single cells or cell detachment
	• No lymphovascular invasion
	• Complex intraglandular growth acceptable, i.e., cribriform, papillae
	• Lack of solid growth, i.e., architecture well-moderately differentiated
	• Depth of the tumour or relationship to large cervical vessels not relevant
Pattern B	• Localised (limited, early) destructive stromal invasion arising from pattern A glands (well-demarcated glands)
	• Individual or small groups of tumour cells, separated from the rounded gland, often in a focally desmoplastic or inflamed stroma
	• Foci may be single, multiple, or linear at base of tumour
	• Lymphovascular invasion +/−
	• Lack of solid growth, i.e., architecturally well-moderately differentiated
Pattern C	• Diffuse destructive stromal invasion, characterised by: diffusely infiltrative glands, with associated extensive desmoplastic response; glands often angulated or with a canalicular pattern, with interspersed open glands
	• Confluent growth filling a 4× field (5 mm): glands, papillae (stroma only within papillae), or mucin lakes
	• Solid, poorly differentiated component (architecturally high grade); nuclear grade is disregarded
	• Lymphovascular invasion +/−

Taken from Roma et al.²³

Fig. 1.

(a) Pattern A endocervical adenocarcinoma showing well demarcated glands with well preserved lobular architecture and pushing border deep into cervical stroma (b) without destructive invasion; (c,d) Pattern B showing limited destructive invasion as clusters of tumour cells with desmoplastic reaction in a background of Pattern A type glands; (e) Pattern C tumour with diffuse destructive invasion that can sometimes show (f) mucin lakes.

Examination of the entire tumour is required to differentiate between patterns A and B, as pattern B tumours by definition are composed of pattern A type glands but additionally have microscopic tumour cells involving the adjacent stroma with destructive features that might be less than 1 mm away. Thus, careful examination is necessary. It is usually recommended that when in doubt, obtaining deeper sections from pattern A tumours with areas that might represent pattern B, such as loose, desmoplastic, or inflamed stroma would be helpful. Careful examination for lymphovascular invasion is also crucial, as the presence of tumour in vascular spaces would by definition exclude a pattern A. Differentiating some pattern A tumours from AIS can be challenging but from results of studies discussed herein of inconsequential relevance. Differentiating pattern B from pattern C depends on both quantitative and qualitative features. Obvious pattern C adenocarcinomas are diagnostically straightforward such that most pathologists can confidently diagnose them as definitively malignant and invasive. However, when the entire tumour does not show pattern C features, quantitative criteria are required to determine the upper end of the spectrum of pattern B and lower end of the spectrum of pattern C. The presence of ≥5 mm (4× field) of one or more of the following features qualifies the tumour as pattern C: destructive stromal invasion, confluent glands, papillary growth, extravasated mucin admixed with tumour cells, or solid growth.²⁴

Of all cases analysed thus far, pattern A endocervical adenocarcinomas have been limited to stage I with none having documented lymph node metastases or recurrences, regardless of tumour size or depth of invasion.^21,23–25 In contrast, pattern C tumours frequently present at stage II or higher with lymph node metastases (22.5%), and are associated with recurrences (19.7%) and death from disease. Interestingly, all pattern B tumours were also limited to stage I and the few cases with lymph node metastases⁴ all had evidence of lymphovascular invasion. The potential utility of pattern analysis in determining the appropriate treatment and predicting outcomes cannot be understated. Additional studies by independent groups have reported similar results with good correlation between pattern of invasion and outcomes, similar to the original studies.^25–27

Another frequent dilemma, the common difficulty in distinguishing AIS from invasive carcinoma, appears to be addressed in this classification system scheme. As mentioned already, the distinction between in situ and invasive adenocarcinomas can be extremely challenging and, in fact, cannot be determined with certainty in about 20% of cases.¹² A study by Parra-Herran et al. showed high interobserver variability (kappa value of only 0.37) in distinguishing AIS from all patterns of invasive adenocarcinoma, although the agreement was worst for AIS versus pattern A.²⁶ Given the excellent outcomes associated with pattern A, a more conservative approach similar to that for AIS (e.g., cone excision with negative margins) may be appropriate. In fact, some authors have proposed that in cases where AIS cannot be definitely separated from invasive carcinoma, the terminology ‘non-destructive endocervical adenocarcinoma’ be utilised, which would obviate the need to absolutely determine whether a lesion is invasive in these circumstances.^26,27 In a study by Douglas et al., the authors compared the ability of gynecological pathologists to correctly separate pattern A adenocarcinoma from benign endocervical glandular proliferations by architecture alone and found that in the majority of cases, pattern A invasion was recognised based solely on the glandular pattern distribution when blinded to nuclear/cytoplasmic morphology with some invasive carcinomas being classified as benign glandular proliferations and vice versa.²⁷ While this attests to the fact that pattern A tumours are associated with a non-destructive form of stromal invasion, there may have been inherent bias on the part of the reviewers, knowing that they were being asked specifically to identify invasive tumours with deceptive patterns. Regardless, this study validates the use of increased glandular density as a criterion in the diagnosis of non-destructive invasive carcinoma, and the combination of complex glandular architecture in addition to the cytological features is helpful to differentiate in situ versus non-destructive invasive lesions.²⁶

