GNS Healthcare (GNS), a leading precision medicine company that applies causal machine learning technology to match health interventions to individual patients and discover new intervention pathways, today announced publication of a study conducted jointly with leading pharmaceutical company Pfizer Inc. by the Journal of Diabetes Science and Technology (JDST). The study is unique in its application of causal machine learning technology to transform electronic health records (EHR) data into computer models that revealed the roles of novel clinical risk factors in progression and speed of progression to type 2 diabetes (T2D) and prediabetes.
GNS and Pfizer applied the patented GNS machine learning and simulation platform, REFS™ (Reverse Engineering and Forward Simulation), to an aggregated EHR data sample provided by Humedica, an Optum/United Health Group company, retrospectively following transitions to prediabetes or T2D in more than 24,000 adults from 2007 to 2012. The analysis dataset consisted of 442 variables, representing demographics; laboratory values, including hemoglobin A1c, fasting glucose, 2-hour oral glucose tolerance, random glucose, triglycerides, total bilirubin, alanine aminotransferase (ALT), creatinine, low-density lipoprotein (LDL), high-density lipoprotein (HDL), and C-reactive protein (CRP); clinical observations, including heart rate, blood pressure, body temperature, body mass index (BMI); ICD-9 diagnosis codes; and prescription data. The paper, “Reverse Engineering and Evaluation of Prediction Models for Progression to Type 2 Diabetes: An Application of Machine Learning Using Electronic Health Records,” was published in the journal’s January 2016 issue as part of a special section on big data and diabetes.
Richard J. Willke, PhD, vice president Pfizer Global Health and Value, noted that Pfizer has initiated several experiments to assess the potential of using machine learning to discover and develop new medicines and optimize their application in the real world. “We believe causal machine learning may yield new insights, even from commonly available forms of health data,” Willke said. “This, in turn, can drive new, data-driven R&D and, ultimately, may help us bring additional innovative therapies to patients.”
REFS, which uses an entirely data-driven, hypothesis-free approach, identified many established risk factors for T2D and prediabetes and concomitantly identified novel predictors of risk for prediabetes, strengthening the body of evidence that these factors are mechanistically linked to diabetes progression. REFS identified the known T2D risk factors of blood glucose measures, age, race, triglycerides, body mass index (BMI), and blood pressure/hypertension. In addition, REFS identified several novel predictors of progression to prediabetes, including the discovery that individuals with the highest levels of high-density lipoprotein (HDL) cholesterol progressed to prediabetes on a pace 24 percent slower than other cohorts, suggesting opportunities for clinical intervention and individual risk assessment. Also, individuals with elevated levels of the liver enzyme alanine aminotransferase (ALT) progressed to T2D more quickly, at a rate 19 percent faster than other cohorts. The findings also suggest a low-magnitude effect for ALT and a role for C-reactive protein (CRP), a marker of systemic inflammation, and body temperature in progression to prediabetes.
“This study delivers new knowledge that makes it possible to identify patients at greatest risk for more rapid progression to prediabetes and to T2D—and to do so early enough that interventions can positively impact the course of disease,” said Iya Khalil, PhD, co-founder and executive vice president of GNS Healthcare. “This capability not only bolsters prevention strategies, it can improve patient care, lead to new therapeutic interventions, and can be used to more effectively stratify patients for clinical trials.”
Knowledge of the underlying clinical processes at work as patients transition between disease states dramatically improves the ability to characterize and identify individuals who are at risk, driving value for a range of healthcare stakeholders, including health plans, biopharmaceutical companies, healthcare providers, and patients. New knowledge from this study can transform prevention strategies by giving health plans and healthcare providers the opportunity to identify at-risk individuals with modifiable risk factors and prioritize these individuals for more vigorous outreach and support. In addition, the ability to more precisely identify at-risk individuals can improve how biopharmaceutical companies select patients for clinical trials, an advance that would reduce the time and cost associated with these trials to more quickly deliver high-value medications to patients.
Insights into drivers of progression may also lead to the identification of novel drug targets, guiding and accelerating the drug development process, and ultimately supporting physicians and their patients with critical information needed to more precisely, quickly, and confidently choose a treatment regimen matched to the unique characteristics of the patient. Together, these capabilities have the potential to more successfully prevent at-risk individuals from progressing to prediabetes and T2D and to more effectively and efficiently treat these conditions, advances that could alter the alarming escalation in prevalence and cost of diabetes throughout the world.
JDST is a bi-monthly, peer-reviewed scientific journal published by the Diabetes Technology Society. In addition to Pfizer, authors represent Biogen, Radial Analytics, Inc., Context Relevant, and Novartis Institutes for Biomedical Research.
Filed Under: Drug Discovery