68. Orthogonal approaches to validate a knowledgebase of interpretations of clinically relevant somatic cancer variants

Aly Abdelkareem

Andrew Bredemeyer

Andy Bredemeyer, PhD is Senior Vice President, Medical Center of Excellence at Velsera. He was a product leader at Pierian, now part of Velsera, creating valuable solutions for clinical genomic laboratories through the development and growth of Pierian’s laboratory developed test and and IVD platforms. In collaboration with Pierian’s Knowledgebase and software teams, he has helped design an automated tertiary analysis and clinical interpretation system and partnered with biocurators to create curation processes and automation for medical device compliance. He previously led Pierian’s Customer Success team and worked with numerous customers to bring workflow and reporting solutions to clinical molecular laboratories.

Dr. Bredemeyer holds a PhD in Immunology from Washington University in St. Louis. He served in policy and finance roles at the NIH and at Massachusetts General Hospital research labs. Dr. Bredemeyer helped lead NGS-based clinical assay development and validation efforts at Washington University’s Genomics and Pathology Services laboratory, an early pioneer in clinical NGS testing.


Andrew Bredemeyera, Gargi Nandab, Anuja Jedheb, Aditi Phatakb, Nikita Deshmukhb, Vidya Iyerb, Rhucha Vatturkarb, Sukanya Senguptab, Pamela Heskera, Rakesh Nagarajana

aVelsera, Creve Coeur, MO, United States; bVelsera, Pune, Maharashtra, India

The interpretation¬† of genomic data by clinical laboratory professionals remains a bottleneck in precision oncology. We applied several approaches to analyze the comprehensiveness and accuracy of variant classification and clinical impact assertions of Velsera’s Knowledgebase (KB), an automated solution to interpret somatic mutations in cancer.

A  panel of four independent pathologists reviewed the accuracy of variant interpretations, classification on the AMP/ASCO/CAP scheme, clinical impact associations, and clinical trial matching. We observed excellent concordance for assertions described in drug labels and clinical practice guidelines, and high accuracy for clinical trial matching. The concordance was lower for assertions from published clinical research, due largely to inter-reviewer disagreement on the definition of well-powered studies, which influences classification.

Most high-confidence solid tumor assertions from the CIViC database (78%) were shared with Velsera KB Tier IA (AMP/ASCO/CAP) assertions. 15.6% of high-confidence CIViC assertions did not qualify as Tier IA evidence because they are not described in drug labels or recommended in clinical practice guidelines.

We also assessed the performance of the Velsera KB for assignment of clinical relevance to genomic variants. The KB agreed with laboratory professional assignment of clinical relevance in signed out clinical cases for all consensus clinically relevant (n=78) and non-relevant (n=22454) alterations.

These approaches demonstrate that the Velsera KB is highly concordant with the consensus opinion of the medical community. Comparisons to additional orthogonal knowledgebases will help ensure quality of the KB and establish best practices for benchmarking the performance of clinical interpretation software.