Health care outcomes often fall short not because doctors and patients lack good treatments or information, but rather because the systems in which they make decisions are poorly designed, according to an NEJM Catalyst report co-authored by LDI Senior Fellow Kevin Volpp, MD, PhD, a professor at both the Perelman School of Medicine and the Wharton School.

Beyond AI Diagnostics

Among other things, the paper, Reengineering Health Decision-Making Environments — Aligning Research, Policy, and Innovation for Societal Benefit, calls for greater emphasis on the use of AI to improve the decision-making environments in which patients and clinicians act, rather than focusing on AI primarily as a stand-alone diagnostic or predictive tool.

A central point of the Catalyst report is that patients and clinicians often make complex health care decisions while confronting information overload, cognitive burdens, time and administrative pressures, poorly timed information, and seemingly minor obstacles that can discourage better choices. The researchers argue that redesigning this “decision environment” could improve care without necessarily requiring new drugs, treatments, or technologies.

The “environment” doesn’t primarily refer to a physical environment such as a hospital, clinic, or examination room. It means the decision-making architecture surrounding both clinician and patient and the authors note that architecture has five features:

• What information is provided and when

• How choices are framed

• What the default course of action is

• What obstacles or “frictions” accompany alternative choices

• What feedback mechanisms allow people to learn from previous decisions

An example the authors point to involves prescription drug prices. Earlier price-transparency efforts made cost information available through websites and patient portals, requiring patients to seek it out themselves outside the moment when prescribing decisions were being made. A redesigned approach instead puts patient-specific cost information and recommendations for lower-cost, clinically appropriate alternatives in front of physicians at the moment they are writing the prescription. In one randomized trial, the approach reduced patients’ out-of-pocket costs and improved prescription fill rates.

Electronic Health Records

Similarly, an electronic health record constitutes part of a clinician’s decision environment. Changing its defaults — for example, making generic prescribing the default choice — alters the environment in which the physician makes the decision without changing the underlying medical knowledge available to that physician.

The Catalyst report synthesizes the findings and discussions of the Social Science Research Council’s (SSRC) June 2025 symposium of the same title at New York University. That SSRC event brought together 31 top researchers from 18 universities, research organizations, health systems and government agencies, including six Leonard Davis Institute of Health Economics Senior Fellows from the University of Pennsylvania.

The Catalyst paper focuses on three approaches to creating better decision environments: behavioral science, learning health systems, and artificial intelligence. Behavioral science can change how choices are presented; learning health systems can continually test different approaches to see what works; and AI can help deliver the right information to patients or clinicians at the moment they need it.

Penn Medicine’s Nudge Unit

Researchers advocate turning health systems into laboratories for continuous improvement. Rather than introducing a new procedure and assuming it works, hospitals can build randomized experiments into everyday operations and quickly determine which reminders, electronic-record defaults, screening approaches, or other interventions actually improve care. Penn Medicine’s Nudge Unit is cited for using these techniques to influence generic prescribing, colorectal cancer screening, opioid prescribing, statin prescribing, and cardiac rehabilitation referrals.

Beyond its use as a diagnostic, AI is highlighted for its potential for assisting in decision redesign rather than simply predicting what will happen. For instance, the authors suggest, AI could help emergency physicians identify patients who can safely receive rapid outpatient care instead of being admitted to the hospital or help health systems match patients with the appropriate specialist based on urgency, clinical need, and appointment availability.

Equity Concerns

However, the researchers caution that such innovations could increase health care disparities if sophisticated, well-funded health systems benefit while safety-net hospitals and poorer communities lack the money and technology to implement them. Even behavioral “nudges” may work better among people already inclined to take healthy actions. They call for equity needs to be built into these systems from the beginning rather than addressed afterward, as often is the current practice.

They also underscore another potential challenge — the regulatory environment, payment systems, and infrastructure may not be structured to support the types of innovation that reengineering decision environments requires.

“Current payment models create an asymmetry: High-cost treatments are reimbursed on the basis of clinical benefit, while delivery system innovations, including decision environment redesign, are typically expected to demonstrate cost savings as a condition of support. This asymmetry structurally disadvantages investments in decision environments even when they contribute clinical benefits with a high degree of cost-effectiveness,” they write.

---