Technology can be intimidating, especially when you do not fully understand how it works. There is also a fear that technology will start taking over and people will lose their jobs. I was one of those people who feared technology, and I kept hearing how artificial intelligence (AI) would take over health care jobs and replace certain roles. I was dead-set against AI being used in health care because I did not think AI would have the level of clinical expertise that is required. But I will be the first to admit I was wrong on so many levels.
Over the last several years of my nursing career, I have worked as a senior quality data abstractor, with a primary focus on psychiatric and stroke core measures, comprehensive stroke certification, and medical record auditing for a large regional health system here in Pennsylvania. As a quality data abstractor, I like to think of myself as a “medical record detective” or “medical record investigator,” where I review patient medical records to extract, validate, and analyze data based on specific guidelines from regulatory bodies like The Joint Commission (TJC), Department of Health (DOH), and the Centers for Medicare & Medicaid Services (CMS), as well as organizations like the American Heart Association (AHA), the American Academy of Orthopedic Surgeons (AAOS), and the American Joint Replacement Registry (AJRR). After abstracting the data, I would enter it into the appropriate clinical registry databases to support quality measurement, regulatory compliance, research, and performance improvement initiatives.
Data abstraction is a specialized role that often requires certification depending on the registry. It requires attention to detail, strong clinical judgment, and a deep understanding of medical documentation. It is also extremely time consuming. Every health care system documents differently. Information may be stored in different parts of the chart or documented inconsistently by providers. Finding the required data often meant combing through hundreds of pages of records. For example, some registry questions require capturing the earliest documented time for a clinical event. That might mean checking four or five different parts of the chart and comparing them before selecting the correct time. More complex questions might require reviewing multiple notes and piecing together the information to determine the correct answer. Until recently, the entire process was manual.
In August 2024, I was approached by a startup company with an interest in the registry world to talk about my registry experiences, what I liked and disliked about my role as a data abstractor, and whether there were things I would change. I remember my biggest complaint was redundancy and having to enter the same information multiple times in the registry, followed by not being able to find the required information because it was not filed correctly in the chart, or because the information in the chart did not add up to the diagnosis. That would require me to send the case back to coding for clarification and could ultimately lead to me having to re-abstract that case. These were just some of the things I called out. After that conversation, I did not think much about it.
Then in December 2024, they contacted me again. This time they wanted to offer me a position as a consultant, where I would provide registry guidance, guideline interpretations, and help with a registry portal that would incorporate AI. I was skeptical, excited, scared, and apprehensive all at once. I knew how hard my job was, how I had to compare data from several different sections of the chart, and I did not understand how AI would be able to do all of that and then determine the correct answer.
In April 2025, I was offered a part-time role as an adjudicator, where I would compare human responses to AI responses. To say I was still skeptical is an understatement. Even though the process was explained to me several times, I still was not sold. It was not until the platform was fully built out and I got to adjudicate my first case that I started to believe this might actually work.
I was comparing a case that was abstracted by a human against the same case that was abstracted by AI. The AI provided snippets showing where the information was found within the chart, and that alone was already a time saver for me. It showed me everything I would have spent time searching for in the chart, right there next to the question. How cool! I did not even need to look through the medical record because everything I needed was right there. The best part was that I had the choice to agree or disagree with AI, so I got the final say on the data I was entering into the registry.
I was not prepared for what happened next. AI captured something the human abstractor did not capture. At first, I just chalked it up to an abstraction error. We are humans. We make mistakes. But the more cases I adjudicated against AI, the more often I found AI was capturing data that humans were missing. I looked at the documentation AI was showing and thought to myself, hmmmm, I normally would not have checked that note, but AI was not wrong. This was getting interesting fast.
As I started chatting with abstractors using the platform, I noticed a common theme. Like me, they were discovering that AI was finding information they normally would not have found. Sometimes it was buried deep in progress notes or in a document they might not otherwise have reviewed. This was huge. I will admit it: I was starting to get really excited.
It was not until I started timing myself abstracting with and without AI that I realized just how helpful it was. I was cutting my time down by about 30 minutes using the AI platform. In addition to saving time, I was led directly to the notes I needed to see. It was so helpful that I started referring to AI as my favorite coworker, or my personal assistant. In November 2025, I was offered a full-time position and fully embraced using AI. I could not imagine going back to abstracting without my own little AI personal assistant. I love that it points me in the right direction and that I get final say over the data I enter into the registry. I think the time savings is probably the best thing overall, because it gives me more time to work on other projects instead of being bogged down for hours trying to find something buried deep within the chart.
It has now been over one year since I started this journey, and I will admit I was wrong. AI is not nearly as scary as I thought. AI will not take abstractors’ jobs. In fact, it is a great tool to help abstractors become more proficient and efficient. If you have the opportunity to give AI a chance, I encourage you to do so. It might just be the best coworker you ever had!
Brandy Sue Greif is a licensed practical nurse with more than 20 years of experience spanning clinical care, quality improvement, and clinical data operations. She brings deep expertise in data abstraction, regulatory reporting, and end-to-end medical record review, with a strong focus on accuracy, compliance, and performance improvement.
She currently serves as registry experience and training lead at Layer Health, where she partners across product, engineering, and machine learning teams to design and optimize AI-driven abstraction workflows. In this role, she helps translate complex clinical guidelines into scalable systems that maintain clinical integrity while improving operational efficiency.
Brandy has built and led training programs, standardized abstraction methodologies, and improved data quality across health systems. With extensive experience in Epic, Cerner, registry reporting, adjudication processes, and quality metrics, she is passionate about using AI to reduce administrative burden, improve data accuracy, and transform how clinical registries operate.
