Identifying health risks through AI and physiotypes

Read the research paper here.

Researchers at UF’s Intelligent Critical Care Center (IC3) recently published a study in PLoS Digital Health that showed their artificial intelligence research could potentially help hospitals improve healthcare decisions for patients by assessing a combination of vital signs during the first six hours of hospital admission.

In the study, IC3 researchers used AI to examine 75,762 admissions of 43,598 patients at UF Health Shands Hospital from 2014 to 2016. These patient admissions were examined for six vital signs over the first six hours of admission: systolic blood pressure, diastolic blood pressure, heart rate, temperature, oxygen saturation, and respiratory rate.

The IC3 researchers were able to classify patients into four physiotypes (categories of people with distinguishing features), and a patient’s physiotype gave clues about the chance of serious health outcomes such as acute kidney injury, blood clotting, sepsis, ICU admission, mechanical ventilation, and dialysis. The physiotypes also identified the likelihood of patient mortality in the short term (within 30 days) and long term (within 3 years).

Though previous studies had used unsupervised AI machine-learning analysis for patients with a common diagnosis (such as sepsis or acute respiratory distress syndrome), this IC3 study used this AI analysis for all patients admitted to a hospital for any reason.

What could this mean for patient care?

Depending on a patient’s physiotype, hospitals may better understand which patients would benefit from:

  • early ICU admission
  • early discharge or low-intensity care settings
  • more attention to post-discharge follow-up due to long-term health struggles

While this study only included patient data from UF Health and would need to be tested in a clinical trial, the IC3 study provided new assessments that did not simply give the same information as existing assessments. Because they could provide rapid insight into potential patient issues, these new assessments could be important for decision-making under time constraints.