ARTIFICIAL INTELLIGENCE IN PHARMACEUTICAL PRACTICE: EVIDENCE OF EFFICACY, METHODOLOGICAL LIMITATIONS, AND CHALLENGES OF RESPONSIBLE IMPLEMENTATION

Abstract

This paper analyzed the applications of artificial intelligence in pharmaceutical practice based on original studies published between 2021 and 2026. The methodology consisted of a systematic literature review in six databases with four search strategies, resulting in 8,622 identified records, screening of 140 full-text articles, and inclusion of 31 studies organized into four thematic areas: clinical decision support and medication adherence, pharmacovigilance and adverse drug reaction prediction, drug interactions and pharmacokinetics, and drug discovery. The results demonstrated clinical efficacy in randomized clinical trials, with medication adherence of 77.4% versus 58.6% in the control group and a return on investment of 12.4:1 in an AI-supported pharmaceutical management program. Pharmacovigilance predictive models achieved AUC values between 0.876 and 0.921, while large language models showed factual hallucination rates between 6% and 23%, with clinically dangerous errors in up to 7.3% of generated recommendations. The analysis concluded that the central barrier to adopting these technologies is not computational performance but clinical translation, evidenced by a drop in F1-score from 0.82 to 0.34 in inter-hospital model transfer and by the absence of prospective multicenter validation in most studies.