A recent Office of Inspector General (OIG) study revealed a startling statistic: U.S. hospitals are missing approximately half of all patient harm events. While this figure undoubtedly represents a failure, it isn’t a failure of intention or individual providers; it’s a failure of our systems, as they are set up today, to capture the full picture of safety risks that patients are facing.
At the core of this report lies a simple fact: If we can’t see harm happening, we can’t prevent it. While healthcare has made remarkable progress in reducing infections, preventing falls and minimizing medication errors in recent years, we’re still operating with a fundamentally incomplete understanding of the extent of patient harm.
The limits of human-dependent reporting
For decades, healthcare safety improvements have relied primarily on voluntary incident reporting systems. Providers identify safety events, document them and submit reports that feed into risk management databases.
In many ways, this approach has served us well, driving significant improvements in measurable areas of patient safety. But humans don’t—and can’t—report everything. We are restricted by time pressures, competing priorities and the inherent difficulty of recognizing patterns across thousands of data points and daily interactions. Because traditional medical coding systems capture only what providers consciously document, countless safety events are destined to slip through the cracks and remain buried in clinical notes and discharge summaries.
Some in healthcare safety have suggested that we abandon safety event reporting altogether, arguing that current systems are too ineffective to justify the investment. But this perspective misses a critical point: the problem here lies not in the act of reporting itself, but in how we’re doing it.
Why AI is safety reporting’s missing link
As AI continues to reshape the way we live, work, and process information, it offers two transformative capabilities that can revolutionize the way we report and address safety events.
First, it can help us find the harm we’re missing. By analyzing millions of clinical documents, AI can identify patterns and safety events in seconds that would be impossible with months of manual efforts. It can also scan unstructured data like nursing notes, narrative reports and clinical documents for subtle markers that might suggest complications, care delays or system failures that medical staff may not have been aware of, much less formally reported.
I’ve seen glimpses of this potential first-hand in my work with Patient Safety Organizations at Press Ganey. For example, when we examined written reports for disability-related markers (terms like “wheelchair,” “interpreter,” or “mobility aid”), we uncovered safety challenges uniquely facing patients with disabilities that had been previously invisible—not because the humans doing manual reporting were negligible, but because their capabilities and time were limited.
Second, AI can dramatically improve the efficacy of our existing reporting systems. Rather than abandoning safety event reporting altogether, AI presents an opportunity to transform these systems into intelligent, actionable devices to improve care at scale. Previously tedious tasks like drafting follow-up communications, clustering similar events, and prioritizing cases that demand immediate attention can be done in seconds rather than hours—and with incredible precision. In fact, innovative event reporting and learning systems are already starting to incorporate AI-powered capabilities to help safety teams work more efficiently and identify life-saving trends faster than ever before.
Healthcare leaders: The time to act is now
For leaders everywhere, the idea that hospitals are missing half of patient harm events should serve as a wake-up call—not to do away with safety reporting, but to enhance it. To reimagine it. To innovate until the systems we rely on are working for us, and for our patients. We cannot achieve our aspirations of Zero Harm without complete visibility into the safety risks our patients face.
Now that tools like AI are available to us, it is our responsibility to accelerate the deployment of systems built to supplement our blind spots—to detect unreported safety events, and to expand our ability to catch what human-dependent systems miss. This requires investment not only in technology, but in the infrastructure and training needed to empower teams to work with AI and act on the insights it brings to light.
On its own, no technology is enough to eliminate patient harm. Healthcare will always require judgment, empathy and commitment to continuous improvement that are inherently human. But it can amplify our human capabilities in extraordinary ways.
Safety is a prerequisite to providing comprehensive care. For the first time, we have the tools at our disposal to revolutionize the way we approach it. The question isn’t whether we can afford to invest in this change—it’s whether we can afford not to. The time for accepting incomplete safety data has passed. The time for AI-enabled, comprehensive patient protection is now.
Tejal Gandhi, M.D., is chief safety and transformation officer at Press Ganey.