How Medial EarlySign’s machine learning model combed EHRs to identify patients at risk of kidney damage.
Image has been resized. Courtesy: Salvatore P, Pexels
Those who keep up with artificial intelligence (AI) news might not be surprised by this: A healthcare start-up has built a machine learning model that digs into electronic health records (EHR) to determine which diabetes patients are at high risk of kidney damage over the next year. Medial EarlySign’s AI technology identified 45% of such patients, a ratio that exceeds the success of existing clinical protocols by a quarter, making the finding yet another promising win for AI in its sweeping campaign against diabetes.
“Immense efforts are invested in developing treatment protocols to reduce the number of patients who will develop renal dysfunction due to diabetes,” says Ran Goshen, MD, the company’s chief medical officer. “Medial EarlySign’s algorithm can aid decision makers, drug developers, insurers, and providers to better allocate their capped resources and secure preferential clinical outcome as well.”
The Israeli AI start-up this week announced the results of a study of the model, which isolated less than 5% of the 400,000 patients with diabetes in Medial EarlySign’s database of 15 million people, according to the company. The algorithm pinpointed patients before they showed any symptoms, the organization notes. It did so by analyzing “dozens of factors” in EHRs, such as lab tests, demographic information, medications, diagnostic codes, and more.
The revelation highlights the importance of using data and high-tech analytics to tackle big medical issues, something that health systems and innovators are pursuing more and more.
“The significant size and rapid growth of digital health databases now allow the application of advanced mathematical tools that can identify patterns in diverse patient populations in order to identify high-risk patients,” says Itamar Raz, MD, who leads the Israel National Council of Diabetes and is director emeritus of Hadassah University Hospital’s diabetes unit. “Rather than relying only on small patient samples based on known risk factors, machine learning tools can reveal the slightest correlations among these parameters and discover additional risk indicators that can lead to improved prediabetic patient risk stratification.”
Medial EarlySign says kidney problems are among the most common complications of diabetes, and anywhere from 20-40% of patients across the world suffer from such issues. Identifying the potential for renal dysfunction early could lead to prevention or delayed damage, which, in turn, could fight against future complications of diabetes end stage renal disease, according to the company.
It also seems as if diabetes is among the most traveled paths for AI innovators looking to make inroads into healthcare. Just yesterday, for example, the Iowa-based start-up IDx announced that the FDA declared the company’s deep learning diagnostic system for diabetic retinopathy a “breakthrough device,” fast-tracking the review process. AI has also helped people with diabetes lose weight, spotted diabetic retinopathy in other settings, and determined the efficacy of basal insulin in routine care. The pervasive nature of the disease has provided fertile ground for disrupters looking to innovate and bring their technologies to market, according to experts, but AI is certainly no magic bullet for diabetes.
Medial EarlySign is using AI to tackle diabetes from several angles. Its predictive models and algorithmic calculators have focused on prediabetes to diabetes progression, cardiovascular disease related to diabetes, and engagement initiatives for patients with diabetes and prediabetes, according to the company.