AI-based tool aims to improve genetic testing referrals in advanced prostate cancer

Physicians at the University of Iowa (UI) Health Care System were recently awarded a grant to support a quality improvement initiative focused on identifying patients with prostate cancer who may be eligible for germline genetic testing.¹

Although clinical guidelines clearly define which patients should be referred for testing, national referral rates remain low. One key barrier is the challenge of reviewing increasingly complex electronic health records, which may contain hundreds of notes, to determine whether a patient meets eligibility criteria. To address this gap, the UI Health team developed an artificial intelligence (AI)–based tool that integrates with the electronic health record to analyze patients’ charts and flag those with relevant disease characteristics or family histories that may warrant genetic testing.

In a recent interview with Urology Times®, principal investigator Kenneth G. Nepple, MD, discussed how this technology could streamline clinical workflows and improve access to high-value genetic testing for patients and their families. Nepple is an associate chief health information officer and a urologic oncologist at UI Health Care.

Urology Times: For some background, why does genetic testing remain underutilized in men with advanced prostate cancer, despite guideline recommendations?

Nepple: That's a great question. At the base of it is that, as providers, everyone gets very busy. As you're seeing patients, it's hard to completely check every single box for each individual patient, just because of the volume of information and [the time constraints]. What we've learned about prostate cancer over time is that there is a hereditary driving component. Depending on the research you read, anywhere from 1% to 10% of patients may have a germline genetic mutation that predisposes them to prostate cancer. The guidelines have developed over time to recommend seeing a genetic counselor and having germline genetic testing done to look for risk in specific populations. One of the challenges is identifying those patients in those specific populations.

For prostate cancer, [this includes] patients that either have specific disease characteristics or a family history. From the disease characteristics standpoint, [this includes] patients with high risk or very high risk prostate cancer, including lymph node involvement or if the cancer's growing outside the capsule of the prostate or into the seminal vesicles, or patients with metastatic disease. This is a good proportion of our patients, but not everyone. On the family history side, it involves not just understanding if there is a family history, but also the severity. If there is a first degree or even a second degree family member that has that history of a similar type of aggressive prostate cancer or any prostate cancer mortality, then they also meet criteria.

In clinical practice, most providers are familiar with this. However, if you look at national data, the rate of patients getting testing when they meet criteria is only about 15%. Our genetic counselor group looked back at a year's worth of patients that get care here at the University of Iowa and found that our rate was similar at only about 15%. So, we [viewed this as an] opportunity to improve the workflow for both patients and providers to be able to offer what can be this high value testing.

Urology Times: What led to the development of this specific project?

Nepple: The project was funded by the American Urological Association in combination with Pfizer and Astellas. They were interested in quality improvement projects, which are projects that make care better for patients. We had some discussions with the genetic counselors about the benefits of genetic testing. For patients with metastatic prostate cancer, it could have an impact on the medication that may be chosen. There's specific BRCA mutations that can make people eligible for a specific group of medications called PARP inhibitors. That's generally the context we've been thinking about with genetic testing. However, when we talked to our genetic counselors, even talked with colleagues that take care of patients with breast cancer, what we identified is that it's also important for patients’ family members. In the past, I've asked patients, “Do you have brothers? Do you have any sons?” But importantly, the most common mutation in prostate cancer is a BRCA mutation, which has a predisposition to prostate cancer, but also to breast cancer and ovarian cancer. So, it potentially impacts not only sons, but daughters also. One of the reasons to consider doing genetic testing is if that mutation is identified, then family members can have additional testing called cascade testing. The health of those family members could potentially be impacted if they have a mutation identified. For me, it was a huge part of having the initial discussions to realize that an appropriate discussion in this space isn't just about the patient's prostate cancer care, but also specifically about their family members and the impact it could have on those family members.

Urology Times: How is this project is being conducted? What some of the ultimate goals are for this work?

Nepple: We [talked about what we could] do in the system to make things better. We got together to talk about this as a group, reviewed the data that's out there, shared the national guidelines in this space, and did all of the standard stuff. We realized that there's a lot of discussion about AI, but not a lot that's embedded within the actual chart that exists when we're seeing patients. What these tools allow is what we call AI chart summary, where they can review the entire medical record, including the typical part of the electronic health record, but also information exchanges with outside institutions, including the VA Healthcare System, plus what we call fax records, or scanned records. Sometimes these can be buried in parts of the chart that the clinician is just not looking at on a regular basis. The tool is able to look at all this information and then provide a summary.

The initial iteration was an overall summary, but then over time, we had access to some advanced tools, specifically a conversational AI interface where you can create a chat with the electronic health record and ask a question to be able to look at specific information. For example, "Does this patient meet criteria for advanced prostate cancer?" "Have they or have they not had germline genetic testing performed?" There is also a feature that is allowing the provider to save custom summaries. I generated a prompt that has those questions, and then the system, pretty quickly, typically within 5 to 10 seconds, can give an output that gives you the information, but very importantly, it's sourced or referenced. It will say, "Hey, this patient has advanced prostate cancer—which we defined for the system—based on this pathology report." One click away is the pathology report. Or "They have had genetic testing; here's the report from that." What it allows is, rather than being dependent on finding somebody who's an expert in coding within the electronic health record, as an individual provider, I could generate these prompts and then within the patient's chart I can see what the output is. It removes one of the potential barriers to quality improvement, which is finding the right people that can do the coding in the chart.

