Is my physician utilizing artificial intelligence to assess my condition during our consultation? Is he capturing our discussion to generate an AI summary of our appointment?
The application of AI in healthcare is still relatively new, leaving many individuals unsure about its implications. According to a 2023 survey, a significant number of Americans reported feeling “considerable discomfort” at the prospect of their physicians employing AI in managing their health care. However, AI is unlikely to disappear. The integration of AI technologies in medical care is on the rise, making it crucial for patients to comprehend its potential benefits as well as the reasons for ongoing caution.
I sought to learn how AI is currently facilitating diagnoses and guiding treatment decisions, what healthcare providers think about AI usage, and what concerns exist, along with efforts to resolve them.
To help answer these questions, I had a conversation with Dr. Leana Wen, a wellness expert at CNN. Dr. Wen is an emergency physician, an adjunct associate professor at George Washington University, and a nonresident senior fellow at the Brookings Institution, where her work includes exploring the intersection of technology, medicine, and public health. Previously, she served as the health commissioner for Baltimore.
What should patients understand about the various applications of AI in healthcare?
Dr. Leana Wen: First, it’s beneficial to differentiate between predictive and generative artificial intelligence, often referred to as AI.
Predictive AI employs mathematical models and pattern recognition to foresee outcomes. For instance, a predictive AI algorithm can ascertain which pneumonia patients are most likely to need hospitalization.
Imagine you are the patient. By analyzing previous cases of patients with similar conditions—such as pneumonia, diabetes, or heart disease—an algorithm could devise a care plan tailored to you, taking into account factors like your age, gender, other health issues, lab results, and ethnic background. The algorithm can assist doctors in determining whether hospitalization is necessary and which treatment options are most likely to be effective for your unique situation.
Generative AI, on the other hand, utilizes large language models to produce interactions that mimic human conversation. Many people might recognize ChatGPT, a form of generative AI designed to respond to user inquiries in a conversational format. Generative AI can condense vast amounts of information in a significantly shorter time frame than any human could. Research has indicated that certain generative AI models can acquire enough knowledge to pass medical licensing exams and generate clear, well-articulated patient instructions on various topics.
However, there are concerns that these models may “hallucinate” and provide responses that are misleading or incorrect. Additionally, the efficacy of both predictive and generative AI depends heavily on the data on which they were trained. When evaluating the usefulness of AI in healthcare, it’s essential to assess each AI tool individually and understand its development process and intended usage circumstances.
In what ways is AI already assisting in diagnosing patients?
Wen: There are several well-established examples of predictive AI being used to enhance and improve diagnoses.
Consider a colonoscopy, which is the standard procedure for diagnosing colorectal cancer. During this procedure, the physician inserts a long tube into the colon to manually identify and remove polyps that may be cancerous or precancerous.
AI can be trained to detect polyps and alert physicians during a colonoscopy. Numerous randomized controlled trials conducted in healthcare facilities worldwide have demonstrated that AI-assisted colonoscopies significantly decrease the likelihood of missing potentially cancerous lesions.
Likewise, AI is being utilized to aid in interpreting mammograms, which are crucial for breast cancer screening. Research has shown that AI-supported mammography screenings are at least as accurate as having two trained radiologists examine the images and may even enhance cancer detection while decreasing the workload of clinicians. The US Food and Drug Administration has authorized around two dozen AI products for mammogram cancer screening; however, their adoption has been slow due to the potential additional costs associated with implementing them in clinical settings.
How is AI being utilized to guide treatment?
Wen: One application is a predictive AI algorithm created by researchers at Johns Hopkins University that identifies hospitalized patients who are at a high risk of developing sepsis, which is a severe body-wide infection that could result in multi-organ failure and death. This early-warning mechanism has been implemented in various hospitals and has proven to decrease the time taken to identify sepsis and consequently to initiate antibiotics and other interventions.
Kaiser Permanente has also implemented a predictive AI tool that monitors indicators of patient decline in hospitals. If data from a patient’s vital signs, lab tests, nursing reports, and additional factors suggests a deteriorating clinical condition, the system sends out an alert to facilitate a prompt evaluation by an in-person clinician. This tool has been linked to a notably reduced mortality rate.
How can I discover how my provider’s office utilizes AI for diagnosis and other functions? What measures can I take to safeguard my privacy?
Wen: Much of what is currently termed AI has been in use for quite some time. Predictive algorithms have long been employed to assist in customizing treatment plans, for example. Medical offices are increasingly using AI to aid in drafting email responses and to support their documentation tasks.
It is reasonable to assume that your provider’s office employs some form of AI in your treatment. You may inquire with your provider and also refer to the policies that the office requires you to sign, which might include permissions to utilize certain technologies.
Your medical records are protected and secured by the Health Insurance Portability and Accountability Act, or HIPAA, which legally ensures the safeguarding of sensitive health information.
What are providers’ perspectives on the application of AI? Does it enhance efficiency and lessen the paperwork burden?
Wen: I believe clinicians have been particularly pleased with AI tools that alleviate documentation and paperwork tasks. For example, generative AI is being employed to assist with medical documentation through a technology referred to as ambient AI. Essentially, the system “listens” to the interaction between the doctor and patient and then transforms it into a medical note that the doctor can modify. Research has shown that this ambient AI scribe shortens note-taking duration and is positively received by both physicians and patients.
Physicians are also leveraging generative AI to aid in the creation of prior authorization letters that must be sent to insurance companies to obtain approval for certain medications and treatments. This can help mitigate an office’s administrative strain and potentially reduce health care worker burnout.
Is it possible for insurance companies to employ AI to deny claims?
Wen: It certainly could be. One could envision that the presence of AI could create a significant obstacle in accessing a human representative within an insurance company to discuss claims and other matters.
This represents one of numerous concerns associated with the use of AI. Other worries include maintaining privacy and data security, as well as the necessity to independently verify algorithms and transparently disclose these findings. Technologists, clinicians, and regulators should consider the numerous beneficial uses of AI in health care while ensuring that each tool is rigorously examined and deployed with care.
