UTA researcher receives NIH grant to advance predictive disease models

Suvra Pal, an associate professor of statistics in The University of Texas at Arlington's Department of Mathematics, has been awarded a $1.8 million grant from the National Institutes of Health to develop advanced predictive models designed to improve disease treatment and potential cures.

These models could potentially transform how doctors treat cancer and other serious illnesses.

Funded by the National Institute of General Medical Sciences, the five-year project aims to improve the accuracy of predicting whether a patient is likely to be clinically cured-particularly when the disease is detected early-by using cutting-edge statistical methods and artificial intelligence, including machine learning.

Using these techniques, researchers analyze large sets of patient data to identify patterns and trends that aren't obvious to the human eye. By training algorithms to recognize which factors are linked to long-term survival or cure, the models can offer more personalized and accurate predictions for patients.

Traditionally, models have focused on survival outcomes, but they haven't been able to predict an actual cure. Our models aim to do both: estimate the probability that a patient will be cured and, if not, predict their long-term survival."

Suvra Pal, associate professor of statistics, The University of Texas at Arlington's Department of Mathematics

By incorporating complex biological factors-like the presence of malignant cells even when they can't be directly observed-Pal's models simulate disease progression and treatment outcomes using what are known as latent variables.

Latent variables are hidden factors that can't be measured directly but affect things that are observable. For example, while doctors might not be able to see every cancer cell, these hidden cells influence test results and patient symptoms. By including latent variables, models can better capture what's really happening inside the body, even when some details are invisible. These models can handle high-dimensional data, including tens of thousands of patient biomarkers, genetic data and clinical features. The goal is to isolate the most predictive features to guide treatment decisions more precisely.

"In many cases, treatments come with serious side effects," Pal said. "If our models can more accurately predict that a patient is likely to be cured without further therapy, we can spare them from unnecessary and potentially harmful treatments. Conversely, if the current models overestimate the cure rate, we can intervene earlier and more effectively."

Pal described this work as a "passion project."

"It's the kind of research that, if successful, could have a real, lasting impact on how we predict, treat and understand complex diseases."