Validation and Development of Claims-Based Algorithms for Identifying Thyroid Eye Disease Using the IRIS...

Area of Science:

• Ophthalmology
• Health Informatics
• Data Science

Background:

• Thyroid eye disease (TED) identification in claims data is challenging.
• Existing rule-based algorithms have limitations in balancing sensitivity and specificity.
• Real-world data offers potential for improved case identification methods.

Purpose of the Study:

• To validate claims-based algorithms for identifying thyroid eye disease (TED) cases.
• To assess if machine learning can enhance TED case identification accuracy.
• To compare the performance of rule-based and machine learning approaches.

Main Methods:

• Evaluated six rule-based algorithms using linked real-world data (Komodo Health, IRIS Registry).
• Developed supervised machine learning models utilizing demographic, diagnostic, procedural, and medication data.
• Employed recursive feature elimination and cross-validation for model development and performance assessment.

Main Results:

• Rule-based algorithms showed a trade-off between sensitivity and specificity.
• Algorithm 6 improved sensitivity while maintaining high specificity and positive predictive value (PPV).
• Machine learning models demonstrated flexibility, allowing performance adjustment for different research needs, with some models outperforming rule-based approaches.

Conclusions:

• Both rule-based and machine learning methods can improve TED case identification in claims data.
• Machine learning offers enhanced flexibility for tailoring performance to specific research objectives.
• No single method consistently outperformed others across all metrics, highlighting the value of both approaches.