The COVID-19 pandemic’s early stages posed a significant problem for healthcare professionals. Doctors found it difficult to anticipate how different individuals would respond to the innovative treatment for the SARS-CoV-2 virus.
As the epidemic spread, caring for patients required caregivers to make decisions about how to prioritize medical resources while having access to limited information.
TransMED is a first-of-its-kind artificial intelligence (AI) prediction tool created by researchers at the Pacific Northwest National Laboratory (PNNL), Stanford University, Virginia Tech, and John Snow Labs to lessen the burden brought on by new or uncommon diseases.
As COVID-19 unfolded over 2020, it brought a number of us together into thinking how and where we could contribute meaningfully. We decided we could make the most impact if we worked on the problem of predicting patient outcomes.
Sutanay Choudhury, Chief Scientist, Pacific Northwest National Laboratory
Khushbu Agarwal, the lead author of the study, stated, “COVID presented a unique challenge. We had very limited patient data for training an AI model that could learn the complex patterns underlying COVID patient trajectories.”
The study was published in the journal Nature Scientific Reports.
To meet the difficulty, a multi-institutional collaboration created TransMED, which uses data from current diseases to forecast the course of the developing disease.
Answering a Call to Help
In response to the COVID-19 epidemic, researchers at PNNL focused on the new challenge head-on. Choudhury was assigned to a team using artificial intelligence to construct the structures of chemicals that would provide good candidates for the creation of SARS-CoV-2 antidotes.
Additionally, he had a strong sense of empathy for the medical staff fighting on the front lines of the COVID-19 conflict.
Choudhury stated, “It was clear we needed to build more effective tools to protect both patients and caregivers better during the next crisis.”
To create such a tool, Choudhury and Agarwal collaborated with Colby Ham, Robert Rallo, the director of Advanced Computing, Mathematics, and Data Division at PNNL, along with computer scientists from Stanford University, Virginia Tech, and John Snow Labs.
One of the scientists was Suzanne Tamang. On a healthcare analytics project, she has previously collaborated with Choudhury, Agarwal, and Rallo. She was excited to take part in this study since it would allow her to put her expertise to use in helping healthcare professionals make decisions.
We all saw a need to contribute. We could leverage our abilities to build a tool with immediate value and utility for healthcare workers.
Suzanne Tamang, Assistant Faculty Director, Data Science, Stanford Center for Population Health Science
She was also the Instructor at the Department of Biomedical Data Science at the Stanford University School of Medicine.
Tamang is used to this kind of generosity. She frequently uses her time and expertise to resolve challenges related to a range of social concerns as a member of Stanford University’s Statistics for Social Good group.
Tamang added, “Sometimes, the best science occurs when researchers are driven by a desire to help.”
A New Approach to Combatting Unknown Diseases
According to preliminary findings, TransMED performs better than the models currently used to forecast patient outcomes, especially for rarer outcomes. Agarwal attributes this in part to TransMED’s capacity to examine a broad range of medical data, including that pertaining to numerous respiratory disorders.
“TransMED considers nearly all types of electronic healthcare records data such as medical conditions, drugs, procedures, laboratory measurements, and information from clinical notes. Taking this holistic view of the patient allows TransMED to make predictions much in the same way a clinician would,” stated Agarwal.
Transfer learning is another element that helps TransMED thrive. Transfer learning essentially involves using a machine learning model to solve a problem using a lot of data.
The model then applies this information to resolving challenges of a similar nature. In the case of TransMED, researchers used the model’s training data on patients with known severe respiratory disease outcomes to predict COVID-19 results.
Choudhury further stated, “Given a patient’s recent medical history, TransMED can predict a patient’s need for ventilators, or other rare outcomes 5–7 days out into the future.”
Although the application of AI in healthcare settings is still in its infancy, this research is an encouraging first step toward developing a viable model for forecasting patient outcomes. Although TransMED has not yet been used in a clinical environment, it provides a positive insight into the future of the healthcare system.
Sindhu Tipirneni and Chandan K Reddy from Virginia Tech, Pritam Mukherjee, Matthew Baker, Siyi Tang, and Olivier Gevaert from Stanford University, and Veysel Kocaman from John Snow Labs are also additional contributors to this study. A PNNL Laboratory Directed Research and Development initiative funded this study.
Agarwal, K., et al. (2022) Preparing for the next pandemic via transfer learning from existing diseases with hierarchical multi-modal BERT: a study on COVID-19 outcome prediction. Nature Scientific Reports. doi:10.1038/s41598-022-13072-w.