A study published in Hospital Pharmacy on July 23, 2026, evaluated the use of large language models (LLMs) to support hospital compounding pharmacy workflows. Researchers E. Castellana and MR Chiappetta assessed the efficiency and reliability of Claude Sonnet 4.6 in determining drug crushability, regulatory compliance, and formulation feasibility for pediatric patients. The model was configured to synthesize data from multiple sources, including the Friuli Venezia Giulia "Do Not Crush" list, the Italian Medicines Agency (AIFA) database, AIFA Law 648/96 off-label lists, and the Stabilis stability database.

Testing involved two drugs, propranolol hydrochloride and imatinib mesylate. The LLM-assisted workflow reduced average information retrieval time to 1 minute 45 seconds per drug, significantly faster than the traditional manual process, which required 20 minutes for retrieval plus 5 to 10 minutes for manual transcription. The model achieved 100% output completeness in all predefined categories. Specifically, for propranolol, it identified SyrSpend-based formulations stable for 146 days at room temperature. For imatinib, it correctly noted cytotoxic handling requirements and identified alternative stability data of 30 days under refrigerated conditions. The researchers concluded that the LLM functions effectively as a decision-support tool, supporting further integration into routine galenical (pharmaceutical preparation) practice.

A study published in Hospital Pharmacy on July 23, 2026, evaluated the use of large language models (LLMs) to support hospital compounding pharmacy workflows. Researchers E. Castellana and MR Chiappetta assessed the efficiency and reliability of Claude Sonnet 4.6 in determining drug crushability, regulatory compliance, and formulation feasibility for pediatric patients. The model was configured to synthesize data from multiple sources, including the Friuli Venezia Giulia "Do Not Crush" list, the Italian Medicines Agency (AIFA) database, AIFA Law 648/96 off-label lists, and the Stabilis stability database.

Testing involved two drugs, propranolol hydrochloride and imatinib mesylate. The LLM-assisted workflow reduced average information retrieval time to 1 minute 45 seconds per drug, significantly faster than the traditional manual process, which required 20 minutes for retrieval plus 5 to 10 minutes for manual transcription. The model achieved 100% output completeness in all predefined categories. Specifically, for propranolol, it identified SyrSpend-based formulations stable for 146 days at room temperature. For imatinib, it correctly noted cytotoxic handling requirements and identified alternative stability data of 30 days under refrigerated conditions. The researchers concluded that the LLM functions effectively as a decision-support tool, supporting further integration into routine galenical (pharmaceutical preparation) practice.