Yi Li, Professor of Biostatistics, and his team collaborated on a project on the risk factors associated with COVID-19 outcomes, utilizing Precision Health resources and data including chest x-ray imaging to support this research.  We connected with Dr. Yi Li (the corresponding author) and Drs. Yuming Sun and Stephen Salerno (both are Dr. Li’s recent PhD graduates) to learn more about their experience, the insights gained, and what’s next:

• Briefly, what was the aim of this project at the outset?

During the COVID-19 pandemic, Yi Li’s lab collaborated with Michigan Medicine and Precision Health to conduct extensive research on the risk factors associated with COVID-19 outcomes.  This work focused on machine learning techniques to assess the prognostic utility of radiomic features for in-hospital COVID-19 mortality. As portable chest X-rays are efficient in triaging emergent cases, we sought to investigate whether these imaging techniques offer supplementary information for predicting survival among patients with COVID-19, beyond the conventional clinical indicators of disease severity.

• How did the Precision Health resources, tools and data you utilized contribute to the ultimate success/outcome of the project?

Precision Health resources were invaluable for this research. Owing to the wealth of information available through DataDirect, we were able to extract demographic, socioeconomic, and clinical risk factors related to COVID-19, as well as physiologic measurements on admission and information on respiratory support, such as mechanical ventilation.  With access to the electronic health record (EHR) and X-ray data from these sources via DataDirect, our group was in a unique position to develop new methodologies for identifying patient characteristics, clinical factors, and radiomic features linked to COVID-19 disease severity and survival outcomes.

• Anything to share on the ease/usability/usefulness of these tools, data & resources that could encourage/help future users?

The Precision Health DataDirect tool and deidentified research data warehouse were incredibly easy to use and provided us with a convenient framework for connecting the imaging studies done by Mike Sjoding’s team to valuable clinical information found in Michigan Medicine’s EHR. In addition, the analytics platform documentation site provides detailed instructions for using the analytic tools and for understanding the data cleaning and extraction processes. This was important for us to communicate our results effectively and provide a reproducible workflow for other researchers at the university, as well as other institutions. Furthermore, the Armis2 high-performance computing environment provided us with the computational power to process and analyze vast amounts of data efficiently. Beyond this, the research facilitators and project managers affiliated with the Precision Health Initiative were incredibly helpful and accommodating as we developed this study and needed additional support with our data management needs.

• If you could summarize your salient findings briefly, and any next steps or further questions this research might have inspired.

We observed notable enhancements in the predictive accuracy of our models when incorporating imaging features extracted from chest X-rays, especially among older and more medically compromised patients. This experience has prompted us to further contemplate the advantages of incorporating radiomic features alongside clinical information when developing predictive models. We have noticed overlapping information between these two data sources, particularly in outcomes like mortality. While our focus has been on COVID-19 within the context of infection disease pulmonary critical care, the integration of data from diverse sources could be explored in other domains, including precision oncology.

• Based on your experience with Precision Health resources etc and this publication/project, will you look to Precision Health in the future for support on other studies?

Absolutely! We believe this project was novel both clinically and statistically, and its clinical novelty benefited immensely from our ability to utilize the extensive COVID-19 data resources available through DataDirect and Precision Health. We are very grateful for the opportunity to collaborate with Precision Health on this important project, and we look forward to continuing this line of work.

Study published here: https://www.nature.com/articles/s41598-023-34559-0