Artificial intelligence (AI) may outperform human doctors in adhering to established treatment guidelines for clinical depression, according to a study published in the open-access journal Family Medicine and Community Health. The research also highlights the potential of AI in addressing gender and social class biases often encountered in primary care doctor-patient relationships.
However, the study acknowledges the necessity of further research to assess AI’s effectiveness in managing severe cases and the ethical implications of its use. Depression is a widespread mental health condition, and those affected often turn to their primary care physicians for assistance. The recommended treatment for depression typically follows evidence-based clinical guidelines, with the approach tailored to the severity of the condition.Â
ChatGPT, an AI language model, presents an opportunity to provide rapid, objective, data-driven insights that can complement conventional diagnostic methods while offering confidentiality and anonymity. Researchers sought to determine how well this technology evaluated the recommended therapeutic approach for both mild and severe major depression and whether it exhibited gender or social class biases when compared to primary care doctors.Â
To conduct the study, the researchers utilized carefully designed and previously validated patient scenarios (vignettes) featuring individuals experiencing symptoms of depression, such as sadness, sleep disturbances, and loss of appetite over the preceding three weeks. Eight versions of these vignettes were created, each incorporating different variations of patient characteristics, including gender, social class, and depression severity.
ChatGPT versions 3.5 and 4 were asked to suggest the appropriate course of action for each of these scenarios, which included options like watchful waiting, referral for psychotherapy, prescription of drugs, or a combination of therapy and medication.Â
The findings revealed significant differences in recommendations between primary care doctors and ChatGPT. In mild cases, only about 4% of physicians exclusively recommended psychotherapy, in line with clinical guidance, while ChatGPT-3.5 and ChatGPT-4 chose this option in 95% and 97.5% of cases, respectively. Most doctors suggested either drug treatment alone (48%) or a combination of psychotherapy and medication (32.5%).Â
For severe cases, most doctors recommended a combination of psychotherapy and medication (44.5%). ChatGPT offered this recommendation more frequently, aligning with clinical guidelines (72% for ChatGPT 3.5 and 100% for ChatGPT 4). Furthermore, ChatGPT did not recommend exclusive drug treatment for severe cases, which was suggested by 40% of the physicians.Â
The study also explored the types of drugs recommended when medication was advised. Doctors typically suggested a combination of antidepressants, anti-anxiety drugs, and sleeping pills (67.5%), while ChatGPT-3.5 and ChatGPT-4 were more likely to recommend antidepressants alone (74% and 68%, respectively). Additionally, ChatGPT versions suggested a combination of antidepressants and anti-anxiety drugs and sleeping pills more often than doctors.Â
The researchers noted that the study has limitations, such as relying on specific iterations of ChatGPT and data from French primary care doctors, which may not be universally applicable. Additionally, the patient scenarios represented an initial visit for a complaint of depression, lacking information about ongoing treatment or other variables that doctors would typically consider.Â
The researchers highlighted the potential of ChatGPT-4 in adhering to clinical guidelines and its lack of gender and socioeconomic biases. They also underscored the importance of addressing ethical issues related to data privacy and security when handling sensitive mental health data, emphasizing that AI should not replace human clinical judgment in depression diagnosis or treatment.Â
Reference Â
Identifying depression and its determinants upon initiating treatment: ChatGPT versus primary care physicians, Family Medicine and Community Health (2023). DOI: 10.1136/fmch-2023-002391.Â
