abstract: Calculator for ovarian carcinoma subtype prediction : Modern Pathology

Abstract:

With the emerging evidence that the five major ovarian carcinoma subtypes (high-grade serous, clear cell, endometrioid, mucinous, and low-grade serous) are distinct disease entities, management of ovarian carcinoma will become subtype specific in the future.

In an effort to improve diagnostic accuracy, we set out to determine if an immunohistochemical panel of molecular markers could reproduce consensus subtype assignment.

Immunohistochemical expression of 22 biomarkers were examined on tissue microarrays constructed from 322 archival ovarian carcinoma samples from the British Columbia Cancer Agency archives, for the period between 1984 and 2000, and an independent set of 242 cases of ovarian carcinoma from the Gynaecologic Tissue Bank at Vancouver General Hospital from 2001 to 2008. Nominal logistic regression was used to produce a subtype prediction model for each of these sets of cases. These models were then cross-validated against the other cohort, and then both models were further validated in an independent cohort of 81 ovarian carcinoma samples from five different centers.

Starting with data for 22 markers, full model fit, backwards, nominal logistic regression identified the same nine markers (CDKN2A, DKK1, HNF1B, MDM2, PGR, TFF3, TP53, VIM, WT1) as being most predictive of ovarian carcinoma subtype in both the archival and tumor bank cohorts. These models were able to predict subtype in the respective cohort in which they were developed with a high degree of sensitivity and specificity (κ statistics of 0.88±0.02 and 0.86±0.04, respectively).

When the models were cross-validated (ie using the model developed in one case series to predict subtype in the other series), the prediction equation's performances were reduced (κ statistics of 0.70±0.04 and 0.61±0.04, respectively) due to differences in frequency of expression of some biomarkers in the two case series. Both models were then validated on the independent series of 81 cases, with very good to excellent ability to predict subtype (κ=0.85±0.06 and 0.78±0.07, respectively).

A nine-marker immunohistochemical maker panel can be used to objectively support classification into one of the five major subtypes of ovarian carcinoma.

Guest Blog: A genome story: 10th anniversary commentary by Francis Collins

".....Already, in its pilot phase, this NIH-supported project has produced comprehensive molecular classification systems for ovarian cancer and glioblastoma, which is the most common form of brain cancer. This information may help doctors do a better job of matching individual patients with the therapies that are most likely to work well for them. What's more, the findings may lead to new therapies directed at the molecular changes underlying various subtypes of cancer.

Some of this is already happening today...."