The Digital Analytic Patient Reviewer (DAPR) for COVID-19 Data Mart Validation

Area of Science:

• * Biomedical Informatics
• * Health Services Research
• * Clinical Data Science

Background:

• * High-quality data is crucial for COVID-19 research.
• * Manual chart review is time-consuming and resource-intensive.
• * Validating derived clinical indicators and machine learning phenotypes requires efficient tools.

Purpose of the Study:

• * To validate derived COVID-19 clinical indicators and machine learning phenotypes.
• * To assess the utility of a natural language processing (NLP)-based chart review tool, the Digital Analytic Patient Reviewer (DAPR).
• * To improve the efficiency and reliability of COVID-19 data validation for research.

Main Methods:

• * Retrospective manual chart review of 150 COVID-19-positive patients.
• * Development and application of 127 NLP logics within DAPR for concept and phenotype extraction.
• * Transformation of DAPR for research purposes, ensuring data privacy and enabling fast access.
• * Survey to evaluate the perceived validation difficulty and usefulness of DAPR.

Main Results:

• * High performance metrics for core COVID-19 concepts (cohort, index date, admission).
• * Three machine learning phenotypes were removed due to performance degradation in the prepandemic population.
• * Survey indicated positive user attitudes, ease of validation, and DAPR's effectiveness in finding information.

Conclusions:

• * NLP technology significantly aided COVID-19 data validation, accelerating the process.
• * DAPR facilitated the prompt provision of reliable research data during the COVID-19 crisis.
• * The benefits of DAPR can be extended to other research domains and user groups.