Keyword search has been the default information retrieval tool in clinical research for thirty years, and it is also systematically misaligned with how clinical knowledge is actually structured. When a safety reviewer searches for "hepatotoxicity" in a pharmacovigilance database, they miss every record that used "liver toxicity", "hepatic failure", "elevated ALT", "drug-induced liver injury", or any of the dozens of other terms that describe the same clinical phenomenon. When a clinical scientist searches for studies in "heart failure", they miss studies that used "cardiac failure", "congestive heart failure", "CHF", or ICD codes rather than text labels. These are not edge cases — they are the normal operating condition of clinical research text.
The Synonym Problem
The most immediate failure mode of keyword search is the synonym problem: the same concept is expressed using many different terms across different documents, databases, systems, and time periods. Medical terminology is particularly rich in synonymy — partly because concepts acquire multiple names through historical accident, partly because different professional communities use different conventions, and partly because the same underlying entity is described at different levels of specificity in different contexts. Keyword search treats each term as a separate query and returns disjoint result sets. Ontology-based search treats synonyms as a property of a shared concept identifier and returns all matching documents regardless of which term they used.
The Hierarchy Problem
Beyond synonyms, clinical knowledge has hierarchical structure that keyword search cannot exploit. A query for "beta-blockers" should return documents about metoprolol, bisoprolol, carvedilol, and atenolol, because these are all beta-blockers. A query for "antihypertensive adverse events" should find records coded to specific drugs in the antihypertensive class, even if the search term "antihypertensive" does not appear in those records. Ontology-driven search expands queries along the hierarchy automatically, using the is-a relationships defined in the drug ontology to retrieve all specific instances of the queried class.
The Consequence for Evidence Review
The combined effect of the synonym and hierarchy problems is that keyword-based systematic literature reviews and pharmacovigilance searches systematically miss a substantial fraction of the relevant evidence. Estimates from the information science literature suggest that comprehensive keyword search in biomedical databases typically achieves 60–75% recall — meaning 25–40% of relevant documents are not retrieved. In a safety context, missing 30% of adverse event reports is not acceptable. Ontology-driven search consistently achieves recall above 90% for well-maintained concept hierarchies, with manageable precision trade-offs.
