July 23, 2019
Converging RWD and AI to Better Identify Cases of Metastatic Breast Cancer
The changing needs of cancer patients require new approaches to clinical drug development. Any significant change and transformation will need true convergence and collaboration across new digital data sources, powerful computing and AI technologies and regulatory authorities.
Concerto HealthAI’s definitive oncology dataset – the largest set of independently sourced, population-scale oncology patient records – allows us to build and train powerful AI models on thousands of EMR data points, in line with FDA guidance on predictive enrichment for clinical trials.
At ISPOR 2019, Concerto HealthAI researchers presented a research abstract poster on an AI model that can correctly identify 32% more cases of metastatic breast cancer compared to simple database queries with traditional business rules – with substantial time and resource savings for life science research. The abstract was also selected as a semi-finalist for the Research Poster Presentation Award.
This AI model was trained, tested and validated using structured electronic health records (EHR) data of ~130,000 breast cancer patients. Clinical validation was carried out against a set of approximately 6,000 patients whose data was manually curated by expert oncology nurses. The model had high rates of precision and recall on a dataset which had ~20-30% metastatic patients.
The outcome demonstrates the feasibility of extracting missing metadata using hundreds of other sparsely populated data points representing the patient’s medical journey, which would otherwise not be possible using simple business rules. This highlights the possibility of creating richer datasets from structured EHR data in order to make that data significantly more useful for clinical research.
Research showed that the AI model could be used to quickly identify eligible patients for clinical trials or retrospective outcomes science studies – saving substantial time and resources when compared to traditional, manual abstraction processes currently used to achieve the same insights.
Our ISPOR abstract confirms the utility of highly validated, precise AI models in life sciences and clinical research. Models like this rapidly enhance and expand existing Real-World Data to finally give researchers a powerful set of tools and technologies with which to test trial design feasibility and construct more effective and efficient study protocols at the outset, with significant cost and time savings.
This class of models will bring speed and scale to the entire life sciences R&D pipeline so more successful studies can be conducted faster, bringing needed therapeutic innovations to patients more quickly. Head to the ISPOR website to access the full abstract and read our full press release here.