Samah Fodeh-Jarad, PhD, associate professor of emergency medicine and of biostatistics (health informatics) and director of the Data Mining Lab, has received two major research grants to enhance health care for underrepresented populations through artificial intelligence (AI). Her work will focus on giving patients a stronger voice in their care.
In August, Fodeh-Jarad was awarded $1,049,990 by the Patient-Centered Outcomes Research Institute (PCORI) to develop PVminer, an AI tool designed to capture and analyze patient concerns and preferences in clinical narratives to assess their impact on mental health outcomes. Earlier, in May, she received an R01 award of about $3.5 million from the National Cancer Institute (NCI) for a project using EPPCminer, a machine learning tool, to study patient-provider communication in cancer care, leveraging messaging in the patient portals, and to assess its effects on outcomes such as adherence and hospitalizations.
Though focused on different health areas—mental health in the PCORI study and cancer in the NIH project—both studies share the goal of incorporating patient experiences into health care through AI. Each project will include national, multidisciplinary collaborations with investigators from Yale University, the Cleveland Clinic, Veterans Health Administration (VHA), and the Texas Association of Charitable Clinics (TXACC). Investigators bring expertise in design and analyses of observational studies using complex statistical modeling, as well as in the development and testing of novel technologies.
The studies will leverage data from diverse patient populations, including underserved groups like the uninsured and underinsured patients of TXACC member clinics, which serve a significant proportion of Hispanic and African American patients. Paula Walker, executive director of TXACC, expressed excitement about the partnership, which will provide de-identified patient messaging data to help train the AI tools. “An important part of our role will be to connect with clinics that will provide data with which to train the machine learning system. We will gather secure messages that the investigative team will help us de-identify. The messages will span across the diverse racial groups to ensure equity.”
At the Cleveland Clinic, from January to August 2023, more than 1.1 million messages were sent by patients using asynchronous messaging in the patient portal. More than 20% of patients using messaging are non-white, and among this group, more than half (52%) of patients using messaging were Black or African American. The study will develop PVminer to mine the patients’ voice representing their wants, needs, fears and preferences from big clinical data.
The NCI study will develop EPPCminer, a multi-class classifier based on a hybrid approach to improve the annotation of patient-provider communication categories, as well as evaluating EPPCminer by comparing its performance to existing models that use topic modeling or deep learning. “My interest in this type of research stems from my passion to better serve patients, address their needs and resolve barriers to seeking care,” says Fodeh-Jarad. “I embarked on this journey early in my career to capture the patients’ voice and translate it into structured format useful for broader data analytics”.
Arjun Venkatesh, MD, Chair of Yale’s Department of Emergency Medicine, highlighted Fodeh-Jarad’s expertise in language modeling and AI as key to advancing ethical research. Fodeh-Jarad aims to create tools that allow patients to express their needs, offering their personal perspectives, and ultimately improving patient-centered care. “I am confident that the project will lead to more grounded and ethical research for model development.” said Venkatesh. Fodeh-Jarad states that “The lived experience of patients and providers can embody subjectivities and randomness which will help developers create more ethical AI models that account for if not preserve space for human factors, outliers, and human agency and autonomy as the models ascribe predictions from large preexisting data sets.”