Marshall Health, which has more than 420 providers in 75 areas of primary and specialty care throughout southern West Virginia, Ohio and Kentucky, has embraced artificial intelligence (AI) for early diabetic retinopathy detection in patients with diabetes.
Using the LumineticsCore AI system (formerly IDx-DR), Marshall Health collaborates with diabetes patients and their primary care providers to identify signs of diabetic eye disease, offering timely intervention to prevent vision loss.
Diabetic retinopathy, a significant diabetes complication leading to blindness, necessitates regular eye exams. Individuals with diabetes, including type 1, type 2 and gestational, are at risk of developing diabetic retinopathy.
While not a replacement for comprehensive eye exams, the LumineticsCore exam offers a convenient option for diabetes patients during their primary care visits.
Using a specialized fundus camera, the system quickly captures retinal images. AI software then analyzes the images for diabetic retinopathy signs, generating immediate diagnostic reports for in-office discussions between providers and patients.
The potential of AI in predicting disease was demonstrated in a study conducted by Geisinger and an AI model developer, Medial EarlySign, published in 2022 by NEJM Catalyst.
The collaboration hinged on the analysis of a significant patient cohort—25,610 individuals—who were overdue for Colorectal Cancer (CRC) screening.
Utilizing a sophisticated machine-learning algorithm developed by EarlySign, high-risk patients were flagged based on factors such as age, gender, and recent outpatient Complete Blood Count (CBC) from the EHR.
Of the patients identified as high risk and screened with a colonoscopy, approximately 70% of those screenings revealed significant findings.
Most impressively, the model showcased remarkable performance in predicting right-sided colorectal cancer.
“This compelling study underscores the transformational role of AI in early disease detection, emphasizing its utility as a predictive tool in healthcare,” says Jeremy Pierotti, general manager, solutions at Lucem Health.
He explains in a future where the precision, efficiency and reliability of clinical AI solutions will be measured, continuous improvement will distinguish the winners.
“One key is ensuring that our training and validation data are extensive, fully inclusive of diverse populations, and correctly labeled,” he says. “It’s crucial that we examine and minimize biases in our training data, as overlooking this can risk perpetuating healthcare disparities we strive to diminish.”
Jim Kean, CEO of Molecular You, points out each year thousands of new research papers are published that validate how certain biomarkers represent an early warning sign of disease progression.
“One of the most promising applications of AI and machine learning is to organize, correlate and summarize these new findings,” he explains. “There are about a million new academic medical papers coming out every year. Humans can’t go through a million articles every year.”
AI can help organize it, correlate it and summarize it–but Kean notes the content itself is always going to be human generated.
“Keeping a human in the loop at all times helps minimize the risk of error from AI, particularly with research and academic content,” he says.
He sees AI as a driver of regulatory change around reimbursement and the way we define which biomarkers are clinically validated to diagnose disease.
“As research progresses, with the help of AI, we discover more and more biomarkers that are precursors and predictors of disease,” Kean says. “I believe we’re going to see a shift to a more dynamic approach to how the system defines early detection, with increased preventive testing and a greater number of recognized disease signatures.”
He adds ultimately, the most promising aspect is that the technology is available, the research is valid; the only thing left is to get this new era of early disease detection into clinics to start impacting as many lives as possible.
Pierotti says while AI brings great promise, challenges and constraints that will hinder its fuller potential can’t be ignored.
“Consider that the fuel propelling AI is data—copious, diverse, and nuanced,” he says. “The challenge lies in obtaining such datasets, not only in sufficient quantity, but also the quality.”
Gaining access to such comprehensive databases can be complicated, often wrapped in privacy concerns and data sharing limitations.
Meanwhile, many are raising ethical considerations in the application of AI. As these algorithms mature, becoming more nuanced and refined, the specter of biases, unfairness in decision-making, and health inequities looms larger.
“It’s a known issue, but solutions are not always easy or obvious,” Pierotti says.
He adds that despite the challenges, he’s optimistic about the prospects of using AI in disease detection, noting the early detection of diseases isn’t just cost-effective; it’s lifesaving.
“Initiating treatment when a disease is in its earliest stages significantly boosts the odds of favorable patient outcomes,” he says. “While we must manage the complexities and challenges of deploying clinical AI in a practical and responsible manner, the potential is transformative in ways that few other technology innovations have offered.”