Q&A: Predictive model using EHR data finds ‘good candidates’ for gastric cancer screening
Click Here to Manage Email Alerts
Key takeaways:
- Older age, male sex and Black or Asian race were associated with a higher likelihood of gastric cancer.
- Additional risk factors included tobacco use, anemia and pernicious anemia.
A new electronic health record-based model accurately predicted risk for gastric cancer using demographic, behavioral and clinical data and may be leveraged to increase screening in at-risk populations.
“We were particularly compelled to look at gastric cancer because it seems tailor made for screening,” Michelle Kang Kim, MD, PhD, chair of the department of gastroenterology, hepatology and nutrition at Cleveland Clinic, told Healio. “For instance, we screen for colon cancer effectively with colonoscopy and other available tests. Gastric cancer is also something that can be screened for in certain high-risk populations. This is an opportunity to actually diagnose patients earlier and to potentially offer them a cure as opposed to treating a more advanced cancer.”
She continued, “In addition to the fact that a lot of our patients with gastric cancer tend to be members of underserved and minority populations, we felt very strongly that this was a cancer worth investigating and worth developing a screening program for.”
In a retrospective, case-control study, Kim and colleagues investigated the feasibility and performance of an EHR-based predictive model using data from 614 patients (median age, 68 years; 58% men; 65% white) diagnosed with noncardia gastric cancer at Cleveland Clinic between 2010 and 2021. They compared demographic, behavioral and medical history in a 1:10 ratio with healthy controls and used area under the curve and the 0.632 estimator for discrimination.
According to results published in Gastro Hep Advances, a higher probability of noncardia gastric cancer was associated with older age (OR = 1.16; 95% CI, 1.04-1.3), male sex (OR = 1.97; 95% CI, 1.64-2.36) and Black (OR = 3.07; 95% CI, 2.46-3.83) or Asian (OR = 4.39; 95% CI, 2.6-7.42) race. Additional risk factors included tobacco use (OR = 1.61; 95% CI, 1.34-1.94), anemia (OR = 1.35; 95% CI, 1.09-1.68) and pernicious anemia (OR = 6.12; 95% CI, 3.42-10.95).
In a Healio interview exclusive, Kim explains the importance of identifying and screening those at-risk for gastric cancer as well as the implications an EHR-based predictive model may have on GI cancer care going forward.
Healio: What is the importance of screening for and identifying gastric cancer early?
Kim: There are certain countries in the world — for instance, Japan and Korea — where gastric cancer is so common that they screen regularly, similar to the way we screen for colon cancer. In those countries, they are able to cure gastric cancer in a very high number of cases and their patients do phenomenally well.
In contrast, most U.S. patients are diagnosed when they have advanced and often metastatic disease. Our 5-year survival is in the range of 32% overall and around 7% for metastatic disease, whereas in East Asia it is as high as 95% for localized cancers.
Other countries are identifying at-risk populations, so we need to find a way to do it here.
Healio: How did EHRs inform this model?
Kim: In order to identify an at-risk population, you have to have some general ground rules. Risk factors that we already know about include things like male sex, certain races and ethnicities, smokers and Helicobacter pylori infection. When we were thinking about how to identify an at-risk population, we realized that a lot of these basic factors are in the EHR that could be used in an effective way.
In addition, the immense amount of data in the EHR, especially with AI models, is a very ripe resource for doing research.
Healio: Tell us more about the development of the model and the results.
Kim: First, we looked at all of the variables that exist in the EHR, including demographic, behavioral, family history, medical history and laboratory parameters.
We were not even sure if this was going to work, so I think one of the things we were reassured by was that the usual risk factors you see with predicting gastric cancer played out here: age, specifically between 40 to 80 years; male sex; family history; and smoking. Other features that are a little bit harder include immigrant status and certain races and ethnicities. There were some laboratory parameters that did come out as significant.
One thing I will say is that while our model performed well at Cleveland Clinic Ohio and Florida, we do not know if this is something that is more generalizable to the rest of the country. What is much more important is that the concept of what we’re doing here is feasible and not only developed in one center but validated in another.
Healio: What key points should readers take away from this study?
Kim: First of all, gastric cancer itself is very much able to be diagnosed early and is curable if diagnosed early. The second is that this concept of having a model derived from any EHR may be used to identify patients who are at risk.
The ultimate question is whether something like this works, not just at our medical center but across the broader U.S., and whether we can use an intervention like this to identify patients who could be good candidates for gastric cancer screening.
Lastly, I do suspect that there will be a change: In the next 5 years, we will be doing gastric cancer screening in the U.S.
Healio: What do these results add to management of patients going forward?
Kim: This is something that we’re doing for gastric cancer, but we actually foresee doing a similar model for other types of cancers. The other thing is that this is also a platform for other diseases that perhaps are not very common but are very devastating in terms of morbidity and mortality.
If there are certain risk factors that could be identified in the EHR for some of these other diseases, perhaps this is something that we can use more widely.
Collapse
Read more about
Click Here to Manage Email Alerts