PILOT IMPLEMENTATION OF AI IN THE GENERATION OF INDIVIDUALIZED PARENTERAL NUTRITION PLANS (TPN PLAN CREATOR) bartosz.w.kowalski@gmail.com

243-06

PILOT IMPLEMENTATION OF AI IN THE GENERATION OF INDIVIDUALIZED PARENTERAL NUTRITION PLANS (TPN PLAN CREATOR)

2

 

 

 

 

Poland

Oral or Poster

Artificial intelligence (AI) is increasingly integrated into clinical workflows to enhance precision and efficiency. This pilot study aimed to assess the clinical utility and limitations of TPN Plan Creator, an AI-based tool for generating individualized total parenteral nutrition (TPN) protocols.

We conducted a prospective pilot study using TPN Plan Creator, a GPT-4-based chatbot trained to formulate TPN plans based on patient-specific data: anthropometry, SGA status, laboratory values, and fluid losses. The study enrolled 150 hospitalized patients requiring TPN due to various indications, including short bowel syndrome, high-output enteric fistulas, bowel obstruction, CRBSI, and postoperative complications. All AI-generated plans were reviewed by a clinical nutrition expert and compared against ESPEN and ASPEN guidelines.

Of the 150 generated plans, 48 (32%) were fully compliant with clinical standards. Moderate errors were identified in 57 cases (38%), including inaccurate electrolyte dosing, unclear fluid balance, or suboptimal micronutrient selection. Notably, critical errors were observed in 45 cases (30%), involving incorrect glucose concentrations, inappropriate lipid dosing, or infusion volumes exceeding venous access limitations. Despite these issues, the AI tool significantly shortened preparation time and supported consistency across cases.

TPN Plan Creator shows considerable promise as a supportive technology in the personalization of TPN therapy. While expert oversight remains essential, the integration of AI into clinical nutrition planning offers a path toward improved efficiency, reduced workload, and standardized protocol generation. With further training on real-world datasets and ongoing validation against clinical outcomes, such systems may evolve into reliable decision-support tools, reshaping the future of parenteral nutrition planning.