Predicting misdiagnosed adult-onset type 1 diabetes using machine learning

Area of Science:

• Endocrinology
• Medical Informatics

Background:

• Adult-onset type 1 diabetes comprises nearly half of new diagnoses.
• Clinical differentiation between type 1 and type 2 diabetes in adults is challenging.
• Misdiagnosis can lead to suboptimal patient care.

Purpose of the Study:

• To develop a machine learning model to identify adult patients with type 1 diabetes who were misdiagnosed with type 2 diabetes.
• To improve diagnostic accuracy for adult-onset type 1 diabetes.

Main Methods:

• A retrospective study utilizing Ambulatory Electronic Medical Records (AEMR).
• Extraction of patient data including age, demographics, risk factors, symptoms, treatments, procedures, vitals, and lab results.
• Development of a machine learning model to detect misdiagnosed type 1 diabetes.

Main Results:

• Key predictors for misdiagnosis included age, BMI/weight, therapy history, and HbA1c/blood glucose levels.
• The model achieved 17% precision at 10% recall for identifying misdiagnosed cases.
• This significantly exceeds the <1% incidence rate of misdiagnosis at initial type 2 diabetes encounters.

Conclusions:

• The developed algorithm demonstrates potential as a clinical decision support tool.
• Integration into Electronic Medical Record (EMR) systems can aid in reducing misdiagnosis.
• Facilitates timely and effective treatment for adult-onset type 1 diabetes.