We had early access to these tools, and we evaluated them to assess the accuracy of the outputs. Over months and many evaluations of these tools, we saw that they were performing effectively. The idea is that we want this information without having to spend a huge amount of time to go and look for it.

The other part we're developing is at the institutional level. We are trying to select the patients that have advanced prostate cancer and haven't had genetic testing and create a cohort that we can potentially reach out to via our communication portal and say, "We're trying to pay attention to genetics and hereditary cancer risk and prostate cancer. It looks like you may meet criteria, but you haven't had testing. Is this something you'd be interested in?" The portal allows also allows you to share things like a video or supplementary information. That's been our two-pronged approach: at the individual patient level, we have a tool that works in the system quickly and accurately to give providers the information for that patient, but then from a bigger system level, we can use the technology to make us a better health care system and to improve quality by identifying what can potentially be a little bit of a blind spot for these patients in terms of their risk.

Urology Times: How do you envision this tool changing the day-to-day workflow?

Nepple: One important concept is how big the chart is. If we're going to develop a tool to pull out patient-specific cancer characteristics, genetic testing information, and family history, you have to first have an accurate assessment of that entire body of the chart. We've done some initial work that we're writing up now. We presented it at AMIA, which is a national informatics conference. In the study, we looked at all of the adult urology patient encounters that we had for a year, which was about 16,000 encounters, and put all of those charts into this AI tool to see what data was there. It’s potentially one of the first real-world evaluations of the chart in that way. There's been some work done for research purposes or pulling information into a research database, but what we were able to do with that evaluation is get an idea of how big the electronic medical record of the chart is for a specific patient.

When we looked at our cohort of patients, the average patient had about 700 notes in their chart. The average note was about 300 words per note. That may seem small, but it could be small telephone notes or nursing notes, but also bigger notes. If we do the math, for the average patient, their notes in their chart had about 180,000 words. [Just for comparison,] that ends up being about a 950-page book. So, if you say that a provider seeing a patient is somewhat responsible for being able to look through this chart and find the right page with the right information, that's extremely challenging, even if you're a speed reader. What we’re developing is a technology that can point us towards the right page of this book where that information is available. We’re also still using physicians to their maximum capabilities; now that we have this information about a patient, we can let the patient and the provider decide what to do with that data.

When I thought about the project to start with, it was quality improvement for advanced prostate cancer in this genetic space, but it really ends up being quality improvement for our whole electronic health record. We’re taking a system that isn't optimized for patient care, and we’re developing a tool within it that improves things. If you do the overall math, for our year of adult patients, when we look at those 16,000 patients, that translates to 3.1 billion words in the charts. What does that look like from a practical sense? This tool is taking those billions of words that are consumed by our physicians and advanced practice providers and making them more easily usable. From a practical side, by adding these tools in, it's freed up time to have some of these important discussions. Rather than looking through the chart for 10 minutes to find the appropriate information, maybe I can find that information in 3 minutes. That opens up 7 more minutes to talk with that patient about specifics related to their care, about survivorship, or even about their family history and their interest in doing germline genetic testing and meeting with our genetic counselors. You're freeing up provider time by improving the technology that's embedded within this software that we're using every day.

Urology Times: What are some of the next steps for this project?

Nepple: I think the cool part about this project and having some support to work on the project has been the opportunity to talk to patients about this functionality. The owner of the chart is the patient; that's their information. Involving them in assessing the accuracy of the tool has been very interesting. We're getting patients involved with a future iteration of this project, which includes having patients review these summaries to let us know if they're accurate. We've just started doing that in clinic a few days ago. I said, "Hey, this is the output, what do you think about this?" and I received some patient feedback on that. That's an area we're exploring, to have this focused on the patient level.

The other big part of it has been developing a team. We have our team of urologists, medical oncologists, radiation oncologists, genetic counselors, and our informatics teams interacting with the different vendors in this space, including not just the main electronic health record, but also the AI chart summary tool from Evidently. We also recently turned on a clinical matching software from a different company called Triomics. We have these tools in place that will be spreading in the future to other places, but we have very early access and are developing a collaborative effort to do that.

Urology Times: Is there anything else that you’d like to add?

Nepple: I think the other part that urologists in general are going to be interested in is the concept of medical education in this space. We’re having discussions with our urology residents about what the right time is to introduce AI tools. We want to use them in a way that we're not stunting the knowledge growth of trainees. If you had an AI tool that always did something for you, then you're probably not going to learn to do that as well on your own. The reality is that this AI functionality is not going away. The more we understand about how to use it and the limitations, that's going to impact current care, care in the near future, and also the future generations that are learning how to take care of patients.

People should feel free to reach out. Just like everything else in urologic care, it’s amazing to be able to share what has worked well and what hasn't worked as well. It’s a cool concept to say that urology providers are going to potentially be leading the development of AI just like we have led the way with the development of robotic surgery and other types of technologies.

REFERENCE

1. UI Health Care awarded grant for AI-enhanced clinical research project aimed at improving advanced prostate cancer care. News release. UI Health Care. November 7, 2025. Accessed December 22, 2025. https://urology.medicine.uiowa.edu/news/2025/11/ui-health-care-awarded-grant-ai-enhanced-clinical-research-project-aimed-improving