Discovering Meaningful Cut-points to Predict High HbA1c Variation

Monitoring HbA1c, the measurement for the average blood glucose level, is important to diabetic patients and may help improve their treatment. This study aims to train learning algorithms that predict patients with high variation in their HbA1c readings. Attributes in clinical data are often continuous, such as age, blood pressure, and lab tests. However, many machine learning algorithms work better – or work only – with categorical attributes. We propose using discretization processes to identify meaningful cut points for continuous attributes. For example, the study could help understand questions like, What is the range of age and BMI for diabetic patients that have high variation in their HbA1c readings? The discretization process finds the number of intervals and the boundaries for the intervals. The process may reveal meaningful cut points in continuous attributes and contribute knowledge to the medical domain. Discretization process also allows us to discover suitable intervals and boundaries for a particular attribute depending on variety of parameters that can be controlled by the experimenter.