REPRODUCIBILITY

The utility of such a pattern based classification system is determined by the ability of pathologists to consistently, accurately and reproducibly diagnose the appropriate patterns in endocervical adenocarcinoma. There have now been several, mostly independent studies, looking at this very issue with promising results. Paquette et al. evaluated interobserver variability amongst four gynecological pathologists with different levels of professional experience and found that the kappa values ranged from fair to almost perfect (κ range 0.24–0.84) with most falling into the moderate to substantial agreement range for the 3-tiered system.²⁵ They also analysed kappa values for a 2-tiered system (pattern A versus patterns B/C combined) and found modestly improved results. Interestingly, these authors included large cervical biopsies in addition to resections, including hysterectomy, cone and trachelectomy specimens and found that in some cases, the biopsy yielded the best result for assessment of the pattern of invasion because most of the tumour was present in the biopsy with minimal or no residual tumour in the resection specimen.

Parra-Herran et al. also reported on interobserver variation among nine pathologists with gynecological pathology varied experience/expertise and diverse schools of training.²⁶ Overall agreement was moderate (κ = 0.65); distinction between non-destructive (AIS and pattern A) and destructive (patterns B and C) carcinoma also showed good agreement (κ = 0.62), being higher in experienced pathologists (>3 years in practice). Although perfect agreement (9/9 pathologists) was seen in only 11 cases (11%), all with destructive invasion, concordance by a majority (≥5/9 reviewers) was achieved in 82 (85%) cases. Surprisingly, even though measuring depth of invasion in endocervical adenocarcinoma poses difficulties in day to day practice, a systematic review in this same study showed excellent agreement amongst gynecological pathologists, better than for pattern classification.²⁶ It is to be noted, however, that while estimation of depth of invasion showed excellent reproducibility (interclass correlation coefficient = 0.82), different measurements by at least one reviewer was sufficient to change the FIGO stage, potentially affecting treatment modalities.

Another interobserver study by Rutgers et al., which included some of the authors of the original risk stratification study as well as additional experts in the field, reported majority agreement in 81% of cases with complete or near-complete (6 of 7) agreement in 50%.²⁴ The authors were provided with additional morphological criteria and algorithmic schema to help differentiate problematic areas, in particular distinguishing pattern A from pattern B and providing quantitative criteria in distinguishing pattern B from pattern C (Table 1, Fig. 1). Overall concordance was 74% with kappa values of 0.54, 0.32 and 0.59 for patterns A, B and C, respectively, with the greatest reproducibility being for pattern C. Of interest, some of the additional morphological criteria provided for pattern C (such as confluence filling a 4× field, labyrinthine glands, or solid architecture) accounted for most pathologists undercalling a pattern C tumour as A or B. Failing to identify a few individual infiltrative cells was the most common cause of undercalling pattern B, and small foci of inflamed, loose or desmoplastic stroma lacking infiltrative tumour cells accounted for misclassifying pattern B as pattern A.

A prerequisite for the application of the pattern based classification system is the histological examination of the entire tumour on a resection specimen with negative margins, at least for non-destructive patterns. However, the decision for radical hysterectomy and lymph node dissection is often based on a pre-operative biopsy or LEEP where it may not be possible to definitively determine the pattern of invasion. To address this issue, Djordjevic and Parra-Herran analysed the concordance of pattern classification in pre-resection biopsies and cone/LEEP specimens to the final pattern in the hysterectomy and trachelectomy specimens.²⁸ They also compared pattern classification from the most superficial 2 mm portion of the resected tumours to the overall pattern of the entire tumour (in cones/LEEPs and radical resections), assuming any biopsy would have sampled the most superficial areas of the tumour. There was excellent correlation between the pattern at the tumour surface and resection specimens (95.7%), as well as between cone/LEEP and resection specimens (92.8%). A significant difference was seen, however, between the initial biopsy material and the overall tumour at resection (37.5% agreement). In all discrepant cases, the biopsy did not sample destructive stromal invasion that was present on excision. Half (8/16) of the biopsies showed non-destructive pattern A glands, only two of which remained pattern A on final examination. In six of these discrepant cases, while there was glandular complexity and exophytic papillary growth, it did not fulfill the quantitative criteria for destructive invasion (filling >5 mm or 4× field for pattern C), and therefore they were classified as pattern A. On excision, marked gland confluence and other patterns of destructive invasion were evident. In contrast, pattern C on biopsy had 100% concordance with the resection, which follows logically since the minimum criteria for pattern C were already met pre-operatively.

