The Collective “Blog” of Health IT
A Companion Blog to The Collective Voice of Health IT Podcast
The Missing Pattern: How Data and AI Could Transform Women’s Health
For women living with endometriosis, getting a diagnosis can take an average of seven years. For these women struggling with their disease, the issue isn’t that their medical information isn’t available, it’s that the health care system often struggles to connect the information it already has.
In a recent episode (#261) of The Collective Voice of Health IT, WEDI’s Michael McNutt spoke with Dr. Niki Panich, a family physician, Chief Medical Officer and Stanford-trained data scientist with nearly two decades of clinical experience focused on women's health. Their conversation explored how fragmented clinical data, interoperability challenges and artificial intelligence (AI) could help reveal patterns that are often difficult to see across years of medical records.
A health care system built for transactions, not the whole patient
Panich's path from medicine to data science grew out of a desire to better understand how data could improve patient care. Her work has included developing machine learning models to predict which patients may benefit from particular therapies and incorporating information, such as food, exercise, and environmental factors, into understanding a patient's health.
But the current health care infrastructure presents a fundamental challenge: it wasn't designed to see the patient as a whole. Much of today's infrastructure was built for discrete transactions, e.g., visits, claims, tests and individual encounters, rather than following a patient's complete clinical journey. Each encounter may generate useful information, but that information can remain fragmented across different providers, specialties, and technology systems. When no one has access to the complete picture, important patterns can be difficult to recognize.
When the pieces of the story don't connect
Panich described examples in which fragmented information contributed to important clinical information being missed, including a hemoglobin result that was overlooked before surgery and a missed cancer diagnosis. These examples illustrate an important distinction between having data somewhere in the health care system and having the same data available when it matters. While the industry has made significant progress in electronic health information exchange, more is required to make sure pieces of information between systems can be connected into a meaningful clinical story.
The value—and challenge—of clinical notes
Another challenge is that not all valuable clinical information fits neatly into a structured field. Panich noted that free-text documentation can contain important context about a patient that may not be captured in discrete data elements. At the same time, free text can be difficult to index, retrieve, and analyze across large numbers of records.
AI-assisted documentation can improve the quality and completeness of information captured during a visit, reducing some of the dependence on how much a physician can type while simultaneously interacting with a patient. But if information remains fragmented across multiple physicians and organizations, an excellent note from today’s visit may still fail to connect with information generated previously by another provider.
Where AI could change the equation
Panich sees the potential for AI to address the data fragmentation comes through its ability to process enormous amounts of information and identify relationships and patterns. In women's health, that could be particularly meaningful for conditions that involve recurring symptoms, multiple specialties, and long diagnostic journeys. Instead of considering each symptom or visit in isolation, AI could help surface the broader trajectory of a patient's health. But AI isn't automatically the answer.
Better AI requires better—and more diverse—data
Panich emphasized the importance of training AI models on diverse and comprehensive datasets. Medical research and data do not necessarily represent the diversity of patients. If those limitations are carried into AI models, the technology can perpetuate existing disparities rather than address them. Mechanisms are needed to identify and detect biases as AI models are developed and deployed.
Clinicians also need “glass box” traceability, which Panich defines as the ability to understand why an AI system reached a particular conclusion. For clinicians to trust an AI recommendation, they need insight into the factors that contributed to that answer.
Keeping humans in the loop
For all the potential of AI, Panich doesn't envision a future in which technology replaces the human relationship at the center of medicine. There are elements of clinical care that cannot simply be extracted from a data set, such as patient's fears, preferences, and personal goals that can influence what treatment makes sense. AI can help clinicians analyze information and identify patterns but the final decisions about goals and treatment plans still require collaboration between the clinician and the patient.
From fragmented information to a complete clinical story
The challenge facing women's health, and health care more broadly, is not necessarily a lack of data. It is the inability to consistently connect the data we already have. For someone living with endometriosis, years of symptoms and visits may contain the clues needed to recognize a pattern. But if those clues remain scattered across notes, providers, specialties, and organizations, the pattern, and diagnosis, can remain hidden. AI, interoperability, and technology do not eliminate the need for clinical judgment. The goal is to move beyond treating each encounter as an isolated transaction and begin to recognize the broader clinical story.
Listen to The Collective Voice of Health: A WEDI podcast at www.wedi.org/category/podcasts