{
  "abstract": "Background and Importance Artificial intelligence (AI) is increasingly used to support clinical decision-making in hospital pharmacy. However, different AI models may affect reliability and safety in high-risk settings. This study compares two OpenAI models–ChatGPT-Instant (GPTI) and ChatGPT-Thinking (GPTT)–in solving clinical queries in hospital pharmacy.Aim and Objectives To compare two AI models, GPTI and GPTT, evaluating accuracy and response time (RT) to determine their suitability for pharmaceutical practice.Material and Methods Eighty questions were designed and grouped into four thematic blocks of 20 questions each: (B1) availability of the medicine in the hospital and therapeutic alternatives; (B2) posology; (B3) intravenous compatibility and (B4) administration via nasogastric tube (NGT). Both models received identical questions with the memory function disabled. A hospital pharmacist assessed responses as correct/incorrect, missing information, and measured RT. Agreement in correct/incorrect responses between models was analysed using McNemar’s test per block.Results GPTI showed a mean RT of 7.8 seconds in B1 and B3, 6.3 seconds in B2, and 4.3 seconds in B4. GPTT exhibited longer times: 123.6 seconds in B1, 85.5 seconds in B2, 84.5 seconds in B3, and 160.9 seconds in B4. GPTI achieved 80% correct in B1, 90% in B2 and B4, and 45% in B3; GPTT achieved 95% correct in B1, B2, B3, and 100% in B4. Missing information for GPTI was 5% in B2 and B3 and 20% in B4, whereas GPTT showed missing information only in B1 (5%). McNemar p-values: B1 p=0.375; B2 p=1.0; B3 p=0.002; B4 p=not defined, as the GPTT variable was constant in this block.Conclusion and Relevance GPTT outperformed GPTI in reliability across all clinical tasks, with statistically significant differences in intravenous compatibility. GPTI was comparable to GPTT in Blocks 2 and 4, with shorter RTs. GPTI was faster across all areas but had higher error rates and more missing information. GPTT is therefore more suitable for assisting with clinical queries; however, its RT require improvement.Conflict of Interest No conflict of interest",
  "authors": [
    {
      "affiliations": [
        "Hospital General Universitario Santa Lucía, Hospital Pharmacy, Cartagena, Spain"
      ],
      "name": "F Lago Ballester"
    },
    {
      "affiliations": [
        "Hospital General Universitario Santa Lucía, Hospital Pharmacy, Cartagena, Spain"
      ],
      "name": "A Lázaro Cebas"
    },
    {
      "affiliations": [
        "Hospital General Universitario Santa Lucía, Hospital Pharmacy, Cartagena, Spain"
      ],
      "name": "Á Varea Béjar"
    },
    {
      "affiliations": [
        "Hospital General Universitario Santa Lucía, Hospital Pharmacy, Cartagena, Spain"
      ],
      "name": "ME Martínez Madrid"
    },
    {
      "affiliations": [
        "Hospital General Universitario Santa Lucía, Hospital Pharmacy, Cartagena, Spain"
      ],
      "name": "EJ Sevilla Alarcón"
    },
    {
      "affiliations": [
        "Hospital General Universitario Santa Lucía, Hospital Pharmacy, Cartagena, Spain"
      ],
      "name": "JM Moya Flores"
    }
  ],
  "title": "5PSQ-023 Evaluating the performance of two OpenAI models for clinical decision-making in hospital pharmacy",
  "uid": "fe095a22-1a5d-556d-9493-e574e86f5297"
}