An automated data verification approach for improving data quality in a clinical registry

Area of Science:

• Medical Informatics
• Data Science
• Clinical Research

Background:

• Clinical registry data quality is essential for study credibility.
• Manual data verification from paper case report forms (CRFs) and electronic medical records (EMRs) is inefficient and labor-intensive.
• Existing verification methods struggle with accuracy and time consumption.

Purpose of the Study:

• To develop and evaluate an automated data verification approach for clinical registries.
• To enhance the efficiency and accuracy of identifying data errors in registry studies.
• To leverage optical character recognition (OCR) and natural language processing (NLP) for automated data validation.

Main Methods:

• Machine learning-enhanced OCR was used to recognize handwritten data and checkboxes on scanned CRFs.
• Natural language processing (NLP) techniques were employed to extract relevant patient information from EMRs.
• Data from CRFs and EMRs were compared against registry data for automated verification.

Main Results:

• The automated approach achieved high accuracy in CRF data recognition (checkboxes: 0.93, handwriting: 0.74) and EMR data extraction (0.97).
• Compared to manual verification, the automated method demonstrated superior accuracy (0.93 vs 0.92) and recall (0.96 vs 0.71).
• The automated verification process consumed significantly less time (0.5 hours vs 7.5 hours).

Conclusions:

• The automated data verification approach is more effective and efficient than manual methods for identifying data errors in registries.
• This method significantly improves data quality by enhancing accuracy and recall while drastically reducing processing time.
• The proposed automated system holds substantial potential for improving the overall quality of clinical registry data.