The Use of Artificial Intelligence in the Development of Economic Models
Author(s)
Michaela Spence, MSc1, Belal Howidi, MSc1, Yixie Zhang, PhD1, Keith R. Kallmes, BSc2, Aidan Dineen, PhD1;
1EVERSANA, Value and Evidence, Burlington, ON, Canada, 2Nested Knowledge®, Saint Paul, MN, USA
1EVERSANA, Value and Evidence, Burlington, ON, Canada, 2Nested Knowledge®, Saint Paul, MN, USA
OBJECTIVES: The use of artificial intelligence (AI) tools in health economics outcomes research is rapidly growing. Through a systematic literature review assisted with AI for screening, this study aimed to identify evidence where AI is used to aid in the conceptualization and/or programming of economic models.
METHODS: Ovid MEDLINE®, Embase, and Cochrane Library were searched for studies of AI-developed economic models published between January 2010 to December 2024, along with conference abstracts retained in Embase and CENTRAL from January 2022 to December 2024. Records were screened by one human reviewer and one AI reviewer using the Nested Knowledge® platform.
RESULTS: Of 1,872 unique records identified, 1,866 were excluded with 32 disagreements between human and AI reviewers after AI training (a 98.2% agreement rate). Five unique studies were included, the majority of which (4/5) reported on the use of Generative Pre-Trained Transformer 4 (GPT-4). Two AI-developed models were coded in Excel and one used R. Overall, the accuracy of AI models was variable with generation of an error-free model ranging from 60% to 97%. One study reported incremental cost-effectiveness ratios of error-free AI-generated models were within 1% of published models. Four studies reported AI modelling was a success compared to published models or experienced developer code. Success of the AI model was largely described as requiring specific prompt development, whereas general prompts failed to generate working code. Only one study reported hallucination of data by AI, which generated a clinically implausible transition in health states.
CONCLUSIONS: AI improved efficiency while maintaining accuracy for article screening for this systematic review. Early research suggests AI can help to conceptualize and program economic models, while also identifying evidence gaps. Use of AI could increase efficiency and decrease model development cost, with potential future growth in adoption of this approach.
METHODS: Ovid MEDLINE®, Embase, and Cochrane Library were searched for studies of AI-developed economic models published between January 2010 to December 2024, along with conference abstracts retained in Embase and CENTRAL from January 2022 to December 2024. Records were screened by one human reviewer and one AI reviewer using the Nested Knowledge® platform.
RESULTS: Of 1,872 unique records identified, 1,866 were excluded with 32 disagreements between human and AI reviewers after AI training (a 98.2% agreement rate). Five unique studies were included, the majority of which (4/5) reported on the use of Generative Pre-Trained Transformer 4 (GPT-4). Two AI-developed models were coded in Excel and one used R. Overall, the accuracy of AI models was variable with generation of an error-free model ranging from 60% to 97%. One study reported incremental cost-effectiveness ratios of error-free AI-generated models were within 1% of published models. Four studies reported AI modelling was a success compared to published models or experienced developer code. Success of the AI model was largely described as requiring specific prompt development, whereas general prompts failed to generate working code. Only one study reported hallucination of data by AI, which generated a clinically implausible transition in health states.
CONCLUSIONS: AI improved efficiency while maintaining accuracy for article screening for this systematic review. Early research suggests AI can help to conceptualize and program economic models, while also identifying evidence gaps. Use of AI could increase efficiency and decrease model development cost, with potential future growth in adoption of this approach.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
P28
Topic
Economic Evaluation
Disease
No Additional Disease & Conditions/Specialized Treatment Areas