Tumour distribution based on pattern seems to be fairly consistent, with 12–21% pattern A, 17–36% pattern B, 43–71% pattern C being reported in various studies.^21–28 In both studies by Paquette and Parra-Herran, interobserver agreement was best when comparing non-destructive (pattern A) versus destructive (patterns B and C) invasion.^25,26 The use of a binary system is appealing since it often allows for better inter and intra-observer concordance. However, as the majority of even destructively invasive tumours have uneventful outcomes, there is a need to further stratify these tumours with more refined clinicopathological parameters.

SUBSTRATIFICATION OF DESTRUCTIVELY INVASIVE TUMOURS

It is important to determine additional features predictive of behaviour in order to strengthen the prognostic value of this tumour pattern classification. Lymphovascular invasion is a well-known feature that aids in determining outcome in cervical cancer.^12,29,30 In one study reviewing squamous and adenocarcinomas of the cervix, the presence of lymphovascular invasion was most closely associated with mortality, while tumour grade, depth of invasion and tumour size were not.³⁰ However, a recent study evaluating the importance of lymphovascular invasion in pattern C tumours showed that tumour size, horizontal spread, and lymph node status were significantly associated with outcome in univariate, but not multivariate analysis, while depth of invasion and lymphovascular status independent of lymph node involvement were not associated with outcome.³¹ It should be noted that this study was looking specifically at 127 pattern C tumours from the original study,²¹ which may account for the apparent discrepancy in results. Most patients (88%) without lymphovascular invasion or lymph node involvement were alive with no evidence of disease. A similar percentage of patients (82%) with lymphovascular invasion but negative lymph nodes were also alive and well but only 45% of patients with metastatic disease within lymph nodes, while 39% had died of disease. These data indicate that the presence of lymph node metastasis is a significant predictor of aggressive behaviour in pattern C tumours and stratifying these into subgroups based on lymphovascular invasion and lymph node status could be an adjunct tool to dictate treatment in patients with pattern C tumours. A different approach was taken by the same group in an attempt to further refine histological features that may predict tumour behaviour, by evaluating the morphological variants of pattern C tumours (i.e., destructive invasion).³² A larger subset of the original pattern C tumours (189) was analysed and further substratified based on different architectural growth features: diffuse destructive, confluent, extensive linear destructive, band-like lymphocytic infiltrate, solid, and micropapillary (Fig. 2a–f). Almost half the tumours (44%) contained mixed patterns, while the majority of the pure tumours were predominantly composed of diffuse destructive (17%), extensive linear destructive (14%), and confluent patterns (12%). Less than 5% of each of the other growth patterns was seen. All patients with tumours showing micropapillary growth had lymph node metastases versus none with tumours showing extensive linear destructive growth.³² Recurrent disease was seen mostly in patients with tumours showing diffuse destructive growth (44%) but not in tumours with band-like lymphocytic infiltrate. Whether or not depth of invasion in pattern C adenocarcinoma is an independent prognostic factor is uncertain. Alvarado-Cabrero showed an 11% increase in the risk of dying from disease for every millimetre increase in the depth of invasion in 189 pattern C cases, while Roma et al. reported that depth of invasion was not associated with outcome in a subset (127) of the same patients.^31,32 This discrepancy shows the difficulty in understanding the significance of depth of invasion based on case selection. Both these publications support further study and stratification of pattern C tumours into subgroups in order to better treat patients and predict prognosis. Therefore, the best approach may be to incorporate the pattern based classification into the current FIGO staging system as previously suggested.³³

Fig. 2.

Pattern C tumours, morphological variants: (a) diffuse destructive; (b) confluent; (c) extensive linear destructive (>5 mm); (d) band-like lymphocytic infiltrate; (e) solid; (f) micropapillary.

CONCLUSION

A novel method of stratifying usual type endocervical adenocarcinoma by morphology has been shown to be applicable to resections (cone/LEEP, hysterectomy, trachelectomy) with good reproducibility. Application to biopsy material is suboptimal, with poor correlation when the biopsy shows a non-destructive pattern. It has been recommended that with destructive invasion seen on biopsy, standard treatment modalities could be followed upfront with further stratification based on lymphovascular invasion, lymph node status and type of destructive growth pattern. In contrast, after biopsies with a non-destructive pattern, a LEEP or cone procedure to better determine the true pattern would be informative. If completely resected and non-destructive, the patient would not need further therapy, avoiding the complications of radical treatment modalities. If there is any component of destructive invasion, additional surgery would be recommended.