AI Detection of Prescribing Cascades Research Presented at ACIAM
Research by Nihit Gurram on AI-guided detection of prescribing cascades in geriatric polypharmacy accepted for presentation at the American College of Artificial Intelligence and Medicine.
Press Release Disclaimer: This is a press release distributed through the XPR Media network. It has not been independently verified by our newsroom.
WICHITA, KS / ACCESS Newswire / September 8, 2026 / Research led by Nihit Gurram, MSHI, has been accepted for poster presentation at the 2nd Annual Scientific Session of the American College of Artificial Intelligence and Medicine, held October 9 and 10 at the McCormick Place Convention Center in Chicago. Gurram presents on October 9.
The poster, “From Prediction to Action: Clinician-Elicited Design Requirements for AI Cascade Detection in Geriatric Polypharmacy,” was co-authored with Evans D. Pope III, PharmD, MS, of the USC Mann School of Pharmacy and Pharmaceutical Sciences, and Christina Eldredge, MD, PhD, of the Morsani College of Medicine at the University of South Florida. Gurram is affiliated with the Kansas College of Osteopathic Medicine and the Morsani College of Medicine.
The study takes on prescribing cascades, which occur when an adverse drug reaction is mistaken for a new condition and treated with an additional medication. In older adults, that pattern contributes to falls, delirium, and hospitalization. Clinical decision support alerts built to catch these situations are frequently overridden because they fail to match how clinicians actually work. The research set out to understand that failure before building a tool rather than after.
Starting With Clinicians Instead of the Model
Rather than beginning with an algorithm, the team began with the people expected to use one. The researchers conducted six iterative semi-structured interviews across four clinical role classes: primary care physicians in internal medicine, family medicine, and geriatrics; specialist physicians in neurology and hospital medicine; clinical pharmacists; and clinical informatics faculty. The work included structured protocols and tailored questions on workflow insertion and success definition, validation of canonical cases against two synthetic patients, direct observation of pharmacist workflow in the electronic health record, and cross-role triangulation to separate the requirements clinicians agreed on from those they did not.
Six themes converged. Clinicians rejected numeric risk scores in favor of categorical tiers. They set a latency ceiling near five seconds and two clicks by role, and ten seconds at the bedside. They wanted three ranked risks rather than exhaustive lists, required alerts to be gated by chief complaint, and refused any alert that did not arrive with a substitutable alternative agent. Auditable reasoning was treated as a precondition for adoption rather than an enhancement. Physicians preferred to assess social determinants themselves while delegating renal, cardiac, and metabolic weighting to the tool.
One question did not resolve. Where in the workflow the alert should appear failed to converge across roles and appears to be setting dependent, a finding the authors report rather than smooth over.
A Design Changed Before It Was Built
The elicitation work altered the design. According to the abstract, conducting it before the build replaced a single-domain burden score with a comorbidity-weighted, setting-adaptive engine that returns an action rather than a prediction. Prospective multi-clinic validation follows.
“The failure mode in this space is rarely the algorithm,” said Gurram. “It is that a correct prediction arrives in a form a clinician cannot act on in the seconds they actually have. We asked clinicians what an alert would need to look like before we built one, and the answers reshaped the tool.”
About the Scientific Session
The American College of Artificial Intelligence and Medicine convenes clinicians, clinician-scientists, researchers, and health policy leaders around evidence-based applications of artificial intelligence in patient care. Keynote addresses at the 2026 session will be delivered by Mark Cohen, MD, Dean of the Carle Illinois College of Medicine, and James Whitfill, MD, Associate Chief Medical Officer for Clinical Artificial Intelligence at Radiology Partners. The activity is accredited for up to 15 AMA PRA Category 1 Credits through joint providership with the Illinois State Medical Society.
The full abstract is available on the conference site at event.fourwaves.com.
Gurram holds a Master of Science in Health Informatics and is a medical student at the Kansas
College of Osteopathic Medicine. He is the founder of Mosaic Health Solutions, a healthcare technology company incorporated as a Delaware C-corporation in 2026.
About Mosaic Health Solutions
Mosaic Health Solutions builds AI-powered clinical decision support that helps clinicians catch medication-related risk in older adults before it causes harm. By translating established clinical criteria into transparent, explainable tools at the point of care, the company focuses on the prescribing cascades, anticholinergic burden, and fall risk that disproportionately affect aging patients. Its mission is to make preventive medication safety practical for the clinicians and families who need it most. Learn more at www.mosaichealthsolutions.io.
Media Contact
Nihit Gurram
Founder, Mosaic Health Solutions
Email: nihit@mosaichealthsolutions.io
Phone: (510) 556-9938
SOURCE: Mosaic Health Solutions
View the original press release on ACCESS Newswire
