Graphic of computer with medical icons

The data behind earlier diagnoses

They sound like the equivalent of medical crystal balls: computer programs that predict whether you’re at risk for a disease well before the first symptoms show up. But in the College of Health Sciences, Associate Professor Rohit Kate and Assistant Professor Jake Luo are researching and developing just such programs.

The secret lies in data science – especially the enormous pool of data contained in digital medical records. Kate and Luo, who specialize in health informatics and administration, use mathematical models that enable machine learning – a process in which computers spot patterns, make associations and fill in missing information.

College of Health Sciences Assistant Professor Jake Luo (left) and Associate Professor Rohit Kate are developing models and algorithms that predict health risks before the first symptoms show up. (UWM Photos/Troye Fox)

Kate’s work focuses on acute kidney injury, a sudden episode of kidney failure or damage. It’s a potentially fatal condition that can be treated if detected early, but it often occurs without symptoms or warning. For that reason, it’s most commonly detected in people who are already hospitalized.

In a partnership with Aurora Health Care, Kate’s model incorporates data variables found in hospital records, such as a patient’s demographics, medications, laboratory tests and comorbid conditions. This builds a risk profile, and the model’s goal is predicting which patients admitted for an unrelated health concern are also likely to suffer from acute kidney injury.

The beauty of this, Kate says, is that most diseases respond to treatment successfully if caught early, so his model could be useful in predicting other threatening conditions. Already, he’s adjusted the model to continually revise its predictions as a patient’s health changes during a hospital stay.

Luo, too, is leveraging data science for multiple health-related applications.

One focuses on people who participate in clinical drug studies. He’s created an algorithm that identifies their risk of experiencing severe adverse effects, including death. Moreover, by linking, extracting and structuring complex data, he can deliver intelligent analysis, which can help identify targets for potential new drugs.

Other models he’s created identify risk in people who experience bouts of dizziness. This could benefit, for example, people with vestibular disorder, an often-misdiagnosed dysfunction of the system in the ear and brain that controls balance.