When Andrew Sheldrick was 8 years old, he died because of a medication error.
Andrew had epilepsy. Because he was too young to swallow pills, his medication had to be specially compounded. A preventable mistake in that process led to a fatal overdose. His mother has spent years telling his story, not because reliving it gets easier, but because she knows: We cannot fix what we refuse to see.
Today, as healthcare races to embrace artificial intelligence, Andrew’s story should serve as a warning.
AI has the potential to help identify dangerous medication interactions, flag deteriorating patients earlier, detect diagnostic errors, and reduce administrative burdens that pull clinicians away from patient care. But AI is only as effective as the data it learns from.
The Harm Caused By Data Gatekeeping
For years, healthcare has treated safety data as something to be guarded rather than shared. Hospitals worry about liability. Health systems worry about reputational damage. Vendors protect the very data their products are purchased for. Meanwhile, patients and families get harmed at an alarming rate 1 out of 4 hospitalized patients get harmed, according to the 2022 report of Office of inspectorGeneral.
The result is a fragmented system in which lessons that could save lives are often confined to individual institutions and data generated by products used in patient care becomes trapped within.
Imagine if aviation operated this way.
Every time a plane experienced a near miss, equipment malfunction, or safety failure, airlines would keep the information private, preventing other carriers from learning from the incident. Or what if the system watching the wings, engines, radar didn’t allow their data to be shared with the overall flight system? The industry would never have achieved the extraordinary safety record passengers now take for granted.
Healthcare should be no different.
AI is Only As Good as its Data Source
The stakes are enormous. Research published by Johns Hopkins University estimated that medical errors contribute to more than 250,000 deaths annually in the United States, making them one of the leading causes of death. Other researchers debate the precise ranking and methodology, but there is little disagreement about the underlying reality: preventable patient harm remains a major public health crisis.
At the same time, diagnostic errors alone are estimated to cause serious harm to roughly 795,000 Americans each year, according to a 2024 cross-sectional analysis. These are precisely the kinds of failures AI could help identify earlier, but only if the systems have access to comprehensive, accurate, and transparent data.
That is where the conversation often breaks down.
Many discussions about healthcare AI focus on algorithms. They focus on computing power. They focus on venture funding and market opportunities.
Far fewer focus on the quality, completeness, and accessibility of the underlying data.
An AI system cannot identify patterns in events that were never collected. It cannot learn from adverse outcomes hidden behind legal settlements. It cannot help prevent future tragedies if institutions are unwilling to share what went wrong. It cannot prevent a serious event if data from the EMR or medical devices are walled off from it.
Transparency is not a technical issue. It is a cultural one.
The healthcare industry has long operated within a framework that unintentionally discourages openness. Clinicians fear blame. Organizations fear scrutiny. EMR and other medical devices treat the patient data as their gold mine and way too many companies are not willing to share it.
Yet without the data from the machines that monitor and provide therapy to the patient, and without data from every hidden mistake represents a missed opportunity to protect the next patient.
This is why the future of healthcare AI must be built on a foundation of radical transparency.
Transparency Must Be the Standard
Hospitals should openly share patient safety data in standardized formats and even report them on their website. Near misses should be treated as valuable learning opportunities rather than embarrassing secrets. Technology companies should share their data and purchasing departments must insist on that with every one of their vendors. Companies developing clinical AI tools should demonstrate how their models are trained, validated, and monitored for bias. Regulators should encourage responsible data sharing while protecting patient privacy.
Most importantly, patients and families deserve access to the information generated about their own care.
Trust cannot exist without transparency. Trust must be verified.
Effective AI needs transparency to be trusted.
Zero Preventable Deaths is Achievable If we Work Together
Through my personal and professional work, I have long advocated for the sharing of data by my peers in the medical technology industry and safety data by my peers in the hospital and research industry because without the flow of data, AI can’t help, nothing can. Through initiatives such as Patient Safety Movement Foundation’s Open Data Pledge, healthcare organizations have an opportunity to move beyond isolated improvement efforts and contribute to collective learning.
The President’s Council of Advisors on Science and Technology (PCAST) report, A Transformational Effort on Patient Safety, recognizes that transparency requires an environment in which clinicians and healthcare workers can report errors and near misses without fear of inappropriate punishment. The Report promotes a “just culture” that supports candid reporting, learning, and improvement. PCAST frames transparency not merely as public disclosure but also as internal organizational openness regarding adverse events, risks, and safety concerns.
AI makes this need more urgent, not less.
The Path Forward
We stand at a pivotal moment. The technology now exists to help clinicians detect deterioration earlier, improve diagnoses, reduce medication errors, and identify risks before they become tragedies. Properly implemented, these tools could save countless lives.
But algorithms cannot compensate for information that is hidden, incomplete, or deliberately withheld.
Andrew Sheldrick’s story should not simply be remembered as a heartbreaking failure. It should be remembered as a call to action.
The question facing healthcare is not whether AI can help save lives.
The question is whether we are willing to give it the transparency and veracity it needs to do so.
Because the future of patient safety will not be determined by artificial intelligence alone.
It will be determined by our willingness to tell and share the truth.
