While 5-year survival rates for non-Hodgkin lymphoma have improved with effective first-line therapies, refractory and relapsed lymphomas continue to suffer from poor response rates and low median overall survival. The increased availability of targeted therapies has been critical to improving response rates in lymphomas. As the number of therapeutic options increases, however, identifying the most appropriate patient-specific drug combination from among a range of available drug sets is virtually impossible by conventional methods due to the large search space. Further complicating this clinical decision is the diversity in interpatient therapeutic responses due to lymphoma patient heterogeneity. Math and AI-based analytical tools are beginning to positively impact all facets of drug development and personalized medicine. Rather than aggregating large datasets from other sources (data repositories, text mining, etc.) to develop predictive models, we developed an experimentally driven small dataset analytics platform, Quadratic Phenotypic Optimization Platform (QPOP), to identify and rank drug combinations from a specific drug set search space. QPOP identifies optimal drug combinations from queried drug sets against specific biologic systems of interests, including patient-derived primary cancer cells. Utilizing small datasets built from drug combination tests designed by orthogonal array composite design, QPOP analyzes drug combination sensitivity data to identify and rank possible drug combinations. Recently, we applied QPOP towards patient-specific clinical decision support applications in refractory and relapsed non-Hodgkin lymphoma. Utilizing 1 million patient-derived primary lymphoma cells for 155 drug combination tests, QPOP ranked 531,441 possible therapeutic options from a 12-drug search set within 6 days of patient sample biopsy. Across a series of refractory and relapsed T-cell lymphoma patients, effective treatment options were accurately predicted that included both standard salvage regimens, such as gemcitabine-dexamethasone-cisplatin, as well as targeted regimens, such as bortezomib-panobinostat. For a hepatosplenic T-cell lymphoma case that had previously progressed following 6 lines of treatment, analysis by QPOP was able to accurately identify patient-specific bortezomib-panobinostat treatment that resulted in a complete response. Furthermore, QPOP-predicted drug combination sensitivity results from our study were compared to historical outcomes of a phase 2 trial of bortezomib-panobinostat in peripheral T-cell lymphoma (NCT00901147), with 20% predicted response in our QPOP-analyzed cohort mirroring 21.7% complete response observed in the previous trial. These results provide evidence that ex vivo drug combination sensitivity platforms may be useful tools for clinical decision support, for both personalized medical applications as well as enhanced clinical trial patient selection and parallel outcome tracking.
Citation Format: Jasmine Goh, Sanjay de Mel, Anand D. Jeyasekharan, Edward K.H. Chow. Drug combination analytics platform for accurate prediction of treatment response in refractory and relapsed lymphomas [abstract]. In: Proceedings of the AACR Virtual Meeting: Advances in Malignant Lymphoma; 2020 Aug 17-19. Philadelphia (PA): AACR; Blood Cancer Discov 2020;1(3_Suppl):Abstract nr PO-51.