Prof. Joydeep Ghosh Receives $660K from NSF

Monday, August 25, 2014

As the adoption of Electronic Health Records (EHRs) increases in the USA, the complexity of EHR data is growing dramatically. EHR data now covers diverse information about patients, including diagnosis, medication, lab results, genomic information and clinical notes. However, such large volumes of information do not readily provide accurate and succinct patient representations for effective and customized healthcare.

According to WNCG Prof. Joydeep Ghosh and his team of collaborators, the trick is to transform data into knowledge by translating complex, interconnected EHR data into concise and meaningful clinical concepts, or phenotypes, about patients. A phenotype is a collection of observable traits that results from the interactions between genetic expression and environmental influence. These phenotypes can be more easily interpreted, accepted and used by physicians. However, current approaches to deriving phenotypes from EHRs are time consuming and demand much human expertise.

Prof. Ghosh was awarded over $660,000 to develop a computational framework for semi-automatic, high-throughput phenotyping of EHR at UT-Austin. His approach employs machine learning primarily based on multi-tensor factorization. Prof. Ghosh’s collaborators include Abel Kho from Northwestern University at Chicago, Bradley Malin and Joshua Denny from Vanderbilt University Medical Center and Jimeng Sun from the Georgia Institute of Technology. The total award across the four institutions exceeds $2 million.

The goal of Prof. Ghosh’s research is to model data as multiple, interconnected relationships, such as the relationship between a patient and their medication and diagnosis, or a patient and their symptoms. His research team will develop scalable algorithms to analyze these relationships and derive hidden concepts from the available data. Clinical experts will refine these concepts into specific phenotypes.

The WNCG team’s computational framework will receive further development and validation in clinical environments at Vanderbilt and Northwestern. The trials will test the framework on high-impact targets such as Hypertension, Type II Diabetes, Crohn’s Disease, Rheumatoid Arthritis and Multiple Sclerosis.

Many research challenges remain, including patient representation, high-throughput phenotype generation from EHRs, the need for expertly-guided phenotype refinements and the adaptation of phenotypes across multiple institutions. Prof. Ghosh predicts his research will lay the foundation for large-scale studies of EHR data that combine computer science and medical informatics to enable new clinical discoveries.