Misdiagnosis-related harm quantification through mixture models and harm measures.

Area of Science:

• Health Services Research
• Medical Informatics
• Biostatistics

Background:

• Traditional chart review for monitoring misdiagnosis-related harm is labor-intensive and lacks scalability.
• Existing statistical methods are inadequate for analyzing the complex data structures in electronic health records (EHRs) for diagnostic performance evaluation.

Purpose of the Study:

• To develop and validate novel statistical methods for monitoring diagnostic performance and patient harm using EHR data.
• To quantify and compare misdiagnosis-related harm across different healthcare institutions.

Main Methods:

• Proposed a mixture regression model and associated goodness-of-fit testing for analyzing EHR data.
• Developed new harm measures and profiling analysis procedures for evaluating diagnostic performance.
• Utilized stroke occurrence data from the Taiwan Longitudinal Health Insurance Database for illustration.

Main Results:

• The proposed mixture regression model effectively analyzes EHR data to identify diagnostic errors and patient harm.
• Analysis revealed key risk factors for harm due to misdiagnosis, offering insights into healthcare quality.
• Special care hospitals demonstrated better diagnostic performance compared to general hospitals in Taiwan.

Conclusions:

• The novel statistical approach provides a scalable and effective method for monitoring diagnostic performance and patient harm.
• Findings highlight the potential for improving healthcare quality by addressing identified risk factors and disparities in diagnostic performance.