Understanding the Current State of AI in Medical Record Review

In a recent webinar, we delved into a pressing topic: the evolution of AI in medical record reviews. From the early days of keyword searches to today’s intelligent AI-driven insights, each step marks a critical advancement in how insurance and legal professionals interact with vast, complex medical data. I’d like to share some highlights from that discussion, focusing on how DigitalOwl’s Case Notes solution represents the powerful next evolution of AI for medical record analysis.

1. The Limitations of Keyword-Based Search

In the past, many systems relied heavily on basic keyword searches—underwriters or claims adjusters would often press Command+F or Control+F to find relevant terms in a document, like “accident” or “cardio.” While this allowed for quick jumps to specific mentions, it had significant drawbacks:

• Limited Scope: Keywords couldn’t capture the full context, and users often needed to jump from page to page to find relevant information.
• Complex Language Variations: Medical records include countless ways to phrase symptoms, diagnoses, and treatments, making it easy to miss critical details.
• Lack of Summary: There was no overarching narrative—just scattered mentions. This often left users to piece together the story themselves, adding time and complexity. Additionally, keywords often appeared with adjectives or modifiers that significantly altered their meaning, such as “negated for insomnia,” “suspected for insomnia,” or “family history of insomnia,” further complicating interpretation.

Despite these limitations, many systems today still rely on keyword search as a foundational tool. However, the next step in AI’s evolution offers a far more comprehensive approach.

2. Entity Extraction: A Step Up

Entity extraction took the use of artificial intelligence in medical reviews one step forward, using Natural Language Processing (NLP) to identify and list entities like conditions, medications, and procedures within documents. This was an improvement, as it provided adjusters with a structured list of medical terms. However, even entity extraction fell short in understanding the complete narrative of medical records:

• Lacks Contextualization: Simply knowing a list of conditions or medications doesn’t tell the story of how an incident or illness occurred.
• Uncodable Information: Some critical information cannot be represented as an entity because there is no corresponding ICD-10 or CPT code for it—for example, whether a driver was wearing a seat belt.

‍

In most current solutions, entity extraction forms the basis of machine learning summaries that deliver information in a list format but lack the depth needed for complex decision-making.

3. Generative Summaries: Adding Context and Narrative

The introduction of generative AI marked a transformative shift in medical record summaries by enabling not only data extraction but also the creation of cohesive narratives. This advancement allowed generative AI to summarize or even generate new, contextually relevant text based on existing summaries or new cases, addressing many of the limitations seen in previous approaches.

• Narrative Building: Generative summaries create a story, linking symptoms, treatments, and outcomes, allowing insurance and legal professionals to understand events like car accidents with details such as seatbelt usage, the reason for hospital visits, and treatment history.
• Greater Depth: This AI application provides a summary that connects events and offers a more complete picture, which is essential in understanding complex cases.

Generative summaries marked a major advancement in applying AI to medical records by adding a narrative layer to entity extraction. However, insurance and legal professionals still had to determine which information was relevant to each case and connect the dots between the data presented and the critical insights needed for decision-making.

4. Q&A and Chat Capabilities: Customized, On-Demand Insights

The next evolution of AI in medical reviews brought forward Q&A and chat functionalities, like DigitalOwl’s Chat, empowering users to ask specific questions and receive immediate, targeted answers tailored to their unique decision-making needs. Previously, summaries were vendor-driven documents containing details deemed important by the vendor at a chosen level of granularity—often helpful but not always sufficient. With these new capabilities, the control was now firmly in the client’s hands, enabling insurance and legal professionals to quickly access exactly the information they need, without limitations.

• Client-Centered Control: Instead of relying on a preset summary, users can ask tailored questions, such as “Was the claimant wearing a seatbelt during the accident?” or “Has the claimant complied with their sleep apnea treatment?”
• Cited Evidence: DigitalOwl’s Chat provides answers along with citations, offering full transparency and a clear trail back to the source document.

These Q&A and chat capabilities deliver precise information in a format users can trust, with click-to-evidence for immediate access to the source documents—offering an unmatched level of customization and transparency.

5. AI Agents: Beyond Summarization to Insights

Case Notes advances AI-powered medical record review by integrating AI agents—intelligent systems designed to tackle complex questions and workflows, delivering actionable insights without manual input. While chat-based Q&A is valuable, many questions in claims, underwriting and legal repeat themselves across similar cases, leading to unnecessary back-and-forth for adjusters, underwriters, and legal professionals.

What’s truly powerful is the next step: AI agents that can address complex tasks based on a clear agenda. Commonly seen in academia and AI research, these agents are now finding their way into enterprise solutions. DigitalOwl’s Case Notes uses AI agents specifically tailored to address real-world insurance and legal needs, effectively solving intricate problems for each unique use case.

• Intelligent Question Sequences: The AI agents that power Case Notes ask and answer critical questions that typically arise in claims analysis or underwriting without manual input, following adaptable, pre-defined workflows tailored to meet specific regulatory and operational needs. This approach saves valuable time by connecting the dots, conducting investigations, and presenting actionable insights. 
• Managing Broad Prompts: Some prompts—such as “underwrite the case”—are too broad to serve as a single chat request. While they may appear straightforward, they actually encompass a series of sub-questions and instructions. Unlike traditional chat-based interactions, AI agents break down these larger questions into manageable sub-questions, ensuring thorough and accurate responses.

• Problem Solving at Scale: The agents don’t just answer isolated questions—they synthesize data to offer insights like risk levels or severity ratings. For example, instead of manually tagging and assessing each data point, an AI agent could determine that a claimant’s untreated condition poses a high risk, alerting the user with an actionable recommendation.

These agents represent the future of AI in medical review, transforming workflows from linear, manual processes into intelligent, flexible decision aids that improve both speed and accuracy.

Ready to learn more about Case Notes? Register for our on-demand webinar to see a live demo and explore its capabilities firsthand.