# ChatGPT in the Management of Chronic Rhinosinusitis with Nasal Polyps: Promising Support or Digital Illusion? Insights from a Multicenter Observational Study

## TL;DR

This study evaluates whether ChatGPT can reliably help choose biologic therapies for nasal polyps, finding it currently unreliable for clinical use.

## Contribution

The study is the first to compare ChatGPT's biologic therapy recommendations with expert opinions in chronic rhinosinusitis with nasal polyps.

## Key findings

- ChatGPT showed 59.2% overall concordance with expert recommendations for biologic therapy selection.
- Concordance was highest for Dupilumab (62.8%) and lowest for Omalizumab (9.1%).
- ChatGPT's reliability for autonomous clinical decision-making in this context is currently limited.

## Abstract

Background/Objective: Chronic rhinosinusitis with nasal polyps is a chronic inflammatory disease with a significant impact on quality of life and is frequently associated, from a pathogenetic perspective, with type 2 inflammation. The introduction of biologic therapies has marked a turning point in the management of severe forms of the disease, offering a valuable treatment option. However, selecting the most suitable biologic agent for a specific patient remains a clinical challenge. Artificial intelligence, and, in particular, ChatGPT, has recently been proposed as a potential tool to support medical decision-making and guide therapeutic choices. To evaluate the concordance between the therapeutic recommendations provided by ChatGPT and those of a multidisciplinary expert board in selecting the most appropriate biologic therapy for CRSwNP patients, based on the analysis of their phenotype and endotype. Methods: A multicenter observational cohort study was conducted. Clinical data from 286 patients with CRSwNP were analyzed. For each case, the therapeutic choice among Dupilumab, Mepolizumab, and Omalizumab was compared between the board and ChatGPT. Concordance rates and Cohen’s Kappa coefficient were calculated. Results: Overall concordance was 59.2%, with a Cohen’s Kappa coefficient of 0.116. Concordance by drug was 62.8% for Dupilumab, 26.5% for Mepolizumab, and 9.1% for Omalizumab. Patients presented with severe clinical profiles, with an average Nasal Polyp Score of 6.22 and an average SNOT-22 score of 64.5. Conclusions: This study demonstrates that, despite its substantial theoretical potential, ChatGPT is currently not a reliable tool for the autonomous selection of biological therapies in patients with CRSwNP. Further studies are necessary to enhance its reliability and integration into clinical practice.