OpenAI: FemHealth Friend or Foe?
Written by Jillian Levovitz, MBA, Chief Strategy Officer, FemHealth Insights.
Healthcare delivery in the US is a constantly evolving landscape. With the often slow, but steady adoption of new technological capabilities, the options for where and how care is delivered have expanded greatly over the past several decades.
As is often the case though, figuring out the best way to integrate innovative technologies without compromising quality can be a complicated process. The repeated controversies around the quality of care offered by specialty telehealth providers that deliver medication such as NextMed and Cerebral are just one example.
Artificial intelligence (AI) chatbots such as Chat GPT are the next groundbreaking technology expected to revolutionize care delivery. With the launch of Chat GPT in November 2022, healthcare leaders began theorizing all the different healthcare applications this technology could be used for.
Now, programs are already being integrated into workflows to assist providers with administrative tasks including receiving prior authorizations through apps like Doximity’s beta product DocsGPT. Other products are being planned to assist with patient education and communication with a future look at diagnostic and treatment functions.
We cannot evaluate the potential of AI to improve healthcare, though, without first understanding the limitations of this technology and how it can result in low-quality care.
Compounding Outdated & Bias Information on Female Health
Chat GPT, and similar types of products, are trained on a vast selection of articles, websites, and social-media posts scraped from the internet, coupled with real-time conversations.
Unfortunately, the data collected from web searches to answer healthcare questions is not always accurate or current. In fact, many physicians and researchers have warned that medical citations used by Chat GPT are often incorrect and sometimes made up.
When considering the implementation of an AI program for female patients the challenges become even more pronounced. It is well known that the data used to train Chat GPT is all dated before 2021, which represents a severe limitation since healthcare knowledge and research is constantly evolving.
This is especially true when answering questions about female health since sex was only required to be considered as a biological variable in clinical research in 2016 and women still account for less than 50% of patients in clinical trials.
Additionally, AI can exacerbate challenges we currently face in how healthcare research is evaluated and surfaced as “top sources.” For example, many research sites use citation-based algorithms that evaluate publications based on the number of times other researchers cite them.
This approach, consistent with most used to evaluate research, is time-dependent and would not include recent studies. Older research that only tracks males, or studies that don’t take hormone fluctuations into consideration, could be used to build educational materials about and for females simply because they’ve been around for longer periods of time and therefore cited more often.
This means outdated, biased research will continue to drive knowledge of the female body instead of newer research that does account for those considerations.
A quick search for menopause research in PubMed, the NIH’s resource for biomedical and life science research, returned 10 top articles on average 6.5 years old with the oldest having been published in 2008.
There have already been 797 menopause articles published in 2023 but not one of them shows up on the first 10 pages of results in the PubMed search.
This includes a review of current treatment options for menopause symptoms and another evaluating the impact of menopause on cerebrovascular reactivity and brain structure. It’s safe to assume that patient education, treatment recommendations, or care algorithms built off older research would be out of date and even potentially harmful.
Finally, without a full understanding of conditions that differently and disproportionately impact women, the algorithms may again turn to studies without diverse patient populations thereby building bias into recommendations.
These educational materials or treatment paradigms would not account for variability in disease presentation between males and females, sex-specific management approaches, and appropriate symptom management.
Chatbots & Privacy
Another worrisome challenge inherent in the current configuration of AI chatbots is the lack of privacy users have. The employees and contractors working at these companies have complete visibility of all inputs to the software and questions asked. This feature would make it very difficult to create a HIPAA-compliant platform that protected its user’s privacy, a consideration that is more important to women today than ever before.
Looking to the Future of Generative AI & Healthcare
Taking all of this into consideration, we still see enormous potential for AI to improve healthcare delivery for women across the world. If care algorithms and educational materials can be purposefully built around current, sex, and condition-specific research, they can fill massive care voids due to the lack of female-specific care paradigms and knowledge.
As an example, provider awareness of female care concerns is alarmingly low. In the US, only 20% of OB-GYN residency programs offer menopause training and only 22% of physicians said they felt prepared to assess heart disease in women adequately.
As a result, females are not getting the disease-modifying or symptom-alleviating care they need for these, and many other, female health conditions from these providers.
Due to the growth of FemHealth startups, the efforts of advocacy groups within the female health and wellness ecosystem, and specific increases in funding from the NIH, we expect to see increased research in some female-specific conditions and those with a different or disproportionate impact. These new research initiatives and clinical trials will likely result in an even faster-changing treatment landscape that will be difficult for providers to stay on top of. Healthcare-specific programs that use AI-powered chatbots can address this knowledge gap and potentially provide better diagnostic and sex-specific care that is safe and comprehensive for all patients.
Lastly, with the ever-present threat of provider shortages, especially in rural communities and developing countries, well-developed treatment algorithms can help empower generalist providers, advanced practice practitioners, or even aid workers, to provide higher-level diagnostics and care delivery. All of which can lead to improved outcomes and quality of life for women throughout the world.
FemTech companies are beginning to use AI chatbots to support female health and wellness:
Ema — An AI-based chat tool that provides guidance and answers questions from fertility to menopause.
Woebot Health — A rules-based AI chatbot that offers mental health services with a focus on helping women suffering from postpartum depression.
About the Author:
Jillian Levovitz, MBA, is an experienced healthcare executive, the Founder, and CEO of OcciGuide, and Chief Strategy Officer at FemHealth Insights where she is responsible for consulting and research projects.
