Assessing Plausibility of Clinical Fact Dates in Real World Data

Area of Science:

• Data Science
• Health Informatics
• Biostatistics

Background:

• Clinical data quality is crucial for research and healthcare.
• Federated healthcare datasets present unique data integrity challenges.
• Timestamp accuracy is a key component of clinical data reliability.

Purpose of the Study:

• To introduce a quantitative framework for assessing timestamp plausibility in real-world clinical data.
• To apply this framework to identify data quality issues within a large federated network.
• To demonstrate the utility of the Jensen-Shannon divergence for anomaly detection in healthcare data.

Main Methods:

• A quantitative framework was developed utilizing the Jensen-Shannon divergence.
• The framework was applied to analyze encounter start dates across the TriNetX US network.
• Abnormal patterns were identified based on statistical divergence metrics.

Main Results:

• The Jensen-Shannon divergence successfully identified healthcare organizations exhibiting abnormal date patterns.
• Date binning or clustering, indicative of potential data quality issues, was detected.
• The method proved effective in highlighting anomalies within the federated network.

Conclusions:

• The proposed framework offers a simple yet effective method for timestamp plausibility assessment.
• This approach facilitates systematic detection of anomalies in federated healthcare datasets.
• Continuous data quality improvement can be supported through this quantitative analysis.