Speak to an expert

Published

0 min read

Finding safety risks sooner: How AI expands the view of patient safety

Finding safety risks sooner: How AI expands the view of patient safety GettyImages 1363097745

JUMP TO

SHARE

By Mike Personett, Senior Vice President, Safety and Reliability Solutions, Press Ganey

Healthcare organizations generate vast amounts of safety data every day. Yet many safety events remain underdetected, and many analyses stop short of uncovering the system-level drivers behind them. Artificial intelligence can help close that gap by identifying more safety risks sooner.

But AI is not a replacement for clinical judgment, local context, or safety expertise. Its greatest value is in expanding what safety teams can see, so they can focus attention where learning and action are most needed.

Voluntary reporting is essential, but incomplete

Most safety event detection still relies on voluntary reporting, especially for failures in systems and processes that can’t be designed to self-detect every potential issue. But when it becomes the primary source of safety intelligence, organizations are left with an incomplete view of harm.

Other sources of safety intelligence include:

  • Experience comments and safety-related survey questions
  • Patient charts
  • Operational systems, such as lab, pharmacy, and billing
  • Huddle conversations
  • Rounding notes

These are well-known, but the volume and complexity of the data make them difficult to analyze consistently at scale. As a result, many organizations remain dependent on voluntary event reporting, even when other signals exist across the organization.

As AI tools evolve, safety teams have new ways to reduce that dependency and build a richer picture of harm.

AI can simplify voluntary reporting

Voluntary safety event reporting is often limited by the complexity of the reporting process itself. Many forms try to anticipate every possible type of failure. The result is often long dropdown menus, complex workflows, and categories that do not quite fit what the reporter observed.

Organizations want people to report unexpected events, but rigid reporting forms can make it difficult to select a classification that fits. When people cannot easily describe what happened, engagement suffers.

AI helps simplify this process. Instead of asking reporters to first navigate complex classification fields, organizations can let them describe what happened in their own words. AI then suggests classifications, identifying possible contributing factors, and routing the event for appropriate review—reducing form complexity without losing the structure safety teams need.

AI can also strengthen feedback loops by drafting notes of appreciation for event reporters and summarizing lessons learned for the broader team. Showing employees how their reports lead to action reinforces the organization’s commitment to reducing harm and encourages teams to remain vigilant in detecting and reporting safety concerns.

AI can expand chart review

Clinical notes, when aligned with structured data in the patient chart, can reveal potential safety concerns that may not appear in voluntary reports. Historically, this kind of review has depended on resource-intensive manual audits, which are often limited to small samples.

AI changes the scope of what’s possible. Instead of auditing fewer than 10% of relevant charts, AI can screen 100% of applicable records for potential safety concerns. For example, AI can compare available documentation against evidence-based practices and flagging records that warrant closer human review.

This creates a strong ROI opportunity: AI screens the charts, so humans don’t have to audit each one. The result is a less costly, less labor-intensive process and a broader view of potential harm.

Of course, comprehensive screening doesn’t eliminate the need for clinical judgment, but it changes where human expertise is applied: validating findings, investigating higher-risk cases, and acting on findings most likely to reveal preventable harm.

AI models informed by voluntary event reports can also create a continuously improving learning cycle. What people report helps train better detection models, and patterns identified in the chart can reduce dependence on human-reported events alone.

AI can listen to patient experience channels

Patient feedback can reveal safety concerns that may never enter the formal event reporting process. The challenge, however, is scale: The volume of experience comments, online reviews, and other unstructured feedback is too great for consistent manual review.

Keyword detection and sentiment analysis were early steps forward. Today, large language models can screen comments, detect safety-related concerns, classify themes, and identify which comments may require follow-up based on severity or risk.

This is where connected safety and experience systems matter. Press Ganey’s text analytics tools can analyze unstructured patient and consumer text to spot safety concerns and route relevant comments into the High Reliability Platform for review, learning, and follow-up.

Detection is only the starting point

Improving detection paints a more complete picture of safety risks across the organization. But more accurate detection must go hand in hand with more effective safety learning systems that help teams understand what happened, why it happened, and how to prevent it from happening again.

The next opportunity is to move from finding more signals to learning from them at scale. And that’s the focus of the next post in this series: how AI can help safety teams review events more consistently, prioritize attention, and turn reporting into learning.

Stay tuned for the second article in our three-part series: “From event reporting to safety learning with AI.” To discuss how AI fits into your safety risk detection strategy, reach out to our team.