Understanding Clinician EHR Data Quality for Reuse in Predictive Modelling

Area of Science:

• Health Informatics
• Clinical Data Management
• Machine Learning in Healthcare

Background:

• Growing volumes of clinical data require robust methods for reuse.
• Clinician trust is essential for adopting electronic health record (EHR) data in research.
• Ensuring data quality is paramount for reliable machine learning predictions.

Purpose of the Study:

• To assess the plausibility and fitness of structured EHR data for predicting work-related musculoskeletal disorder outcomes.
• To evaluate data quality by comparing individual clinician data entry against group data.
• To identify factors influencing clinician trust in EHR data for research.

Main Methods:

• Utilized a proprietary, structured electronic health record system.
• Assessed data plausibility by comparing individual clinician entries to group norms.
• Employed machine learning techniques to evaluate data suitability for outcome prediction.

Main Results:

• Individual clinician data entry generally correlated positively with group data, suggesting fitness for reuse.
• Inconsistencies were identified from the clinician's perspective, potentially impacting data trust.
• The study highlights the need for clinician engagement in data quality assessments.

Conclusions:

• Clinician engagement is crucial for assessing and improving EHR data quality.
• Recognizing clinicians as local knowledge experts can enhance data reliability for machine learning.
• Data quality assessments must consider the clinician's viewpoint to build trust and ensure data utility.