Artificial intelligence is taking the world by storm, and its impact will be felt in all corners of society, including medicine. Is AI in healthcare a good thing or something to be wary of?
Here are four ways AI is transforming the medical field.
1 | Patient Care
The first way AI is changing medicine is by assisting with diagnosing and treating patients.
Misdiagnosing a disease happens relatively infrequently but it can occur due to factors such as physician fatigue, errors in diagnostic modalities, or limited resources in underprivileged areas. While the idea of AI taking over disease diagnosis may sound alarming, we should view it as a supportive tool for physicians.
For example, the convolutional neural network, or CNN, is a diagnostic modality that can analyze thousands of images from public datasets and patient medical records to identify patterns, enabling them to quickly and accurately diagnose diseases.
Researchers recently used CNNs to diagnose Kawasaki disease, or KD, an inflammatory disease of the blood vessels in children that can prove fatal if left untreated. A diagnostic hurdle with KD is that symptoms are usually vague and can overlap with other childhood illnesses.
To reduce the rate of misdiagnosis, researchers compiled images from KD patients from all over the world, building a CNN that can identify its common signs. The CNN proved both sensitive and specific for diagnosing Kawasaki disease, making diagnosis possible by merely taking a photo with a smartphone.
Of course, clinical judgment currently remains irreplaceable, and physicians should still rely on a thorough history, physical exam, and the relevant labs or imaging. These CNNs can help guide physicians in the right direction, especially in resource-restricted locales where it can be expensive for patients to do more thorough testing.
Preliminary efforts have also demonstrated AI can aid clinicians in diagnosing colorectal cancer, lung cancer, liver cirrhosis, and many more diseases. In one study, board-certified pathologists diagnosed colorectal cancer with 96.9% accuracy, and AI slightly outperformed them, reaching an accurate diagnosis 98% of the time.
However, doctors shouldn’t forget about the limitations of AI in diagnosing disease. Biases can exist in any of the datasets which can influence the way AI interprets the diagnosis. For example, if a dataset is mostly made up of older patients, AI might not be able to accurately interpret findings for a younger age group.
After diagnosis, AI can also aid physicians in treatment, particularly in the era of personalized medicine. Personalized medicine is a form of medicine that uses information about a person’s genetics to prevent, diagnose, or treat disease.
One specific example of this is rheumatoid arthritis, an autoimmune disease where the body attacks itself, especially in joints like the wrist and finger joints. Since rheumatoid arthritis is such a complex and chronic disease, there are a variety of medications used to treat patients, and treatment often depends on what each patient best responds to.
Researchers at Mayo Clinic used genetic data and patients’ clinical characteristics to develop a machine-learning algorithm to predict patient response to methotrexate, one of the most important rheumatoid arthritis drugs. Instead of having both doctors and patients wait months to determine the efficacy of a certain drug, these models can direct both doctors and patients toward more effective treatments immediately, saving the patient both time and money.
Similarly, researchers at the Georgia Institute of Technology and Ovarian Cancer Institute utilized machine learning algorithms to determine treatment effectiveness with 90% accuracy for certain chemotherapies in ovarian cancer patients. Utilizing AI to predict a patient’s response to chemotherapy can save valuable time and spare patients from the destructive physical side effects, emotional burden, and costs associated with weeks of treatment that may prove ineffective.
Beyond diagnosis and treatment, AI has powerful potential in predicting the occurrence and progression of chronic diseases like hypertension, diabetes, and kidney disease, which can help patients live longer, healthier lives.
Let’s take a closer look at diabetes. Using various machine learning models, researchers are developing predictive models to estimate a patient’s current glucose level based on multiple factors, such as their previous glucose levels, body-mass index, external stress, and even hours of sleep. This can help patients anticipate when their blood sugar levels are critically high or low, enabling them to better prepare for emergencies.
Researchers have also used AI to develop models that predict the likelihood of developing diabetes based on various risk factors. Early diagnosis and treatment of diabetes can prevent complications like diabetic kidney disease and blindness, which are emotionally distressing and costly for both patients and the healthcare system.
