AD-CDO: A Lightweight Ontology for Representing Eligibility Criteria in Alzheimer's Disease Clinical Trials

Zenan Sun, Rashmie Abeysinghe, Xiaojin Li, Xinyue Hu, Licong Cui, Guo-Qiang Zhang, Jiang Bian, Cui TaoarXiv 2025 · DOI · arXiv

Introduction

Clinical trials for Alzheimer’s Disease (AD) are essential for therapeutic progress, yet they are hindered by the complexity of eligibility criteria. These criteria, which define participant suitability, are predominantly documented as unstructured free text. This lack of standardization leads to significant challenges: inconsistency in terminology across trials, difficulty in cross-study comparisons, and obstacles in linking trial requirements to Electronic Health Records (EHR) for automated recruitment. While existing ontologies like the Alzheimer’s Disease Ontology (ADO) exist, they are often too broad or focused on molecular biology, making them impractical for operational clinical trial modeling.

The AD-CDO Framework

To address these issues, researchers developed the Alzheimer’s Disease Common Data Element Ontology (AD-CDO). This lightweight ontology is specifically designed to standardize and represent AD clinical trial eligibility criteria. The development process follows a data-driven pipeline encompassing data preparation, ontology construction, evaluation, and application.

Overview of the AD-CDO development workflow, from data collection to practical use cases.

Methodology

The construction of AD-CDO involved several key stages:

  1. Data Collection and Annotation: The team sourced 1,509 interventional AD trials from ClinicalTrials.gov. A subset of 300 trials was manually annotated to establish a gold standard for entity recognition.
  2. Extraction and Normalization: A hybrid NLP pipeline utilizing the CLAMP toolkit for extraction and SapBERT for entity linking was employed to process the text. This resulted in 38,399 normalized entity mentions.
  3. Semantic Categorization: Extracted concepts were organized into seven clinically relevant categories: Disease, Medication, Diagnostic Test, Procedure, Social Determinants of Health (SDoH), Rating Criteria, and Fertility.
  4. Standardized Mapping: Concepts were manually mapped to global biomedical vocabularies, including UMLS (CUI), OMOP (AthenaID), DrugBank, and NLM VSAC value sets. This ensures interoperability with hospital databases using standards like ICD-10 or RxNorm.
  5. Optimization: To maintain a "lightweight" profile, the researchers used the Jenks Natural Breaks clustering method. This statistical approach helped identify the optimal number of concepts to include, balancing ontology size with data coverage.

Results and Evaluation

The final AD-CDO is a compact yet robust structure consisting of 293 classes and 305 logical axioms. Despite its streamlined nature, the ontology achieves a 63.4% coverage of all entity mentions across the 1,500+ trials analyzed. This high coverage-to-size ratio demonstrates that a targeted, data-driven approach can effectively capture the core requirements of clinical research without the overhead of massive, generalized ontologies.

Use Cases and Applications

The utility of AD-CDO was validated through two primary applications:

  • AD Clinical Trial Simulation System: By integrating the ontology with Temporal Event Logic (TEL), researchers can formally specify and simulate trials against real-world data, facilitating better trial design.
  • Entity Normalization: The ontology serves as a foundation for pipelines that convert unstructured clinical text into standardized entities, enabling consistent phenotyping and data integration across diverse research environments.

Conclusion and Future Work

AD-CDO provides a critical bridge between unstructured clinical trial protocols and computable data. While currently focused on a hierarchical class structure, future work will aim to incorporate complex relational semantics (e.g., drug-disease interactions) and automate the mapping process to external vocabularies. Expanding the scope beyond eligibility criteria to include biomarkers and longitudinal outcomes will further enhance its utility in the AD research ecosystem.

Summary updated

BibTeX
@misc{sun2025ad,
  author = {Zenan Sun and Rashmie Abeysinghe and Xiaojin Li and Xinyue Hu and Licong Cui and Guo-Qiang Zhang and Jiang Bian and Cui Tao},
  title = {{AD-CDO: A Lightweight Ontology for Representing Eligibility Criteria in Alzheimer's Disease Clinical Trials}},
  year = {2025},
  eprint = {2511.21724},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2511.21724},
}
Figure