Using machine learning to improve anaphylaxis case identification in medical claims data.

Area of Science:

• Medical Informatics
• Computational Biology
• Public Health Data Science

Background:

• Anaphylaxis is a severe, life-threatening allergic reaction.
• Accurate identification in healthcare databases is crucial for public health and big data initiatives.
• Existing methods for detecting medical outcomes in claims data can be laborious and expensive.

Purpose of the Study:

• To evaluate the utility of machine learning (ML) in identifying incident anaphylaxis cases using healthcare claims data.
• To develop and test ML models for detecting anaphylaxis with varying data quality.
• To identify novel features that can enhance existing case-finding algorithms.

Main Methods:

• Utilized claims data from October 1, 2015, to February 28, 2019, from the CMS database.
• Implemented a feature selection pipeline to identify critical data features.
• Employed unsupervised and supervised ML methods, including Sammon mapping and eXtreme Gradient Boosting, to train models.

Main Results:

• Machine learning model accuracies ranged from 47.7% to 94.4% when tested on ground truth data.
• Identified new features that can assist experts in refining current case-finding algorithms.
• Demonstrated that ML models can achieve performance comparable to expert-developed algorithms.

Conclusions:

• Machine learning models show significant potential for accurately identifying anaphylaxis in large healthcare datasets.
• ML can streamline algorithm construction and potentially improve performance over existing expert-driven methods.
• Identifying key features using ML can enhance rule-based algorithms for medical outcome detection.