Researchers develop machine-learning mannequin that precisely predicts diabetes, research says


TORONTO – Canadian researchers have developed a machine learning model that accurately predicts diabetes in a population based on routinely collected health data, a new study says.

The study, published in the JAMA Network Open Journal, tested new machine learning technologies on routinely collected health data that examined the entire population of Ontario. The study was carried out by the ICES non-profit data research institute.

Using linked administrative health data from Ontario from 2006 to 2016, researchers created a validated algorithm by training the model on information gathered from nearly 1.7 million patients.

Using the validated data that was collected, including variables such as body mass index and high blood pressure, the algorithm was 80 percent accurate in predicting who was at risk for type 2 diabetes in the population over several years.

“Effective prevention and specific measures to prevent type 2 diabetes are in place. However, it is often a challenge to ensure that these approaches are targeted to those who need them most in the context of a health system, ”said ICES scientist and lead study author Dr. Laura Rosella in the press release. “Predictive models for type 2 diabetes are particularly useful in enabling more effective and efficient targeting of health system interventions that help prevent type 2 diabetes.”

The role of the machine learning model is not in individual patient care, but in the context of population health planning and management in predicting diabetes across health systems across the board and promoting health equity.

Using machine learning and health data, population health planning tools that accurately differentiate between high and low risk groups can “be used as a guide to investment and targeted interventions to prevent diabetes,” the study said in five years’ time.

“The model was tested for accuracy to predict the incidence of diabetes for a total of 5 years and in key demographic and socio-economic subgroups to ensure it worked well for the entire population,” said Rosella.

The study found that in 2009 the number of patients with diabetes in Ontario was estimated at 785,000, with a cost of $ 3.5 billion. Those numbers rose to $ 1,144,000 and $ 5.4 billion, respectively, seven years later in 2016.

The researchers found that the cohort of patients with diabetes grew at an average rate of 51,800 new patients per year between 2009 and 2016, with an additional $ 242 million per year as related costs.

The model also showed that patients who were predicted to be at the highest risk of diabetes according to the data accounted for the bulk of the associated healthcare costs: Medium to high risk patients account for five percent of the cost, but the population studied accounted for 26 percent of the total with it related diabetes costs.

The study found that it was difficult to transfer diabetes prevention from individual patients to the population because of system-level barriers such as differences in socio-economic status, lack of access to healthy food and medicines, and lack of access to health care.

The patients at highest risk averaged 58 years old, including a larger cohort of immigrants and people more likely to live in areas with lower socioeconomic indices and higher unemployment.

The new model aims to solve the problems governments, health insurers, and public health planners face when it comes to identifying those who are most in need of diabetes intervention and prevention, especially patients who are older and older come from “marginalized areas in terms of ethnicity and material disadvantage”.

With files from the Canadian press