2 | Research
In addition to patient care, AI is also furthering medicine by transforming the way medical research is conducted, particularly in clinical trials.
Clinical trials study different interventions in patients, usually in the form of new vaccines or medications, to determine which treatment is more effective in clinical practice.
One of the first steps is identifying eligible patients, which is both time and resource-consuming since researchers must design and print brochures and manually screen clinics for eligible patients. With access to medical records, AI can quickly identify which patients fit the right criteria to streamline the process.
In addition, AI has already proven effective in the development of new treatments.
The first step in certain clinical research is discovering possible drugs. However, despite extensive lab testing, the discovery process can result in a waste of valuable time and resources. Many drugs that are effective in a lab fail in human trials.
Verge Genomics was one of the first companies that discovered a potential drug for amyotrophic lateral sclerosis, also known as ALS or Lou Gehrig’s disease, by using AI instead of animal or cell testing.
Instead of cell or animal data, they used AI to analyze human data points to provide researchers with more accurate representations of effective treatments in humans. This avoided the risk of drug failure when translating animal studies to human trials.
If you’ve experimented with ChatGPT, you’ve experienced AI’s writing capabilities first-hand. While these AI writing tools are far from perfect, they can significantly reduce the time needed to prepare and revise manuscripts.
ChatGPT has been used in scientific research and has even been credited as a co-author on multiple papers. However, it’s not without its limitations, including the hard to ignore fact that it can reference incorrect data and create fake citations.
Other software, such as Consensus, can be used to guide your initial literature review and provide summaries for papers that answer your research question.
Don’t expect these tools to take over your academic writing just yet; however, they can catch grammatical errors, brainstorm ideas, and collect and synthesize data.
We dug into the possibilities and drawbacks of using AI writing tools in another video on the future of medical school applications. If you think you can exclusively rely on ChatGPT for your medical school or residency applications, think again.
3 | Administrative Tasks
The third way AI is improving medicine is by streamlining administrative tasks in healthcare.
For example, BotMD is a company with an AI service that assists patients with clinical issues like finding physicians on call, scheduling appointments, or answering prescription-related questions, such as the availability of certain medications or their alternatives. Allowing AI to take over these tasks can free administrative staff to focus on other obligations.
AI can also simplify medical scribing since physicians don’t have to take their own notes or employ medical scribes. As opposed to human scribes who are subject to human error, AI works instantly and immediately understands medical terminology.
AI technology can pre-authorize insurance and optimize billing for physicians, since billing is reliant on accurate, consistent documentation. Reducing the burden of these tasks can help prevent physician burnout, a significant problem that leads to psychological distress for doctors and worse outcomes for patients.
4 | Medical Education
Lastly, AI is quickly being integrated into medical education at all levels of training.
For example, Oscer, an Australian medical education company, allows medical students to practice their history-taking skills on AI patients. Using this tool, students can learn to ask the proper questions and consider various diagnoses for any specific presentation.
Integrating AI more thoroughly into formal medical curricula can prepare future physicians for AI’s increasing role in healthcare. Some universities, like Duke and Stanford, have already introduced courses to help medical students and residents learn to use AI to solve healthcare issues. The Mayo Clinic and Stanford offer courses that teach physicians how AI is currently influencing medicine, as well as how they can use it to their advantage in their practice.
AI is even influencing the medical school application process, which we discuss in a separate article: ChatGPT and the Future of AI in Medical School Applications.
AI has huge potential to revolutionize many different facets of medicine, but with any change, we should be cognizant of any drawbacks or possible biases we could be introducing. AI, after all, is designed and programmed by humans, so it’s susceptible to the same biases we are.
As AI technology rapidly evolves in all areas of our lives, from medicine to personal wellness to writing, we’ll continue to cover emerging topics here on the Med School Insiders blog.
If you haven’t already, sign up for the Med School Insiders newsletter, which sends our latest videos, guides, study strategies, job opportunities, and music recommendations straight to your inbox once a week.