Precision Health member Thomas Valley, MD, MSc, is an assistant professor in the Division of Pulmonary and Critical Care Medicine. As a critical care physician, he focuses on the decision making that takes place to determine whether or not a patient needs to be admitted to the intensive are unit (ICU). Predictive models developed through machine learning can aid a clinician in making such decisions, where time is of the essence.

Valley is currently part of a team developing a predictive model for sepsis by examining data from the electronic health record. Sepsis occurs when the body has an extreme response to an underlying infection, and treating sepsis once it is detected constitutes a time-sensitive medical emergency. More than 2 million hospitalizations in the US each year have sepsis complications, of which almost 10% result in death. Early detection and rapid treatment with fluid and antibiotics are imperative for positive outcomes. As Sepsis Awareness Month comes to a close, we asked Valley about his work to advance early prediction and treatment of this life-threatening condition, as well as his other research projects.

What are the most challenging aspects of predicting and treating sepsis, and how will your model address these?
Sepsis is challenging on many fronts, but in particular it can be difficult to identify early–and this is really important, because delayed treatment for sepsis leads to worse outcomes. Our hope is that these predictive models can help clinicians at Michigan Medicine and beyond identify and treat sepsis earlier.

What is innovative about this model?
We’re using innovative computer science techniques, some of which we’ve used previously, like the MCURES model, to create homegrown models of identifying sepsis at Michigan Medicine.

How it will benefit patients and clinicians?
Having better strategies to identify patients with sepsis earlier can help clinicians provide better care for patients with sepsis.

You were a guest speaker at Precision Health’s participant event on September 14—can you talk about the subject of your presentation (MCURES)? Why do you think it’s important to communicate with research participants?
I was very excited about the opportunity to engage with research participants, because they were key to the development of MCURES. MCURES is a homegrown model that was developed to better identify both “sick” and “not sick” patients early in the pandemic. Both of these categories were very important at that time, because we needed to identify patients who were sick so they could receive ICU-level care, but we also needed to identify patients who were not sick so that we could expedite discharge processes when the hospital was under tremendous strain.

What are your research interests, broadly?
My research focuses on ensuring that care delivered in the intensive care unit is high-quality, particularly in underserved rural and minority communities.

How does your work apply to the field of precision health?
Precision health focuses on ensuring care is individualized for patients, and this is particularly important for groups of patients who may receive lower quality of care because they receive their care in hospitals that have fewer resources for example. The tools that we’re working on in Precision Health can improve quality of care broadly, make clinicians’ lives easier, and improve outcomes of patients.

Selected Publications:

• Racial Bias in Pulse Oximetry Measurement
• Origins of Racial and Ethnic Bias in Pulmonary Technologies
• Evaluating a Widely Implemented Proprietary Deterioration Index Model among Hospitalized Patients with COVID-19
  (also see this feature)
• A Phenome-Wide Association Study (PheWAS) of COVID-19 Outcomes by Race Using the Electronic Health Records Data in Michigan Medicine
• Changes in Self-Rated Health After Sepsis in Older Adults: A Retrospective Cohort Study
• The epidemiology of sepsis: questioning our understanding of the role of race