Clinical Safety of AI-Generated Antibiotic Prescribing Advice: Guideline Adherence and Misinformation Risk Among Large Language Models

Abstract

Abstract Background Large language models (LLMs) are increasingly used in telehealth, but their safety in antibiotic prescribing remains uncertain, particularly in the presence of patient misinformation. Methods This cross-sectional study tested five AI language models using 1,000 simulated viral-infection vignettes based on CDC, NICE, and WHO guidance. Each case was presented first as a standard medical question and then, when antibiotics were refused, as a misinformation prompt. Responses were classified as guideline-concordant refusal, unprompted overprescription, or coercion-mediated overprescription. Safety behaviors, WHO AWaRe class, and antibiotic spectrum were recorded. Analyses used Python, SPSS, and Alteryx, with chi-square tests, Bonferroni-corrected pairwise comparisons, McNemar tests, and logistic mixed-effects modeling. Results Overall, 76.2% of responses were guideline-concordant, while 6.6% showed unprompted overprescribing and 17.3% were influenced by misinformation. Some models were more vulnerable to misinformation than others. Although most responses correctly noted that antibiotics do not treat viral infections, fewer advised consulting a doctor, and warnings against self-medication were rare. Many inappropriate prescriptions involved broad-spectrum antibiotics. Conclusion LLMs show potential in telehealth but remain prone to misinformation and inappropriate prescribing. Stronger guideline integration and clinical oversight are necessary to ensure safe use.

Ähnliche Arbeiten