SMASH: A Data-driven Informatics Method to Assist Experts in Characterizing Semantic Heterogeneity among Data...

Area of Science:

• Health Informatics
• Data Science
• Biomedical Data Standards

Background:

• Semantic heterogeneity (SH) poses significant challenges to data interoperability and integration within healthcare systems.
• Accurate assessment of SH is crucial for developing effective strategies to improve data consistency and usability.
• HIV-associated data elements (DEs) present a complex case for assessing SH due to variations in terminology and meaning.

Purpose of the Study:

• To assess semantic heterogeneity among HIV-associated data elements using both expert-based and data-driven approaches.
• To introduce and evaluate a novel method, String Metric-assisted Assessment of Semantic Heterogeneity (SMASH), for identifying SH.
• To compare the effectiveness of expert review and data-driven methods in identifying and characterizing SH.

Main Methods:

• Extracted eight data dictionaries from Clinicaltrials.gov to create a comprehensive inventory of HIV-associated data elements (DEs).
• Developed and applied the SMASH method to vectorize DEs and identify semantic similarities and differences between studies.
• Utilized an HIV expert to assess semantic equivalence of data element pairs and contextualize usage.

Main Results:

• Out of 1,175 data element pairs analyzed, 1,048 (87%) exhibited semantic heterogeneity, while 127 (13%) were homogeneous.
• The majority of heterogeneous pairs (97%) were characterized by semantically equivalent but syntactically different terms (e.g., 'HIV-positive' vs. 'HIV+' vs. 'Seropositive').
• SMASH effectively aided in the identification of SH, particularly highlighting semantically equivalent/syntactically different DEs.

Conclusions:

• Combining expert-based and data-driven methods, including the novel SMASH approach, is highly effective for assessing semantic heterogeneity in healthcare data.
• Semantic equivalence with syntactic differences represents the most common form of SH in HIV-associated data.
• The study recommends the adoption of complementary expert-driven and data-driven solutions to resolve semantic heterogeneity issues.