Data electronically extracted from the electronic health record require validation.

Area of Science:

• Health Informatics
• Clinical Data Management
• Neonatal Research

Background:

• Electronic health records (EHRs) offer vast data but require accurate extraction for research.
• Automated data extraction from EHRs can introduce errors, impacting study validity.
• Neonatal hypoglycemia research relies on precise data for effective analysis.

Purpose of the Study:

• To identify sources of error in electronically extracted data from EHRs.
• To assess the accuracy of extracted categorical and continuous variables for neonatal hypoglycemia research.
• To evaluate the effectiveness of validation procedures in improving data quality.

Main Methods:

• Electronically extracted data for 100 neonates with early-onset hypoglycemia were analyzed.
• Extraction language was developed by an IT specialist.
• Data validation involved blinded manual chart review, using Kappa coefficients for categorical and percent validity for continuous variables.

Main Results:

• 35% of categorical variables achieved acceptable agreement (Kappa ≥ 0.81); 22% showed fair-slight agreement (Kappa < 0.40).
• The variable "hypoglycemia" demonstrated poor agreement (Kappa = 0.16).
• Continuous variables showed higher validity (≥ 94%); improvements were noted after refining extraction language, though "hypoglycemia" remained problematic.

Conclusions:

• Data extraction without validation, particularly for categorical variables (e.g., using International Classification of Diseases-9 codes), often leads to inaccurate data.
• Implementing built-in validation processes is crucial for ensuring the reliability of electronically extracted EHR data.
• Specific variables, such as "hypoglycemia," may require tailored validation strategies to overcome extraction challenges.