Development and Validation of HealthImpact: An Incident Diabetes Prediction Model Based on Administrative Data

Area of Science:

• Health Informatics
• Epidemiology
• Predictive Modeling

Background:

• Type 2 diabetes is a growing public health concern.
• Accurate risk stratification is crucial for early intervention.
• Existing methods often rely on patient-provided information or laboratory tests, limiting accessibility.

Purpose of the Study:

• To develop and validate a predictive model for incident type 2 diabetes.
• The model, named HealthImpact, is based solely on administrative claims data.
• To assess the model's efficacy in identifying individuals at high risk for developing diabetes.

Main Methods:

• Utilized the Optum Labs Data Warehouse (OLDW), a national commercial administrative dataset.
• Developed and internally validated the HealthImpact model using a nested case-control design with large training and validation cohorts.
• Externally validated the model in a prospective cohort of 2,000,000 adults over 3 years.

Main Results:

• The HealthImpact model, comprising 48 variables, demonstrated strong predictive performance with a c-statistic of 0.808 in internal validation and 0.817 in external validation.
• A HealthImpact score threshold of 90 was identified as indicative of high risk for incident diabetes.
• The model achieved high specificity (94.92%) and negative predictive value (96.90%) for predicting new diabetes diagnoses within 3 years.

Conclusions:

• The HealthImpact model provides an efficient and effective method for risk stratification of incident type 2 diabetes.
• It leverages readily available administrative data, eliminating the need for patient-reported data or laboratory results.
• This approach facilitates broader application in identifying at-risk populations for targeted preventive measures.