Abstract
BackgroundOvid MEDLINE and PubMed are widely used for clinical and biomedical literature searches, but they differ in syntax, field handling, controlled vocabulary and available content. These differences matter in evidence synthesis, where search sensitivity and reproducibility affect the completeness of the evidence base. Manual translation can introduce errors, while simple find-and-replace approaches may fail on wildcards, fields and controlled headings.ObjectiveTo describe ClinSearchBridge, a deterministic, recall-oriented and auditable framework for translating Ovid MEDLINE strategies into executable PubMed strategies while preserving source concepts and recording approximations, fallbacks and manual-review conditions.MethodsClinSearchBridge separates source-file ingestion from a deterministic rule framework operating on normalised, numbered Ovid MEDLINE rows. Its ordered pipeline converts Boolean references, wildcards, truncation, free-text fields, controlled vocabulary, publication types, comments/corrections relations and selected clinical-review filters, with selective MeSH metadata resolution and line-level audit. Development used 61 Cochrane review strategies, with an additional 14-study external non-Cochrane test set. Comparators were Medline Transpose in Ovid-to-PubMed Accuracy mode and one-shot GPT-5.6 Luna, Terra and Sol. Each LLM used five independent stateless requests per strategy, medium reasoning effort, no tools or external retrieval, and the same frozen prompt.ResultsOn the 61-review development corpus, ClinSearchBridge retained 99.30% of Ovid-reference PMIDs on average. In the five-system common-study external comparison, it achieved the highest mean Ovid-reference coverage (97.48%, SD 3.57%); GPT-5.6 Sol was the strongest generative baseline (94.14%). In the 13-study failure-adjusted analysis, mean coverage was 97.85% for ClinSearchBridge, 74.17% for Medline Transpose, 72.69% for GPT-5.6 Luna, 83.65% for GPT-5.6 Terra and 85.98% for GPT-5.6 Sol. Terra had the highest LLM run-success rate, while Sol showed the strongest repeated-run retrieval consistency. Retrieval-set overlap, burden and reproducibility are also reported because no expert gold-standard PubMed translation is assumed.ConclusionsA rule-governed, auditable translation layer can make Ovid-to-PubMed conversion more reproducible and safer for clinical evidence-synthesis pipelines than opaque or silent transformations. The aim is not exact emulation of Ovid semantics, which is impossible for some constructs, but conservative preservation of searchable concepts with explicit warnings and a retrieval gate when validation fails. ClinSearchBridge is available at https://clinsearchbridge.github.io/, with the implementation and supporting materials openly accessible.