Hopkins Students Undertake Data Visualization
When a child is diagnosed with pneumonia—the leading infectious cause of death in children worldwide—the treatment depends on whether the pneumonia is viral or bacterial, and tests to differentiate between the two are both expensive and invasive.
In an undergraduate course this spring, a team of students developed a web-based application that could help doctors predict the cause in an individual patient by comparing characteristics of that child’s health with a data set of 1,000 pediatric pneumonia patients. When the physician enters the child’s age, HIV status, and number of viruses and bacteria found in the patient’s nasal swab or blood sample, a scatterplot graph appears showing the causes of the disease for similar patients, allowing the doctor to infer the most probable cause of that patient’s pneumonia and tailor her treatment accordingly.
“It gives doctors quantitative data and evidence to add to their expertise,” says Rebecca Yates Coley, instructor for the course and a postdoctoral research fellow in the Department of Biostatistics at the Bloomberg School of Public Health.
The course, “Data Analysis and Visualization Practicum for Individualized Health,” introduced undergrads to two big ideas, Coley says. One idea is data visualization, or the art and science of graphically communicating statistics and research findings in meaningful ways. It’s an important part of individualized health, allowing doctors not only to tell patients about the specific nature of an illness and/or its expected progression, but to show them. “Data visualization is kind of an afterthought for lots of researchers, but it’s actually a skill with best practices,” Coley says.
The second idea is the technical component—in this case, the app that serves as interface for patients and physicians to enter data and get responses, which students devised using the open-source statistical software called R, along with app development software Shiny. The course was funded by a Narrative and Visualization Grant from Johns Hopkins’ Center for Educational Resources.
While one team of students built the pneumonia diagnosis app, other teams tackled a mental health app that predicts symptoms for patients with schizophrenia after several weeks on a particular medication, and a prostate cancer app to predict a patient’s outcome using a surveillance approach versus treatment.
“This was the class where I learned how to display data: by using art to graph the trends of data,” says Audrey Garman, a public health and history of science double major who just completed her junior year.
Addressing the challenges involved in creating a meaningful display of 1,000 data points gave Garman both a marketable skill and experience with an approach she believes may be a wave of the future in both health care and public health. “I think being able to figure out what variables have the greatest impact statistically will be able to shape where policy is headed and how we try to fix these issues,” she says.
Zhenke Wu (PhD ’14, biostatistics)—like Coley, a postdoctoral research fellow in biostatistics and visualization enthusiast—gave the pneumonia team the simulated data set based on his own research, and guided them as they analyzed it. Their final product succeeded in highlighting the information important to physicians seeking to individualize pneumonia treatment, he says. “Diagnosing a disease is not simple; it requires one to integrate multiple specimens of distinct clinical values, and to visualize all this information together in an effective way. The students did a good job of communicating the uncertainty,” Wu says.
Written by Rachel Wallach