Background: Decision making in the treatment of rheumatoid arthritis is a complex process. The opinion on the use of Artificial Intelligence (AI) in therapeutic decision-making is a controversial topic, while some see AI as an ally, others as a threat.

Objectives: Compare the attitude of Spanish rheumatologists in different clinical situations with the answers provided by AI.

Methods: An online Google form with 15 questions was sent through social networks to several groups of rheumatologists in the national territory. Descriptive statistical analysis was carried out, subsequently the survey was completed by ChatGPT 3.5 and ChatGPT 4.

Results: 108 surveys were collected. In patients with recent-onset RA with poor prognostic factors, in addition to corticosteroids, half of those surveyed (50%) begin treatment with csDMARD + rapid escalation to bDMARD/sd if response is insufficient, while chatGPT 3.5 leans towards bDMARD/sd±MTX from baseline and chatGPT 4 due to combined therapy with ≥ 2 csDMARDs. The majority of rheumatologists (47.2%) and ChatGPT 3.5 and 4 agree that the patient’s profile is the most important factor when choosing the drug. The most relevant factors when choosing each DMARD are: anti-TNF, rheumatologists (47.2%) due to its cost-effectiveness; AI for its effectiveness; anti-IL6, due to its effectiveness (rheumatologists (72.2%) and AI); abatacept, for its efficacy and safety in RA-ILD patients (rheumatologists (53.7%) and AI); rituximab, for its safety in patients refractory to other treatments (rheumatologists (76.9%) and ChatGPT 3.5), for its effectiveness in seropositive patients (ChatGPT 4); JAK inhibitor, due to the possibility of use in monotherapy (rheumatologists (40.7%) and ChatGPT 3.5) and efficacy (ChatGPT 4). The most important factor that makes rheumatologists (57.4%) and Chat GPT 4 change treatment is the measurement of activity, while ChatGPT 3.5 responded “existence of other potentially more effective or safe therapeutic alternatives.” If there is a good therapeutic response, the majority of rheumatologists (50%) and ChatGPT 4 would first optimize the bDMARD/sd administration interval, while ChatGPT 3.5 would simultaneously optimize the csDMARD and bDMARD/sd. In case of using combined therapy of csDMARD and bDMARD/sdDMARD, both the majority of the rheumatologists surveyed (59.3%) and IA recommend maintaining csDMARD in addition to the bDMARD/sdDMARD. Regarding the use of corticosteroids, the vast majority of colleagues (79.6%) and ChatGPT4 prefer to try to stop them as soon as possible, while ChatGPT 3.5 would choose to maintain low doses, unless there are specific comorbidities. In the event of pregnancy, 59.3% of those surveyed replace the drug with a safer one, while IA opts to maintain the drug if it is anti-TNF. Incident diagnosis of cancer is a reason for discontinuation of all bDMARDs/sd for 47.2% of respondents, however the IA would prefer to maintain/switch to rituximab if the patient is receiving bDMARD/sd. Finally, biosimilar drugs are considered equally effective and safe as the originals by both rheumatologists (80.6%) and AI.

Conclusion: A striking heterogeneity has been observed in the way of acting in complex clinical situations, both among rheumatologists and in comparison with ChatGPT. Among rheumatologists, there is consensus on the need to limit the use of corticosteroids, on the use of activity indices to evaluate therapeutic response and on the wide acceptance of biosimilar drugs, the latter is also supported by AI. ChatGPT 4 shows greater agreement with the rheumatologist’s opinions than ChatGPT 3.5.

REFERENCES: NIL.

Acknowledgements: NIL.

Disclosure of Interests: None declared.