Accurate and timely synthesis of clinical drug information is essential for pharmacists engaged in evidence-based practice and formulary evaluation. However, generating structured summaries from diverse data sources remains time-intensive. Advances in large language models (LLMs) offer new opportunities to automate this process. To establish and evaluate a pilot inference framework that automatically generates structured preliminary clinical drug reports by integrating outputs from multiple prompt-specific language models. The framework was designed for compatibility with transformer-based LLMs, including domain-adapted and instruction-tuned variants. Nine independent prompts were implemented to extract key clinical sections: FDA-approved indications, efficacy evidence, summary of clinical findings, dosing recommendations, and adverse reaction profiles. Each section was processed individually, standardized, and automatically merged into a cohesive, human-readable report. Output reproducibility, formatting consistency, and clinical usability were qualitatively assessed across models. The framework generated structured, readable reports that preserved section-level accuracy and coherence across all evaluated models. Outputs demonstrated uniform formatting, reproducibility, and clinical interpretability, independent of the underlying LLM. The automated process effectively reduced manual synthesis time and maintained fidelity to primary clinical content. This inference framework demonstrates the feasibility of using multiple prompt-specific LLMs to automate generation of preliminary clinical drug reports. Its modular, reproducible design supports integration into pharmacy informatics and drug evaluation workflows, offering a scalable tool to enhance clinical decision-making and streamline literature synthesis.
Khan et al. (Wed,) studied this question.