Type 2 Diabetes Risk
Description
Type 2 diabetes risk is the likelihood that the user has type 2 diabetes (impaired processing of blood sugar) and corresponds to the percentage of people with the user’s risk profile who have been diagnosed with type 2 diabetes (irrespective of their current blood sugar reading).
Type 2 diabetes diagnosis considers blood sugar readings taken under specific conditions, and so individual measurements that are abnormally high (e.g., after meals) do not necessarily mean the user has diabetes. BitDoctor determines the user’s diabetes risk based on blood flow patterns and demographic information.
A user with an elevated type 2 diabetes risk might consider having their blood sugar tested by a medical professional to determine if they meet the diagnostic criteria for diabetes.
Participants
6,840 adults (18+ years of age) recruited at a hospital health clinic in China.
Additional Data Collection
Subjects were also asked whether they have been diagnosed with diabetes.
Modeling Approach
Blood flow signal was extracted and processed from facial video, and then blood flow features were extracted from blood flow signal.
Feature selection was carried out on blood flow and demographic features to identify features predictive of diabetes status. A machine-learning based classifier was then created to predict an individual’s diabetes status based on those features.
The model was created/trained using 5,472 subjects. The distribution of sex and diabetes status in this group is summarized in Table 3.
Sex distribution
51.5 % Male; 48.5% Female
Diabetes distribution
25% Diabetic
Table 3: Diabetes distribution of training set
Model performance
The Diabetes Risk model was then validated for accuracy on an independent portion of the dataset (n=684) that was not used in training - validation set. Then the final accuracy was obtained on an independent portion of the dataset (n=684) that was not used in training - test set. The distribution of sex and diabetes status in these groups is summarized in Table 4.
Sex distribution
53.4% Male; 46.6 % Female
e 51.5% Male; 48.5% Female
Diabetes distribution
25% Diabetic
25% Diabetic
Table 4: Diabetes distribution of validation and test sets
The accuracy on the test set calculated as area under the curve was 82.1% as shown in Figure 3.
Figure 3: AUC of diabetes type 2 risk prediction
Predictions are displayed to the user as a percentage likelihood of having diabetes. A percentage of 45% or more suggests a risk of diabetes, while 55% or more suggests high risk of diabetes.
Last updated
