Tag: healthcare

The year is 2025, or more than fifteen years since the enaction of The HITECH Act and Meaningful Use.  Almost all of the clinical data recorded from ordinary Americans’ physician office visits and hospital stays have now shifted to electronic format.  Therefore, increasing emphasis is gradually being placed on the value of real-world data, with the hope that medical knowledge resulting in care improvements can be extracted from the vast amount of information that exists in electronic health records.

Structured vs. Unstructured Clinical Data 

This electronic clinical data can be subdivided into two categories – structured and unstructured.  Examples of structured data include demographic information, diagnoses and procedures (in the form of clinical codes), medication prescription information, insurance records, and vital signs, while examples of unstructured data include free-text clinical narratives and imaging and test reports. Both types of clinical information are important and perform complementary functions in real-world data.  Structured data contains many basic data elements and is traditionally easier to process, due to its tabular nature.  However, unstructured data has been estimated to comprise 80% of clinical data by volume and often provides insights that are absent from structured clinical data and claims data [1].  There is an old saying among medical professionals that “90% of diagnoses can be made using the patient history, and 10% using the physical exam [2]” (notably, both elements are virtually absent from structured EHR data). What are some of the details captured in the well-written clinical note that are typically excluded from structured EHR data?

Information Extraction from Unstructured Data in an LLM-World 

A well-written clinical note contains many details about the patient that are absent from both structured tabular data and claims data.  Until just a couple of years ago, the challenge was extracting information from a clinical note into a usable format.  However, with the advent of large language models (LLMs), one can present a note as context and ask a favorite LLM questions about the note, such as “Where is the location of this patient’s pain?” or “Why did the patient discontinue lisinopril?”  Adaptation of this method enables extraction of information from the note as structured categorical data, which can then be used as structured data.

OMNY Notes: A First-of-its-Kind Clinical Notes Data Product  

OMNY Notes is one of our exciting new data products that makes billions of de-identified clinical notes from diverse health systems available to the end-user.   Researchers no longer must rely solely on structured EHR and claims data; they can now view the full patient journey with our HIPAA-compliant de-identified linked structured EHR, claims and notes solutions representing more than 75M individuals. No other solution available today provides the combined depth, breadth, and scale of OMNY structured and unstructured data to support improving quality, safety, and efficiency of healthcare delivery and overall public health. Contact us at info@omnyhealth.com to learn more about our OMNY Notes product and our other data products:

• OMNY Foundation
• OMNY Linked Claims
• OMNY MedTech

References:

[1] https://healthtechmagazine.net/article/2023/05/structured-vs-unstructured-data-in-healthcare-perfcon.

[2] Tsukamoto, Tomoko, et al. “The contribution of the medical history for the diagnosis of simulated cases by medical students.” International Journal of Medical Education 3 (2012).

OMNY Health’s recent study is shedding light on the role of cognitive impairment in patients diagnosed with Alzheimer’s disease.   Alzheimer’s disease (AD) is the most frequent type of neurodegenerative disease that develops over several years. It is characterized by multiple cognitive deficits that progress over time, including memory deterioration. Newly approved monoclonal antibodies, unlike traditional medications that primarily relieve symptoms, have been shown to slow the progression of Alzheimer’s disease by approximately 30%.

At OMNY Health, researchers recently completed a retrospective analysis (2017 to 2024) of electronic health records from over 150,000 patients that had received Alzheimer’s disease care in the United States.  The study aimed to characterize the cognitive state of patients and to describe their treatment regimens.  OMNY’s researchers focused on nearly 7,000 Alzheimer’s disease patients in the dataset who had either the Montreal Cognitive Assessment (MoCA) or the Mini-Mental State Exam (MMSE) scores available.  

Key Findings  

The findings provide insight into the cognitive state of patients as they are first observed within the health system and receiving a diagnosis of Alzheimer’s disease.   On average patients were 78 years of age at the time of their first observed diagnosis.  Following the first observed diagnosis of Alzheimer’s, patients were categorized into four groups based on their first reported cognitive scores: normal, mild, moderate, and severe.  

The distribution of cognitive severity among patients was as follows: 

• Normal: 15%  
• Mild: 37%  
• Moderate: 35%  
• Severe: 13% 

OMNY’s research also found that with increasing cognitive impairment (normal, mild, moderate, severe), there was a monotonic rise in the proportions of female patients (51%, 60%, 64%, 67%) and nonwhite patients (15%, 17%, 22%, 27%). 

The study additionally assessed treatment patterns within 30 days of diagnosis. Many patients (59%) received cholinesterase inhibitors or NMDA receptor antagonists—therapies aimed at symptom management—with usage rates consistent across impairment levels. Fewer than 1% of patients had received treatment with newer monoclonal antibodies.

Why It Matters  

As disease-modifying therapies become more accessible, OMNY’s insights can support the research community in evaluating treatment effectiveness, informing clinical decision-making, and advancing understanding of disease progression across patient populations.

An OMNY Health Study on Severity Measure Extraction in Respiratory Care

 

Understanding disease severity is essential to supporting treatment decisions, yet many critical severity metrics—especially in respiratory conditions—are often buried in unstructured EHR notes. At ISPOR 2025, OMNY Health presented findings on how large language models (LLMs) can accurately extract FEV1/FVC scores, a core pulmonary function test (PFT) metric, directly from free-text clinical notes.

Why It Matters 

Clinical measures like FEV1/FVC are pivotal in evaluating lung function and diagnosing COPD and asthma. However, these measures are not consistently captured in structured EHR fields, making them difficult to access at scale. OMNY Health’s study explored whether retrieval-augmented LLMs could help close that gap—automating severity extraction while preserving accuracy. 

Study Overview

OMNY Health researchers sampled 50 random note excerpts containing the phrase “FEV1/FVC” from the OMNY Health platform and categorized them as:  

• Simple (S): One FEV1/FVC score present 
• No Value (NV): No score provided 
• Complex (C): Multiple scores present 

OMNY Health researchers tested two Gemini LLMs (Flash and Pro) using a structured prompt and evaluated:

• Accuracy in correct extraction 
• Hallucination rate (false outputs when no score existed) 
• Latency (processing time in slot milliseconds) 

Key Findings 

1. High Accuracy in Simple Notes 
Flash extracted FEV1/FVC scores with 90% accuracy, outperforming Pro (73.3%).

2. No Hallucinations in NV Notes  
Neither model generated phantom data when no score was present.

3. Pro Recognized Complex Context Better 
While Flash returned one of the multiple scores, Pro acknowledged the complexity—useful for nuanced documentation.

4. Flash Was Faster 
Flash processed data using 6,210 slot milliseconds versus Pro’s 23,576—offering speed advantages for scale. 

Example Outputs

Performance Snapshot

What’s Next 

This study highlights how LLMs can reliably extract clinical severity measures from real-world EHR data. Moving forward, the team plans to evaluate performance across more complex note structures, investigate how interpretability and robustness vary by model, and assess the trade-offs between cost, speed, and depth of output. The long-term vision is to expand this approach to other therapeutic areas where structured fields fall short—enabling deeper, more scalable evidence generation.