I’ve spent more than 30 years in nursing, and in recent years my work has shifted to something most people never think about: clinical data abstraction. My job is to review the medical record and extract the pieces of information that clinical registries rely on for research, quality reporting, regulatory requirements, and everyday decision-making. It’s specialized work that depends on clinical judgment, a deep understanding of clinical documentation in the EHR, and a commitment to getting every detail right.

Abstraction is complicated and requires clinical expertise. The day-to-day work can be challenging and arduous. Information lives in many different places in the patient’s chart and often in multiple systems, not just the EHR. Clinical documentation also varies from one clinician to another. One patient stay can produce more documentation than most people realize, each piece with its own wording and structure. My role is to make sense of it all and ensure the data we report to the registry bodies reflects the real story of the patient.

But, almost all of this process is still manual. Abstractors spend a large chunk of their day searching for specific details buried in documentation, checking them, and checking them again. It’s steady, concentrated work. You can’t rush it because a missed value or a misinterpreted line in a note can change the meaning of the data. Even with years of experience, it’s easy to lose time to the volume of information in a single case.

When our team announced we’d be using artificial intelligence (AI) from Carta Healthcare and their Lighthouse solution to support abstraction, I didn’t know what to expect. I had built my own habits and methods over many years. I trusted them. Introducing a new tool into a workflow that depends so heavily on judgment made me curious, but also a little uneasy.

Could AI really understand the way clinicians think and document? Would it notice the small details? How accurate would it be? I was about to find out.

Early apprehension

Those first few weeks, I double-checked everything the AI suggested, which only added to my workload. With manual abstraction, it took about 30 minutes to complete a straightforward case, and complex cases could take five hours or more.

While I was initially skeptical that AI could replace my 30+ years of clinical expertise, I decided to keep working with the system and share my feedback. I joined sessions with the Carta Healthcare product team, who asked thoughtful questions and made quick improvements. Over time, I began to see progress. The solution captured data correctly, pulled over clinical data, and identified relevant notes, labs, and occurrences with increasing consistency and precision. It was learning. The turning point came when I realized Lighthouse was finding details I may have missed, such as a lab value buried deep in a note or a discharge medication mentioned in documentation but omitted from the summary. That was when my skepticism began to shift toward trust.

Embracing a new workflow

As I became more comfortable, my workflow changed completely. Instead of manually searching through dozens or even hundreds of documents in the EHR, I could review Lighthouse-generated answers to registry questions, linked to their source notes and clear justifications why the AI answered a specific question in a specific way. Clicking through the summary view at the start of each case felt like receiving a concise patient report that displayed medical and surgical history, key labs, and procedures in one place. From there, I could validate what the system found rather than manually hunting for data. This change cut my per-case abstraction dramatically. For average cases, what once took over 30 minutes now takes 15 to 22 minutes. For complex cases, which used to take five hours or more, the time has been reduced to about 90 minutes.

The difference became especially clear one evening when an IT network issue prevented me from accessing Epic and Lighthouse. I still had eight cases due that night and had to finish them manually. I thought it would be easy since that was how I used to work. But I quickly realized how much I depended on Lighthouse’s efficiency and accuracy. Without the system’s fast answers to registry questions, I had to return to reading through lengthy progress notes and comparing lab results across multiple days. That night, I understood how much time and effort the AI had been saving me.

Earning my trust

Today, I trust the Lighthouse AI output completely. For straightforward day-surgery or overnight-stay patients, I accept its results with confidence. For longer, more complex stays involving multiple physicians and numerous occurrences, I still review the discharge summaries and verify that all conditions are captured correctly. That human check remains essential, but the system’s precision gives me a strong foundation. My inter-rater reliability (IRR) scores have actually improved, confirming the AI technology’s accuracy.

AI-enabled abstraction has transformed how I work. The technology performs the initial review, surfacing relevant data from structured and unstructured fields and showing exactly where it found each value. This allows me to spend less time searching and more time validating and confirming.

The impact goes well beyond efficiency. For health systems that manage thousands of registry cases each year, AI-enabled abstraction leads to measurable improvements in both speed and consistency, while also reducing our abstraction costs.

I have also come to see how this technology strengthens, rather than replaces, human expertise. In health care, clinical judgment and experience will always matter. There are still moments when something in a chart just doesn’t feel right, and that instinct only comes from years of experience. No system can do that part for you.

What AI does well is clear away the routine work so I can focus on the details that require human insight. It automates repetitive work, flags potential inconsistencies, and allows me to apply my decades of experience and knowledge where it matters most. On a personal level, it also keeps me connected to patient care even though I stepped away from bedside nursing years ago.

What I’ve learned … so far

The success of AI in data abstraction depends on collaboration between people and technology. The system can move through large amounts of data in seconds, but human oversight ensures that context and clinical reasoning remain part of the process. When you put the technology and human review together, you get what many describe as “hybrid intelligence”: a blend of advanced AI and clinical expertise that leads to reliable, high-quality data.

For abstractors who are beginning this journey, my advice is simple: Give the technology a fair chance. Continue using your clinical instincts to confirm accuracy, but allow the system to show what it can do. Once you experience how efficiently it identifies the right data and how reliably it supports your work, you will see its value. My own journey began with skepticism, but it has led to genuine trust in a tool that now makes my work faster, more consistent, and more rewarding.

Pamela Ashenfelter is a nurse and clinical data abstractor.