Testing Algorithms for the Diagnosis of Malignant Glandular Tumors of the Uterine Cervix Histotyped per the International Endocervical Adenocarcinoma Criteria and Classification (IECC) System

The International Endocervical adenocarcinoma Criteria and Classification (IECC) categorizes tumors into human papilloma virus (HPV) associated (HPVA), not associated (NHPV), and invasive adenocarcinoma not otherwise specified (IA NOS). HPVA and NHPV encompass 11 histotypes and an algorithm of mucin content, HPV ribonucleic acid (RNA), estrogen receptor and GA-TA3 is proposed for the diagnosis of most. In this study, the IECC algorithm's diagnoses were compared with hematoxylin and eosin (H&E) based IECC histotyping. Kappa statistics measured performance agreement. With additional markers, hierarchical clustering by random forest (RF) classification identified the most discriminating between tumor types, and investigated other algorithms. Three pathologists independently reviewed digitized H&E images of n =152 primary cervical adenocarcinomas for IECC histotype and mucin content, and tissue microarrays for expression of HPV RNA by in situ hybridization and 16 antibodies by immunohistochemistry. Results were finalized by consensus. There were n = 113 HPVA, n = 22 NHPV, and n =17 IA NOS. Mucin was obvious inn = 36 and limited in n = 116. Among n =124 with satisfactory test results, HPV RNA was positive in n = 96, estrogen receptor in n = 72, and GA-TA3 in n =15. The IECC algorithm diagnosed n = 99 which agreed with H&E histotyping in n = 64 for a fair kappa of 0.36 (95% confidence interval, 0.21-0.50): n =12 were undiagnosed and n =13 were IA NOS. Small sample sizes restricted RF to HPVA versus NHPV which were discriminated by p16, HPV RNA, and MUC6 with an area under the curve of 0.74 (95% confidence interval, 0.58-0.90). The IECC algorithm for histotyping under-performed. The RF algorithmin for categorization was favorable, but validation in larger studies and investigation of additional algorithms to discriminate between all IECC histotypes are needed.

Automated summary

This study aimed to compare the diagnoses of the International Endocervical adenocarcinoma Criteria and Classification (IECC) algorithm with H&E-based IECC histotyping and validated the use of random forest (RF) classification for more accurate discrimination. The results showed that the IECC algorithm performed poorly in histotyping, while the RF algorithm showed favorable results in distinguishing between HPVA and NHPV.