Statistical and deep learning methods for real-world clinical data — Yale BIDS & Pediatrics.
The Loza Lab focuses on development of statistical and deep learning methods to leverage Real-World Data to improve clinical care with a focus on multimodal medical foundation models.
Instructor of Biomedical Informatics and Data Science
Department of Biomedical Informatics and Data Science and Department of Pediatrics, Yale School of Medicine, VA Connecticut Healthcare System
Andrew Loza, MD, PhD, recently joined Yale BIDS as an Instructor of Biomedical Informatics and Data Science. His research centers on applying statistical and deep learning methods to improve patient care, particularly by enhancing predictive models and refining data collection processes. Dr. Loza also emphasizes how his experience at the VA Connecticut Healthcare System, combined with the resources at Yale BIDS, will further advance his work.
Andrew Loza is a physician-scientist whose research focuses on predictive analytics and population health. He received his PhD from Washington University in St. Louis in biophysics studying mechanisms of collective cell migration using time lapse microscopy, computer vision, and simulation. He completed his MD degree at the Yale University School of Medicine and residency in Internal Medicine – Pediatrics also at Yale. After residency, he completed a Clinical Informatics fellowship in the ACGME Yale/VA program and a postdoctoral fellowship in the VA Biomedical Informatics program.
Q
What is your primary academic focus or research specialty, and how did your background in biophysics, internal medicine, and clinical informatics shape your research direction in healthcare delivery?
A
My primary focus is in developing statistical and deep learning methods to better understand patients’ health trajectories and clinical care delivery. I think that each step of my background has emphasized that as systems get more complex, the history and trajectory matter more and more compared to a single snapshot. Human health is complex, and to make the right diagnosis and deliver the best care requires knowing not just the current vitals or symptoms, but knowing the whole patient narrative. My background in biophysics and clinical informatics have provided a toolkit to translate these ideas into a mathematical framework.
Q
What long-term goals do you have in biomedical informatics and data science, particularly in the context of healthcare delivery and patient outcomes?
A
My long-term goals are to enhance our ability to transform the data we collect into actionable information, to improve what data we collect, and to improve the software infrastructure we use to deliver this information to physicians and patients. Predictive models are a great intersection of all these ideas – a successful prediction model relies on the right inputs, the right model, and the right channels of delivery to ensure they are integrated into, rather than interrupting workflow.
Q
How do you think the resources and collaborative environment at Yale BIDS, as well as your experience with the VA Connecticut Healthcare System, will help you achieve your goals in using transformer-based generative models for healthcare?
A
Three things come to mind. First, creativity: our best ideas often do not emerge fully formed, but grow and develop through conversations. Having great colleagues can accelerate this process. Second scale: the ability to work with data between Yale and the VA can provide insights into practice patterns and trends beyond what a single system might show and the scale of the data can be used to reveal patterns that would otherwise be missed. And last logistics: the best ideas remain just words on a page without the ability to translate them into action. Whether this is understanding the data dictionaries within massive databases or scaling up an intervention across multiple sites, logistical knowledge is critical. This knowledge often doesn’t live in a book or come as the answer to a well-crafted prompt, but requires a group of talented individuals.
Q
If you could meet any scientist, living or deceased, who would it be and why?
A
I would have to pick John von Neumann. I think that few individuals have had such impact across the breadth of fields from pure mathematics to economics to nuclear and quantum physics to chemical engineering to computing. As examples, he founded the mathematical field of game theory, developed the merge sort algorithm, provided critical insights for the Manhattan project, and developed the first computational climate models. To hear his thoughts on the current state of computing and deep learning would be fascinating, not to mention that his ideas would probably accelerate the field by 10 years if I could write them down.
Fellows, students, postdocs, and staff working on clinical AI, EHR modeling, and foundation models.
Postdoctoral Associate
Huan Li is a Postdoctoral Associate in the Loza Lab, holding a PhD in Biomedical Informatics. Her research leverages machine learning and foundation models to enhance electronic health records (EHR), optimize physician productivity, and advance clinical decision support. She is deeply committed to translating complex informatics methodologies into practical, scalable tools that seamlessly integrate into real-world clinical workflows.
Clinical Informatics Fellow · Emergency and Palliative Care Physician
Kent McCann, MD, is a Clinical Informatics Fellow in the Loza Lab and an active emergency and palliative care physician. Bridging the gap between bedside care and computational research, his work focuses on clinical data tokenization and mapping care representations to patient outcomes. He actively applies models like multivariateGPT (mvGPT) to real-world clinical sequences, pioneering trajectory-aware modeling directly informed by EHR and related data.
PhD Student, Computational Biology & Biomedical Informatics
Erica Stutz is a second-year PhD student in Computational Biology and Biomedical Informatics in the Loza Lab. She specializes in engineering conditional attribute and target-guided transformer architectures, aligning token-level predictions with broader sequence-level outcomes. Drawing on applications in reinforcement learning and controlled language generation, she is currently extending these frameworks to clinical sequences, enabling patient outcomes to directly inform treatment-oriented predictive models.
PhD Student, Computational Biology & Biomedical Informatics
Giacomo Marino is a first-year PhD student in Computational Biology and Biomedical Informatics in the Loza Lab. His research focuses on developing multi-modal drug representation frameworks for medical foundation models. By advancing the tokenization of small molecules, he aims to bridge complex chemical structures with the diverse clinical signals these models are designed to understand.
PhD Student, Computational Biology & Biomedical Informatics
Tahamid Siam is a first-year PhD student in the Loza Lab. His research centers on the tokenization of high-frequency sensor data, such as EKG and EEG recordings, for integration into medical foundation models. He is passionate about developing robust data representations that fuse continuous physiological signals with broader clinical and textual contexts, enabling a more holistic view of patient trajectories.
Postgraduate Associate
Ikgyu "Tom" Shin is a Postgraduate Associate in the Loza Lab, bringing a strong foundation in statistics and health informatics. His research targets the tokenization of electronic health record narratives to train and refine medical foundation models. Tom is dedicated to building scalable computational pipelines that transform unstructured clinical text into actionable models, ensuring they align flawlessly with day-to-day patient care.
Interested in collaborating or joining the lab? Reach out to Dr. Loza.