Assessing accuracy and completeness of drug information in mobile apps compared with Micromedex®: An AI-based study...

Ensuring access to accurate and complete drug information is fundamental to rational medication use. Mobile medical applications (MMAs) are increasingly used by healthcare providers; however, their quality compared with institutional databases remains underexplored, especially in non-English and resource-limited settings. Natural Language Processing (NLP), particularly using Thai-language transformer models such as WangchanBERTa, enables automated screening and classification of real-world drug-related queries derived from public online communities.

This study aimed to compare the accuracy and completeness of drug information from three MMAs-Lexicomp®, Medscape®, and Epocrates®-against the institutional gold standard, Micromedex®, using AI-classified Thai-language clinical questions.

A total of 1500 Thai-language questions about drug therapy were collected from online health forums (Pharmacafe, Pantip, Reddit). Using WangchanBERTa for text classification and stratified sampling, 194 representative questions were mapped to 13 pharmacoinformatic domains. Each question was answered using the three MMAs and Micromedex®. Three licensed pharmacists independently scored each response for accuracy and completeness using a validated binary checklist (1 = correct/complete; 0 = incorrect/incomplete). Inter-rater consensus was achieved through group discussion. Accuracy and completeness were expressed as percentages and analyzed via one-way ANOVA with Tukey HSD post-hoc testing.

Micromedex® demonstrated the highest accuracy (55.7 %) and completeness (53.2 %), significantly outperforming Epocrates® (p < 0.05). Among MMAs, Lexicomp® showed superior performance (accuracy = 32.3 %; completeness = 29.4 %), whereas Medscape® (accuracy = 31.6 %; completeness = 28.8 %) and Epocrates® (accuracy = 20.7 %; completeness = 18.0 %) ranked lower. The weighted composite score used previously (60 % accuracy + 40 % completeness) was removed for simplicity and clarity.

While Lexicomp® demonstrates potential as a practical alternative to Micromedex® in ambulatory and community-based pharmacy environments, none of the MMAs achieved equivalent reliability. Real-world Thai-language data analyzed through NLP pipelines provide a reproducible framework for pharmacoinformatic benchmarking. This approach supports rational medication use and guides digital-tool selection in low-resource healthcare systems.