Towards a Computable Data Corpus of Temporal Correlations between Drug Administration and Lab Value Changes

Area of Science:

• Pharmacovigilance
• Health Informatics
• Data Science

Background:

• Automated detection of adverse drug reactions (ADRs) using electronic health records offers advantages over traditional methods.
• Assessing the temporal relationship between drug intake and reaction occurrence is crucial for causality but lacks adequate public data.
• Existing methods like spontaneous reporting and manual chart review have limitations in efficiency and scope.

Purpose of the Study:

• To create a publicly available dataset of drug administrations and corresponding laboratory observations.
• To establish a computable ground truth for the temporal correlation between drug administration and lab value alterations.
• To facilitate the development and validation of algorithms for automated ADR detection.

Main Methods:

• Retrospective extraction and transformation of routine data from a university clinic.
• Expert evaluation of drug administration and temporally corresponding laboratory observations for reasonable time relationships.
• Normalization of laboratory parameter values for a generic approach.

Main Results:

• A data corpus of 400 episodes detailing normalized laboratory parameter values in temporal context with drug administrations.
• Manual classification of each episode indicating potential temporal correlation, absence of correlation, or indeterminacy.
• Assignment of a concordance value to each episode, reflecting assessment difficulty.

Conclusions:

• The dataset serves as a vital resource for research in systematic, computerized ADR detection.
• It provides a ground truth for comparing algorithm-based assessments with human expert evaluations.
• Enables the development and immediate validation of algorithms for detecting temporal relationships in retrospective data.