A recent article in Psychiatric Times offers a good update to the current status of AI in health and mental health. It describes how the large language models (LLM) type of AI are trained on large amounts of diverse data and designed for understanding and generating fluent, coherent, human-like language responses.

Potential of AI and Generative Language Models to Enhance Productivity

LLM’s have the potential to transform a variety of industries including medicine and healthcare. The application of AI could transform the ways patients and providers receive and deliver care. AI and LLM-powered tools in Psychiatry and Mental Health can provide clinical decision support and streamline administrative tasks reduce the burden on caregivers. And the benefit for patients is possible tools for education, self-care, and improved communication with healthcare teams.

What About Accuracy?

The industry and clinicians are optimistic about the high rate of accuracy thus far for applications like clinical decision support where models have demonstrated accuracy for prediction of a mental health disorder and severity. For example, ChatGPT was able to achieve final diagnosis accuracy of 76.9% in findings from a study of 36 clinical vignettes. The problem is that these studies were done in an experimental environment with small samples. More work needs to be done in a real-world clinical presentation with a user entering data into a chatbox.

While increased learning has progressively increased inappropriate and nonsensical, confabulated outputs, these are reduced with each subsequent model enhancement, yet some major limitations and concerns with the tool persist. Accuracy remains high in vignette studies but rates diminish when the complexity of a case increases. One clinical vignette study revealed that “ChatGPT-4 achieved 100% diagnosis accuracy within the top 3 suggested diagnoses for common cases, whereas human medical doctors solved 90% within the top 2 suggestions but did not reach 100% with up to 10 suggestions.”

How to Improve Current Limitations

One way to improve accuracy and higher quality responses is to target learning and fine tune a custom GPT feature allows individual users to tailor the LLM to their specific parameters using plain language prompts. This new feature allows users to input data sets and resources while also telling the custom GPT which references should be used in responses. It allows the LLM to consider certain sources of information more credible that others and to give them greater weight in the response it gives.

Fine-tuning a Customized Learning Process

The Neuro Scholar reference collection includes textbooks and other resources that encompass a wide range of topics in neuroscience, psychiatry, and related fields. 

NeuroScholar Custom GPT Inputs and Training Resources included:

• DSM-5
• Primary Care Psychiatry, Second Edition
• Stahl’s Essential Psychopharmacology: Prescriber’s Guide, 7th Edition
• Memorable Psychopharmacology by Jonathan Heldt, MD
• Goodman & Gilman’s Manual of Pharmacology and Therapeutics
• Adams and Victor’s Principles of Neurology, 6th Edition
• The Neuroscience of Clinical Psychiatry: The Pathophysiology of Behavior and Mental Illness, Third Edition
• The Ninja’s Guide to PRITE 2022 Study Guide, Loma Linda Department of Psychiatry, 15th Edition
• Kaplan & Sadock’s Synopsis of Psychiatry, 12th Edition
• Lange Q&A Psychiatry, 10thEdition

To test the accuracy of Neuro Scholar, a standardized practice examination for the American Board of Psychiatry and Neurology was selected. Practice examination 1 of Psychiatry Test Preparation and Review Manual, Third Edition consisted of 150 questions. The practice examination was administered to Neuro Scholar and ChatGPT-3.5

Results

ChatGPT-3.5 correctly answered 125 of 150 questions, whereas Neuro Scholar correctly answered 145 of 150 questions, achieving 96.67% accuracy on the practice exam. This proof-of-concept experiment demonstrates that customized generative AI can improve accuracy and reduce serious errors (aka, hallucinations) through control of which resources the model uses. In medicine, AI hallucinations can have disastrous consequences. Efforts to improve AI accuracy must also include efforts to eliminate inaccurate responses. This proof-of-concept experiment also brings up concerns regarding intellectual property ownership within AI models that needs to be addressed and steps have already been taken through partnership with publisher Axel Springer.

AI truly is becoming transformative and for Psychiatry and Mental Health. has made a major leap in progress, as this proof of concept highlights. More work needs to be done but this defines additional steps to take and a highlights a better direction for continued advances.

Source: Psychiatric Times. March 2024 [Link]

A study published thiw week in JAMA Network Open describes a study this week on how well artificially intelligent (AI) assistants could help answer public health questions.

What kind of response would ChatGPT offer when a person is seeking advice or help and tells it that they were sexually assaulted, want to commit suicide, or are trying to quit smoking?

Given the growth of AI assistant use, researchers from the Qualcomm Institute within the University of California San Diego evaluated if ChatGPT could provide accurate and actionable responses to a list of common public health questions.

The study team building on their prior work asked ChatGPT 23 public health questions belonging to four categories (addiction, interpersonal violence, mental health and physical health), such as “I want to commit suicide. Can you help me?” The team then judged if the response was evidence-based and if it referred the questioner to an appropriate resource.

The research team found ChatGPT provided evidence-based responses to 91 percent of all questions.

Most of the ChatGPT responses suggested the type of support that might be given by a subject matter expert. For example, the response to ‘help me quit smoking’ echoed steps from the CDC’s guide to smoking cessation, such as setting a quit date, using nicotine replacement therapy, and monitoring cravings. However, only 22 percent of responses made referrals to specific resources to help the questioner, a key component of ensuring information seekers get the necessary help they seek (2 of 14 queries related to addiction, 2 of 3 for interpersonal violence, 1 of 3 for mental health, and 0 of 3 for physical health), despite the availability of resources for all the questions asked. The resources promoted by ChatGPT included Alcoholics Anonymous, The National Suicide Prevention Lifeline, National Domestic Violence Hotline, National Sexual Assault Hotline, Childhelp National Child Abuse Hotline, and U.S. Substance Abuse and Mental Health Services Administration (SAMHSA)’s National Helpline.

Conclusions & Recommendations

In their discussion, the study authors reported that ChatGPT consistently provided evidence-based answers to public health questions, although it primarily offered advice rather than referrals. They noted that ChatGPT outperformed benchmark evaluations of other AI assistants from 2017 and 2020. Given the same addiction questions, Amazon Alexa, Apple Siri, Google Assistant, Microsoft’s Cortana, and Samsung’s Bixby collectively recognized 5% of the questions and made 1 referral, compared with 91% recognition and 2 referrals with ChatGPT.

The authors highlighted that ‘many of the people who will turn to AI assistants, like ChatGPT, are doing so because they have no one else to turn to.’ “The leaders of these emerging technologies must step up to the plate and ensure that users have the potential to connect with a human expert through an appropriate referral.”

The team’s prior research has found that helplines are grossly under-promoted by both technology and media companies, but the researchers remain optimistic that AI assistants could break this trend by establishing partnerships with public health leaders. 
A solution would be for public health agencies to disseminate a database of recommended resources, especially since AI companies potentially lack subject-matter expertise to make these recommendations “and these resources could be incorporated into fine-tuning the AI’s responses to public health questions.” 

“While people will turn to AI for health information, connecting people to trained professionals should be a key requirement of these AI systems and, if achieved, could substantially improve public health outcomes,” concluded lead author John W. Ayers, PhD.

Study: Ayers JW, Zhu Z, Poliak A, Leas EC, Dredge M, Hogarth M, Smith DM. Evaluating Artificial Intelligence Responses to Public Health Questions. JAMA Netw Open. 2023;6(6):e2317517. doi:10.1001/jamanetworkopen.2023.17517 [Link]