Back

Clinical Information Needs Among Latin American Physicians: A Multi-Country Analysis of Semantic Clinical Search

Monsalve Barrientos, K.; Villa, M. C.; Castano-Villegas, N.; Zea, J.; Velasquez, L.

2026-07-09 health informatics
10.64898/2026.06.26.26356340 medRxiv
Show abstract

Background: Physician clinical information-seeking behavior has been studied in high-income settings but remains poorly characterized in Latin America. Objective: To describe clinical information needs, geographic variation, and temporal usage patterns among physicians using a semantic clinical search platform across Latin America. Methods: We conducted a retrospective query-log analysis of physician searches performed between May 2025 and June 2026. The dataset included 235,803 queries generated by 9,443 physicians across 20 Latin American countries. Query topics were classified using a 10-category canonical taxonomy derived from platform metadata and validated through manual review of a stratified random sample of 200 queries. Results: The physician activation rate was 84.4%. Among categorized queries, Management Plan (35.6%), Differential Diagnosis (20.6%), and Work-up and Test Selection (11.4%) accounted for 67.6% of all searches. This category hierarchy was broadly consistent across the 17 countries included in country-level analyses despite differences in cohort size and healthcare settings. Query intensity was also similar across countries, with a mean of 29.6 queries per active physician over the study period. Manual validation confirmed agreement between reviewer assessment and taxonomy assignment in approximately 93% of sampled queries. Conclusions: Clinical information seeking among Latin American physicians was dominated by management planning, diagnostic reasoning, and test selection, with broadly consistent patterns across countries. These findings provide a regional behavioral baseline for understanding physician information needs in semantic clinical search systems.

Matching journals

The top 3 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.