Assessing the Feasibility of Pathology Informatics Approaches for Subtyping Diffuse Large B-Cell Lymphoma (DLBCL)

Introduction

Patients (pts) with DLBCL can experience marked difference in overall survival (OS) based on clinical characteristics, race, and biological subtype. Immunohistochemistry (IHC) algorithms can stratify pts into the germinal center B-cell-like (GCB) or activated B-cell-like (ABC) subtype, which is associated with worse OS with current standard of care therapy with rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone (RCHOP; Meyer et al. 2011). However, these algorithms can be challenging to perform in clinical practice. We evaluated the frequency of GCB and non-GCB subtypes in a retrospective cohort of patients and evaluated the feasibility of performing automated pathology informatics computer segmentation algorithms to assess DLBCL subtype.

Methods

We included all patients diagnosed with DLBCL at Emory University from 1995-2016. We collected demographic, clinical, laboratory and pathologic data on all patients. We evaluated GCB and non-GCB status using the Hans algorithm as performed during routine clinical care, collected IHC slides for H&E, CD10, CD20, CD30, BCL2, BCL6, C-MYC, and MUM1, and assessed the feasibility of performing automated computer segmentation algorithms to assess DLBCL subtype using these materials.

Results

Out of 364 patients with DLBCL in our database, 151 had available data for DLBCL classification by the Hans algorithm and 103 had available slides for computational algorithms. For the entire dataset the median age was 61 years, 51% were male, 31% had stage III/IV disease. Among patients who had classification data, the median age was 61 years, 53% were male, 45% had stage III/IV disease, and 47% GCB DLBCL and 53% had non-GCB. Among patients with slides available for computational algorithms the median age was 64 years, 53% were male, 48% had stage III/IV disease.

Conclusions

In this retrospective cohort study, the group of patients with available slides for performing computer segmentation algorithms had similar demographics and disease characteristics to patients in the general population of DLBCL at an academic medical center. This dataset and associated pathology images provides a useful resource to evaluate whether computational algorithms can aid in defining the prognosis for DLBCL patients. Future directions for this work will involve the establishment of an online DLBCL digital archive to develop a more precise, repeatable and objective image-analysis based scoring of DLBCL tissues to improve the prognostic accuracy and classification of DLBCL.