Validation of classification algorithms for childhood diabetes identified from administrative data

Area of Science:

• Pediatric Endocrinology
• Public Health Surveillance
• Health Informatics

Background:

• Childhood type 1 diabetes is common, but type 2 diabetes cases are rising.
• Canadian National Diabetes Surveillance System (NDSS) uses administrative data but doesn't distinguish diabetes types.
• Accurate classification of pediatric diabetes type is crucial for epidemiological research.

Purpose of the Study:

• To validate algorithms for classifying diabetes type in children (<20 years) using NDSS data.
• To develop and test methods distinguishing type 1 from type 2 diabetes in administrative health records.

Main Methods:

• Applied NDSS case definition to British Columbia children (1996-2007).
• Developed four classification algorithms using demographics and drug-utilization patterns.
• Validated algorithms against a gold-standard clinical database for accuracy.

Main Results:

• Age-based rules were sensitive for identifying type 1 diabetes.
• Drug-utilization restrictions (no insulin) were sensitive for identifying type 2 diabetes.
• One algorithm achieved optimal sensitivity and specificity for both types (Type 1: Sn 98.6%, Sp 78.2%; Type 2: Sn 83.2%, Sp 97.5%).

Conclusions:

• Demographic and drug-utilization data can reliably differentiate pediatric diabetes types in administrative databases.
• Validated algorithms enhance the utility of surveillance systems like NDSS for childhood diabetes research.
• Further validation in other regions is recommended to ensure broad applicability.