Childhood Cancer Data Initiative Annual Symposium (Abstract Registration): Submission #68

Abstract Submission for Poster Presentation
-------------------------------------------
Abstract Title:: Hybrid Semantic–Lexical Retrieval for Source-to-CDE Mapping in the NCI caDSR
Abstract::
Accurate mapping of source data elements to standardized common data elements (CDEs) is essential for biomedical data integration, but short, abbreviated, and heterogeneous source descriptions make both lexical and semantic matching unreliable. We constructed a benchmark of approximately 69,000 expert-linked source-to-CDE mappings from the National Cancer Institute’s caDSR and used it to evaluate bi-encoders, text representations, and two-stage fine-tuning. We then developed a hybrid system that combines semantic retrieval, CDE Match-inspired keyword retrieval, cross-encoder scoring, and supervised reranking. Across an internal test set and five distribution-shifted holdouts, the system achieved Recall@5 of 0.971 internally and 0.802–0.972 externally. It achieved the highest Recall@5 among evaluated methods on five of six datasets; on GDC, performance was near ceiling at 70/72, within one query of the official NCI CDE Match service and its Python approximation at 71/72.

Authors::
1. First Name: James
   Middle Initial: H
   Last Name: Tanis
   Degree(s): Ph.D.
   Organization: NCI
2. First Name: Denise
   Middle Initial: B
   Last Name: Warzel
   Degree(s): M.Sc.
   Organization: NCI
3. First Name: Rakesh
   Last Name: Khanna
   Degree(s): B.S.
   Organization: NCI
4. First Name: Qingrong
   Last Name: Chen
   Degree(s): Ph.D.
   Organization: NCI
5. First Name: Chunhua
   Last Name: Yan
   Degree(s): Ph.D.
   Organization: NCI
6. First Name: Ravi
   Last Name: Valleleth
   Degree(s): B.S.
   Organization: NCI
7. First Name: Matthew
   Middle Initial: T
   Last Name: Nash
   Degree(s): B.S.
   Organization: NCI
8. First Name: Deepali
   Last Name: Manathara
   Degree(s): B.S.
   Organization: NCI
9. First Name: Bradley
   Middle Initial: B
   Last Name: Biggers
   Degree(s): B.S.
   Organization: NCI
10. First Name: Joyce
    Middle Initial: B
    Last Name: Soares
    Degree(s): M.Sc.
    Organization: NCI
11. First Name: Daoud
    Last Name: Meerzaman
    Degree(s): Ph.D.
    Organization: NCI

Presenting Author:: James H Tanis
Institution:: NCI
Email Address:: james.tanis@nih.gov