By Benjamin Ross
March 19, 2018 | Medial EarlySign recently announced the results of a clinical data study in which their machine-learning based model was able to analyze Electronic Health Records (EHRs) and identify diabetic patients who are at the highest risk for having kidney disease within one year.
According to Medial EarlySign, 20-40% of diabetics worldwide suffer from diabetes-related kidney complications. Early identification and treatment may help prevent or slow the progression of damage to the kidney, reducing the likelihood of future complications, like end stage renal disease (ESRD).
Tomer Amit, VP in Corporate Marketing at Medial EarlySign, sees the value of this test as getting ahead of the game in terms of diagnosing and treating diabetes and diabetes-related diseases.
“Our goal is to find patients who are very likely to develop kidney diseases before they actually do,” Amit told Bio-IT World. “It’s not just predicting who’s going to get sick, it’s who’s going to get sick in a short time frame so we can take preventative action.”
Medial EarlySign’s machine-learning-based model analyzed dozens of factors found in EHRs, including laboratory tests results, demographics, medication, and diagnostic codes among others.
Amit says that this model is part of a suite of predictive models and algorithms, including the company’s 2017 Bio-IT World Best Practices award-winning project with Maccabi Healthcare System.
“This study looks at our diabetes suite, with a few new models that look at prediabetic and diabetic patients,” says Amit. “One identifies prediabetic patients who are most likely to develop diabetes within a year, which would give healthcare organizations the time to intervene before diabetes progressed. The second part looks at diabetes complication, such as kidney disease, where we identified individuals who are most likely to develop kidney disease also within 12 months.”
Medial EarlySign began the study late 2016, isolating less than 5% of the 400,000 diabetic patients that were selected from their database, a collective of patient data from the UK, Israel, and the US.
The company’s algorithm was able to identify 45% of those selected patients who would progress to kidney damage within a year. The company says these results represent 25% more patients than would have been identified by commonly-used clinical tools and judgment.
Amit says the focus now for Medial EarlySign’s suite of tools is to mimic their encouraging results within healthcare organizations, extending beyond population health on a predictive scale.
“We want to work with doctor-to-patient relationships in the hospital setting, helping doctors make better decisions while the patient is in their care,” says Amit.
The first step is looking at how to embed these solutions within healthcare organizations’ systems, Amit says. He sees the current system as being too wide open. Rather than looking at 200,000 patients, for example, and trying to identify those who are going to be sick and get to them in time, Medial EarlySign wants to use their suite of machine-learning based models to focus on a lower percentage of the population, vastly improving the chances that diseases will be detected and treated in time.
“Right now, you only get to treat people who show symptoms or who are beginning to get sick,” says Amit. “We want to catch [the disease] beforehand in order to prevent the disease all together.”