How Much Clinical Data Is Actually Unstructured?
An estimated 80 percent of clinical information in electronic health records is unstructured, meaning it lives in physician notes, operative reports, discharge summaries, and similar free text rather than in coded, structured fields. The figure originates with a widely cited estimate from 3M Health Information Systems and has since been repeated across health IT research and clinical trial analysis. A 2025 peer-reviewed methodology study notes that the figure has not been validated at scale. Still, the broader point remains: a substantial amount of clinically meaningful information lives outside structured fields.
Any AI strategy built primarily around structured data is missing a meaningful share of the available clinical picture.
Unstructured Clinical Data Still Requires Manual Review
Information trapped in free text still has to make it into registries, quality reporting programs, accreditation submissions, and research, and today that mostly happens through manual review. A trained reviewer reads the note, finds the relevant detail, and enters it where it needs to go.
This work is necessary because coded fields alone tend to miss nuance. A diagnosis, a complication, or a patient’s response to treatment is often documented in narrative form long before, or instead of, a corresponding code. Skipping the narrative means skipping real clinical detail.
Manual Review Can’t Scale with Growing Data Volumes
Healthcare data volume keeps growing, and the amount of documentation each reviewer has to work through keeps climbing with it. A few factors are driving that growth at once:
- Expanding quality reporting requirements tied to value-based care
- Growth in registry participation across specialties
- Increasing documentation volume per patient encounter
- Rising accreditation and research data requests
Manual review scales linearly. Twice the data means roughly twice the reviewer hours needed to keep up. Documentation volume is growing faster than most abstraction teams can add staff, which puts a hard ceiling on how far manual-only review can scale.
AI Is Redefining Clinical Data Abstraction
Clinical documentation is genuinely hard to automate end to end. A few realities explain why:
- Patient histories have gaps where care was delayed or spread across providers who don’t share systems
- Records arrive as faxes, scanned images, and files in formats that were never designed to talk to each other, sometimes scanned in sideways or upside down
- Every health system, and often every facility within one, charts differently
That last point is one that MRO’s own team has said directly. As VP of Software Engineering, Michael Cordell, put it in a recent interview: “When you’ve seen one EMR, you’ve seen one EMR.”
That kind of variation is exactly why the most effective AI systems in healthcare are built to work alongside clinical experts rather than instead of them. The technology doesn’t need to solve every edge case on its own. It needs to handle the volume and surface what matters, while a clinical expert makes the final call. That combination is where the real gains are, and it’s a much more solvable problem than trying to build an algorithm that never needs a second opinion.
Closing the Gap Between Manual and Automated Abstraction
AI can take on the volume, evidence discovery, and pattern recognition required to process unstructured clinical data at scale, while a purpose-built workflow helps ensure results meet the quality standards healthcare demands.
This ability to turn complex documentation into usable clinical data is where healthcare AI is delivering meaningful value. Organizations no longer have to choose between speed and accuracy. With the right technology and workflow, both can improve together.
Putting the Model into Practice with Prodigy™
Prodigy™, MRO’s AI engine for clinical data abstraction, puts this approach into practice by combining AI-powered technology with clinical expertise. Prodigy helps health systems identify eligible cases and automate the registry abstraction workflow, while keeping clinical judgment and oversight where they matter most.
The result is a more scalable approach to abstraction that helps health systems manage growing data volumes without choosing between efficiency and data quality. Contact us to learn more today.
Frequently Asked Questions About Unstructured Clinical Data
What Is Unstructured Clinical Data?
Unstructured clinical data is information recorded in free text form rather than coded fields, including physician notes, operative reports, discharge summaries, and scanned documents. Industry estimates commonly put this at around 80 percent of the information in a typical electronic health record, though the exact figure varies by source and hasn’t been validated at large scale in peer-reviewed research.
Why Can’t AI Fully Automate Clinical Data Abstraction?
Clinical documentation contains gaps, inconsistent formatting, and variation across health systems that make full automation on its own unreliable. The strongest results come from AI handling volume and pattern recognition while clinical experts apply judgment and make the final call.
How Is Unstructured Clinical Data Used Today?
Health systems primarily rely on trained reviewers to read unstructured notes and enter relevant details into registries, quality reporting systems, and research datasets. It’s an accurate approach, but each reviewer can only process a limited volume, which becomes a constraint as documentation grows.