One Biostatistician, Dozens of Research Questions
Linda Valeri was a master’s student in economics and social sciences at Bocconi University in Milan, Italy, when she landed an internship in India. Her mission: Use survey data to uncover the factors implicated in dropout rates among teen girls. Valeri found that a girl’s age at the start of puberty significantly limited her educational prospects—especially among those with longer walks from their rural homes to school. “I was struck by how interested I was in analyzing the data and what it could tell me about the intersection of health and social sciences,” says Valeri, whose own educational trajectory was profoundly altered by the experience. Four weeks after receiving her M.Sc. from Bocconi in economics and social sciences, Valeri enrolled at Harvard. She received her PhD in biostatistics in 2013.
Now an associate professor and PhD Program Director for the Columbia Mailman Department of Biostatistics, Valeri develops methods to reveal causal relationships among factors related to mental health, environmental determinants of health, and health disparities. She also collaborates widely; she’s coauthored studies that assess the use of text messages to promote healthier beverage choices, reveal how uncontrolled eczema exacerbates food allergies among some children with asthma, test ways to promote smoke-free compliance in public housing, and even uncover how Chilean teachers’ classroom style affects their students’ literacy development.
“The nice thing about biostatistics,” says Valeri, “is we can play in anyone else’s backyard.”
What brought you to Columbia Mailman in 2018?
Valeri: I received a five-year K01 award from the National Institute of Mental Health to develop novel methods to analyze digital health data for a psychiatric population. At the time, I held a joint appointment at Harvard Medical School and the affiliated McLean Hospital. This grant was giving me the opportunity to build my biostatistics lab while collaborating with Psychiatrists. It was time to find a new home in a Biostatistics department and I was really excited about Columbia Mailman’s connections to NewYork-Presbyterian’s Department of Psychiatry and the New York State Psychiatric Institute. I knew I could complete my proposal here. I reached out to Dubois Bowman, who was chair of Biostatistics at the time, and said I could cover 70 percent of my salary with the K01. The Department of Environmental Health Sciences, led by Andrea Baccarelli at that time, also supported my work for the first few years.
What spurred your interest in digital interventions for psychosis?
Valeri: Pharmacological solutions are not enough. We need complementary and behavioral strategies to promote treatment adherence, combat side effects, improve quality of life for affected people, and learn more about people’s social context and social behavior. And with better analytical tools, there’s so much digital data can reveal and great potential—improving screening, prevention, delivery of care, encouraging lifestyle changes in the moment. Maybe the tools we’re building will be used by people where the whole family shares one phone, or they have concerns about battery life. Working with NYSPI I want to include people of low-income, people of different racial populations. I want my work to be truly relevant for everyone.
You also work closely with Ana Navas-Acien, now the Leon Hess professor and chair of Environmental Health Sciences and director of the Columbia University Northern Plains Superfund Research Program. What fuels your collaboration?
Valeri: I love working with Ana. She has this expertise in metal exposure and cardiovascular health, and I come with expertise in how we can extract causal evidence from longitudinal observational data about complex pollutants. As a co-investigator on investigations in the Strong Heart Study and the Multiethnic Study of Atherosclerosis, I help her and her collaborators refine their scientific questions, improve their study design, and plan the analytic strategies to investigate the causal link between these pollutants and their impact across the lifespan on heart health and dementia.
How does machine learning affect your search for causal variables?
In public health, we’re dealing with very complex questions because of the confounding factors that we have to tease apart—how a particular variable among hundreds affects an outcome. Machine learning allows us to account for thousands of variables together, extracting the true causal variables from those in the background. In the case of estimating causal effects, it’s especially important to quantify uncertainty. Say you decide to take specific actions based on a causal finding—we need guarantees of how confident we are in these recommendations.
What value does AI have for your work?
AI can play a huge role in study design. Data gathering and organization suddenly becomes possible for huge systems. We can be comprehensive in our investigations. For example, I want to link data from multiple medical systems to investigate the impact of hormone replacement therapy on the incidence of dementia and understand whether menopause—when women experience specific symptoms for which there are treatments—could be a window to screen for cognitive changes in later life. That volume of data allows us to better characterize the patient population, rather than just focusing on a specific insurance code, but a human couldn’t compile it in a reasonable amount of time. With the resulting body of data, we can improve measurement quality and our statistical analyses can become much more detailed
