Dr. Paul A. Testa, chief health informatics officer, MCIT department of health informatics, at NYU Langone Health
Photo: NYU Langone Health
The most important health IT priority today is the responsible integration of artificial intelligence – particularly generative AI and autonomous clinical agents – into the fabric of patient care, governed by frameworks that center on trust, experience, safety, quality and equity.
So says Dr. Paul A. Testa, chief health informatics officer, MCIT department of health informatics, at NYU Langone Health. He's also an ER physician and a clinical professor of emergency medicine in the Ronald O. Perlman Department of Emergency Medicine at the NYU Grossman School of Medicine.
How to steward transformation
"We are at an inflection point: The technology is no longer aspirational, and the question has shifted from whether AI will transform healthcare to how we steward that transformation to serve patients and clinicians," he stated. "The foundational challenge is not technical but rather one of incentive alignment, governance and institutional readiness.
"AI deployed without a unified digital backbone, rigorous benchmarking and clear accountability structures risks amplifying the very disparities and inefficiencies it promises to mitigate," he continued. "At the same time, the opportunity is extraordinary. AI can compress the 'time to therapy' – that frustrating gap between diagnosis and effective treatment that exists not because clinicians lack knowledge, but because of friction and complexity."
AI can step into workforce gaps that will never be filled by hiring alone, from medication titration to a trusted clinical co-pilot – watchfully attentive and contextually informed, he added.
"And it can serve as a translator across data standards, making true interoperability less about rigid adherence to a single specification and more about the frictionless exchange of rich, actionable information," he explained. "But realizing these gains demands that we treat AI not as a standalone product but as something woven into an integrated ecosystem – one patient, one record, one experience – where every innovation is benchmarked for accuracy, monitored for bias and deployed only when it demonstrably improves outcomes.
"At the intersection of medicine, law and informatics, I have come to believe the most durable advances in health IT come not from chasing the newest technology, but from building the organizational discipline to deploy it safely and at scale, which might be the truest definition of innovation," he continued. "Healthcare already leverages the currency of narrative, but what changes when we bring AI into that narrative is the speed, scope and fidelity of the decisions this narrative can drive."
Keeping patients at the center
Safe deployment will require governance, oversight and leadership that keeps patients at the center and deeply respects clinicians' commitment to clinical mastery within the joyful practice of caring for others, he added.
"That means investing in primary care as the engine of interoperability, shifting from optimizing inputs to elevating the quality of outputs from the digital systems we have implemented, and empowering our clinical peers through structured engagement and education – like Prompt-a-Thons – so clinicians and patients are co-creators of AI systems," he said.
"The organizations that stand out as leaders are those that are committed to an integrated digital foundation, because you cannot scale responsible AI on a fragmented infrastructure," he continued. "At NYU Langone, it is not AI for its own sake – it is AI in service of the highest quality care. The institutions willing to do the hard, unglamorous work of governance and integration in service of the very best digital experience for their patients and clinicians are the ones that will earn and keep the trust this moment will engender."
For NYU Langone Health, responsible AI adoption is not a single project but an institutional commitment to infrastructure, clinical processes, the patient digital experience, workforce development and governance – all simultaneously.
"NYU Langone Health stood up UltraVioletAI, one of the very first private, secure, HIPAA-compliant generative AI environments, and made it available to our entire workforce to safely leverage these tools," Testa noted.
"Thousands of employees access such tools daily. But access without structure is just chaos, so we paired it with formal processes through our collaboration hub within our enterprise IT, connecting a multitude of collaborators from our clinical, research and educational missions, thereby creating a structured project pipeline that routes ideas from exploratory use to mentored development to real-world deployment.
A group effort for generative AI
In 2023, the organization ran the first Generative AI Prompt-a-Thon in healthcare to bring clinicians, researchers and educators together as co-creators of AI systems. Simultaneously, the health system is using AI to improve the quality and safety of clinical documentation itself: a combination of machine learning and large language models that provides physicians with actionable feedback on note quality, driving standardized documentation from less than 5% adoption to more than 75% across specialties, Testa reported.
"We have added ambient AI tools to the mix to bring joy to clinical engagement for physicians and nurses, first providing care and secondly manifesting the documentation of such care," he said. "We are pushing toward the frontier – building AI assistants for our home blood pressure monitoring program that, today, still have a human-in-the-loop for medication titrations, but we believe can operate semi-autonomously within well-defined clinical guardrails in the near future, thereby compressing the months-long 'time to therapy.'
"Similarly, we are deploying agents within the EHR that watch clinical reasoning unfold to guard against anchoring bias, nudging reconsideration of high-risk diagnoses if not already considered when clinical thresholds are reached in high-cognitive-load settings like the ED," he continued.
One effort Testa is particularly proud of that reflects the organization's approach is the "About Me" initiative within the EHR.
"The goal was to make patient voice and identity visible and actionable, not held only in narrative notes or dependent on individual clinicians' curiosity," he explained. "We leverage an SMS prompt to patients that invites them to share whatever they wish about their non-clinical lives – information that becomes visible to clinicians across encounters and supports the care team in connecting quickly, respectfully and empathically.
"That is interoperability at the human layer: It reliably carries personhood across handoffs," he continued. "Alongside that, we've pursued pragmatic, safety-oriented improvements in clinical communication – such as using AI to improve readability of patient-facing communications and to evaluate and improve note quality through structured feedback – because safer care depends on clearer documentation and clearer patient understanding, not just better 'data storage.'"
Digital access pathways
NYU Langone Health treats equity as a design constraint, not a retrospective metric, reinforcing why systems must intentionally design digital access pathways that do not widen existing gaps, he added.
"Through our informatics training programs, we are growing the next generation of informaticists with intentionality who understand that the hardest and most important work is not building the algorithm – it is earning and keeping the trust of the patients and clinicians who will live with its consequences," he said.
Testa has clear advice from the frontlines of AI for hospital and health system C-suite executives and other health IT leaders.
"C-suite executives and health IT leaders will succeed by treating AI as an organizational capacity driving transformation – one that demands the same level of institutional commitment, governance discipline and cultural investment that the best health systems brought to EHR adoption over the past two decades," he said.
"As I mentioned, we are at an opportune moment to shift from prioritizing inputs to clinically valuable outputs," he continued. "The first and most urgent imperative is to consolidate a unified digital infrastructure. You cannot scale responsible AI on a fragmented foundation of standalone apps, siloed data systems and disconnected vendor systems."
The organizations that will lead in this era are those that run on deeply integrated systems, enabling audacious learning from every patient at every encounter, he added.
"For those who haven't yet made that commitment, the time to act is now, because every AI tool you layer onto a fragmented infrastructure will compound complexity rather than reduce it," he concluded. "This is a stewardship moment, and we must lead with bold curiosity, vision and a disciplined commitment to the highest-quality care."
Follow Bill's health IT coverage on LinkedIn: Bill Siwicki
Email him: bsiwicki@himss.org
Healthcare IT News is a HIMSS Media publication.
WATCH NOW: The model context protocol at work in AI and clinical workflows
