Skip to content

Knowing diabetes risk years ahead could mean targeted prevention, lower costs: study

The data can help develop targeted population-wide strategies to reduce disease prevalence among high-risk groups
25288228_web1_20210521190536-60a8443d44fe76d48ef45897jpeg
People gather next to the Lachine Canal on a warm spring day in Montreal, Saturday, May 15, 2021, as the COVID-19 pandemic continues in Canada and around the world. A machine learning model used health data collected routinely to predict the majority of people most likely to develop the disease, says the lead author of a study that suggests the findings could be used to create targeted prevention programs years before someone develops the disease. THE CANADIAN PRESS/Graham Hughes

A new study suggests Type 2 diabetes can be predicted years in advance by a machine learning model that can scour and analyze routinely collected health data for a variety of risk factors.

Lead author Laura Rosella, a scientist at the non-profit research group ICES, formerly known as the Institute for Clinical Evaluative Sciences, says researchers looked at health information as well as social and demographic data from a random sample of nearly 1.7 million residents in Ontario between 2006 and 2016.

The model turned out to be 80 per cent accurate at predicting those at highest risk of developing Type 2 diabetes five years later. They tended to be 58 years old on average, and included a greater proportion of immigrants, and people who were more likely to live in neighbourhoods with lower incomes and higher unemployment.

But the goal of the study, published Tuesday in the journal JAMA Network Open, was not to guide individual treatment.

Researchers say the data can help develop targeted population-wide strategies to reduce disease prevalence among high-risk groups, ease the burden on the health-care system and save millions of dollars in costs related to diabetes.

“The whole idea of just making better use of the data we have in the health-care system and actually putting it towards driving decisions is what this study is all about,” said Rosella, pointing to potential policy measures such as income supplements, food subsidies and housing support.

“Now we want to think through how do we actually implement it, which involves a lot more than statistics and modelling and actually involves talking to people, to providers, to decision-makers, and making sure they can take the output and actually understand it and use it.”

The machine learning model considered variables including high blood pressure and body mass index, but also looked at the social and demographic determinants of health, which the study says are important predictors but are often missing in clinical data.

Out of a test cohort of 236,506 individuals, 1,967 developed diabetes. The model predicted slightly more at 2,000.

Scaling this cohort to the entire population of Ontario, researchers estimated the number of patients with diabetes to be 785,000 in 2009, with an associated cost of $3.5 billion. The numbers rose to 1,144,000 patients and $5.4 billion seven years later.

Rosella said insurance companies provide clinicians in some countries with these types of analytics, and sometimes with recommendations for patients. But that only addresses a segment of the at-risk population when a much broader strategy could do much more, she said.

“What’s not being done to the same extent is taking that information and saying, ‘I’m trying to contain costs and reduce risk in this whole population. Who do I need to target?’ But we’re seeing more and more examples and more and more of that type of thinking in health systems across the world.”

Dr. Harpreet Bajaj, an endocrinologist and director of late-phase research at LMC Healthcare in Toronto, said earlier interventions can make a huge difference.

“The more downstream we start the better it will be. In pre-diabetes, if we talk about the disease state itself, despite their best efforts, for many people it will still convert to diabetes. So if we started before people would actually even develop pre-diabetes that’s a better way to go,” he said.

Jagjeevan Singh of Brampton, Ont., who is one of Bajaj’s patients, developed Type 2 diabetes two years ago at age 32.

Singh said he believes the stress of his job at a large warehouse contributed to the chronic disease. But he improved his condition by switching jobs, changing his diet, losing the weight he’d gained from not eating properly while doing shift work, and started practising yoga and meditation.

Learning about his high risk of diabetes earlier would have helped him better handle the stress and make quick changes to protect his health, Singh said.

“Right now, it’s under control,” he said.

Camille Bains, The Canadian Press

Like us on Facebook and follow us on Twitter.

Want to support local journalism? Make a donation here.