LLMs Recognize TDM Problems but Reason Overconfidently on NTI Drugs
This benchmark shows current large language models interpret phenytoin and digoxin levels well but falter on pharmacokinetic reasoning and uncertainty acknowledgment, reinforcing that pharmacist oversight remains mandatory.
phenytoindigoxintherapeutic drug monitoringTDMnarrow therapeutic index drugslarge language modelsartificial intelligenceclinical decision supportChatGPTClaudeGeminiclinical pharmacybenchmark studyPharmacokinetics / TDM
LLMs show promise for drug monitoring, yet caution remains due to potential reasoning errors.
Large language models (LLMs) are being considered as tools for therapeutic drug monitoring (TDM) but their reliability in clinical settings remains uncertain. Claude Sonnet 4.6 had the highest performance (94.9% ± 8.3%), outperforming ChatGPT 5.5 and Gemini 3.1 Pro, particularly in reasoning-dependent tasks. LLMs can identify TDM issues but struggle with reasoning tasks, highlighting the importance of pharmacist oversight to ensure patient safety. Pharmacist supervision is crucial when using LLMs for therapeutic drug monitoring to prevent safety concerns from overconfidence errors.
📊 Deep Analysis Available
Subscribers get the full breakdown: study design, outcomes, pharmacist implications, clinical pearls, GRADE evidence grade, and a downloadable PDF report. Cancel anytime.
Free account, no credit card. Includes 1 full deep analysis every month.
Authors:Azmakan H, Azamakan T
Citation:Azmakan H, Azamakan T Performance of large language models on narrow therapeutic index drug monitoring: Implications for clinical pharmacy practice. Journal of the American Pharmacists Association : JAPhA. 2026:103524. doi:10.1016/j.japh.2026.103524