Presentation Detail
- Speaker: Sungwon Lim, PhD, CEO & Co-Founder of ImpriMed, Inc.
- Date: April 9, 2022
Research Abstract
Introduction
Most canine B-cell lymphoma patients respond well to first-line CHOP therapy, but the response rate is much lower after relapse occurs. To help clinicians find effective anticancer drugs for individual relapsed B-cell lymphoma patients, we developed artificial intelligence (AI) models that make personalized drug response predictions and analyzed the performance of these models in a retrospective clinical study.
Methods
The dataset used to train the AI models includes real-world clinical outcome reports from 1,497 canine patients as well as our immunoprofiling and drug sensitivity testing data. A study population of 66 patients with relapsed B-cell lymphoma was selected from the database and removed from our AI training dataset. Response predictions were made for 10 different drugs for each patient in the study population. To assess the quality of these predictions, we divided the study population into two groups—a population where the treatment was highly concordant with our AI predictions and a group where concordance was lower—and compared their clinical outcomes.
Results
Both the overall response rate (ORR) and the complete response rate (CRR) were substantially higher in the high concordance group than in the low concordance group (68.9% high vs. 35.1% low for ORR with p=0.013; 41.3% high vs. 24.3% low for CRR with p=0.19).
Conclusion
Our results suggest that AI models trained to predict individual patients’ clinical outcomes may help veterinarians select drugs likely to elicit a clinical response in patients with relapsed B-cell lymphoma that are refractory to standard treatment options.
