Using association rule mining for phenotype extraction from electronic health records

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Clinical and Translational Research

Background:

• Electronic Health Records (EHRs) adoption offers opportunities for secondary data use.
• Phenotype extraction from EHRs is crucial for population-based studies but is resource-intensive.
• Existing methods for developing phenotyping algorithms are time-consuming and require significant resources.

Purpose of the Study:

• To propose an efficient machine learning technique for semi-automatic modeling of phenotyping algorithms.
• To leverage distributional associational rule mining (ARM) for faster and more robust phenotype extraction.
• To improve the development and testing process of phenotyping algorithms for EHR data.

Main Methods:

• Utilized distributional associational rule mining (ARM), a machine learning technique.
• Applied ARM to large datasets for discovering phenotype definition criteria and rules.
• Compared ARM performance against traditional machine learning methods like logistic regression and support vector machines.

Main Results:

• ARM demonstrated a highly efficient and robust framework for phenotype definition.
• Preliminary results show significantly improved performance compared to logistic regression and support vector machines.
• ARM generated rule patterns that are amenable to human interpretation, aiding clinical understanding.

Conclusions:

• Distributional ARM is an efficient and effective method for semi-automatic phenotyping algorithm development.
• This approach accelerates the secondary use of EHR data for clinical and translational research.
• ARM offers a promising alternative to traditional methods, enhancing both performance and interpretability.