JMIR Publications

ABSTRACT

Background:

We presented a substantially improved diabetes risk prediction model based on a large retrospective population cohort of >230,000 people enrolled during 2006-2017 using sophisticated machine learning algorithms.

Objective:

To construct a machine learning-based diabetes risk prediction tool using only self-reported information.

Methods:

We collected demographic, medical, behavioural and incident type 2 diabetes mellitus (T2DM) information in >230,000 diabetes-free participants from the 45 and Up Study. We predicted and compared the risk of diabetes onset in these participants in 3, 5, 7 and 10 years after enrolment based on three machine learning approaches and the conventional regression model.

Results:

Approximately 6.05% (14,313/236,684) of the participants developed T2DM during an average 8.8-year follow-up period. The 10-year diabetes incidence in males was 8.30% (8.08-8.49%) and significantly higher than 6.20% (6.00-6.40%) in females (OR = 1.37 [1.32, 1.41]). In obese participants, the incidence was doubled (males: 17.78% (17.05-18.43%); females 14.59% (13.99-15.17%)). Gradient boosting machine (GBM) performed the best among the four models (‘area under curve’ score 79% in 3-year prediction and 75% for 10-year). All machine learning models predicted BMI to be the most significant contributing factor to diabetes onset, as it explained 12-50% of the variance in the prediction of diabetes. The model predicted that, if BMI in obese and overweight participants could be hypothetically reduced to healthy, their 10-year probability of diabetes onset would be significantly reduced from 8.3% to 2.8% (t-test, p<0.001).

Conclusions:

A one-time self-reported survey can accurately predict the risk of incident diabetes using a machine learning approach. Reducing BMI can significantly reduce the risk of developing type 2 diabetes mellitus.