Season 2 - Episode 2

Helpful Links

We wanted to take this theme and highlight an AI tool for patients who need assistance writing insurance appeals. The website is Fighthealthinsurance.com. This tool assists people in writing insurance appeals that could directly combat bills that were deemed the patient's financial responsibility. If you believe you were wrongly billed for certain services, checking out this tool could assist you in writing an appeal.  

Transcript

Hello, welcome to mirroring medicine with Kodi and Isabel! Today we will be discussing the advancement of AI and medicine. This includes going over the intricacies of AI, how AI is currently discussed in medicine, and AI as a potential tool in medicine, or even a potential detriment to human interactions. This podcast is meant to provide an overview and it is not comprehensive, we will be going over the tip of the iceberg in regards to this topic and we recommend the audience researching this topic even further. Additionally, this podcast does not reveal medical advice, if you would like to inquire further, please contact your doctor. 

What is AI?

AI is also known as artificial intelligence. This is a common buzzword thrown around nowadays, but what makes it so interesting? There are multiple AI systems out there but people might have heard of the generative AI system ChatGPT. ChatGPT is one of the main AI systems like Gemini on Google or the new Siri AI system. According to IBM, “Artificial intelligence (AI) is technology that enables computers and machines to simulate human learning, comprehension, problem solving, decision making, creativity and autonomy.” Additionally, “generative AI (gen AI), a technology that can create original text, images, video and other content.” [1]

What is the nuance behind AI?

AI is different from other models like machine learning and deep learning. Machine learning is a way that models can be predicted through bootstrapping and an algorithm. By using previous information, it can use an algorithm and bootstrapping by replicating and producing something from data and generating new forms of data from old ones. People can use bootstrapping through this tool as well. This includes repeated drawing from a set of data (with replacement into the set or without replacement into the set of data)

So what is deep learning? Deep learning is the idea from machine learning that includes formulating a kind of a brain like algorithm, where it has deep neural networks and complex connections between different sources of information, this idea came into fruition shortly after the creation of machine learning. Furthermore, it actually enables this process where these computers can make their predictions about what the data represents.

There are different types of learning under the umbrella of deep learning. This includes: semi-supervised learning, self-supervised learning, reinforcement learning, and transfer learning. 

Generative AI is the most modern type of system of artificial intelligence that we have to date, this doubles down on deep learning models and evolves this system to produce models that can create original data and content. 

Generative AI works by creating a foundation model, adapting and applying this model, and improving accuracy on this model. This is summarized in three steps: the first step being the training phase, the second being the tuning phase, and the final is the generation, evaluation and more tuning phase. With these three phases, it allows for this generative AI to learn from mistakes and improve its platform constantly. 

Ethically, according to IBM, it shows explainability, interpretability, fairness, inclusion, robustness, security, accountability, transparency, privacy, and compliance all need to be considered when understanding the intricacies in this booming field of technology. 

We wanted to take this theme and highlight an AI tool for patients who need assistance writing insurance appeals. The website is Fighthealthinsurance.com. This tool assists people in writing insurance appeals that could directly combat bills that were deemed the patient's financial responsibility. If you believe you were wrongly billed for certain services, checking out this tool could assist you in writing an appeal.  

AI in Medicine

While AI has infiltrated all sectors of our technologically advanced world, one interesting aspect of this application is the integration of AI in medicine. This includes making diagnosis, plausible treatments, and development of screenings for certain diseases. Beyond that there are some applications of data mining that allow large quantities of data to be analyzed into trends, providing new information on data. This includes early symptoms of diseases, demographic groups at risk, etc from thousands of patient chart samples as an example. 

Interestingly, since AI is self learning the more time that passes and as it is used more frequently in this sector, the more accurate and detailed this application can become. 

And how does this work? Most AI systems use free information that is published on the internet and instantaneously integrates information from multiple sources into a concise summary. Allowing one to get a brief overview about health conditions, diseases, etc.

Importantly, this tool will benefit the lives of patients and increase efficiency of medical diagnosis. It is unlikely that this will replace physicians, or at least replace complex specialties such as surgery, critical care, obstetrics and gynecology, because of the need for human decision in these acute and time sensitive fields. Split second decisions need to be made, and it will not be possible for a provider to input all patient information into an AI system to get an answer in a split second. 

However, there are some specialties that are expected to be likely replaced or downsized by AI, this includes, primary care, pathology, interventional radiology, radiology, infections diseases, etc. I want to point out that importantly, there are aspects of medicine that cannot be replaced by AI due to the human experience. Medicine requires empathy, compassion, consideration of socioeconomic status, radial inequities, gender inequities, discrimination, prejudice, religious tolerance, and many other factors affecting the lived experience which is only possible to have due to being a fellow human. 

I wanted to provide an example of a real world application of artificial intelligence to the field of predicting disease. On Jan 08, 2025, published in the Journal of Clinical and Experimental Hypertension, researchers in Japan looked at roughly 16,000 patients in a retrospective machine learning study trying to decipher if in fact AI would be able to predict if these participants would develop hypertension (high blood pressure) [2]. They divided the same into two categories: 70% of participants were part of the learning group, and 30% were part of the test group. Looking at 58 qualities in these groups, the most meaningful quality was concluded to be fatty liver index (FLI) which is a non-invasive tool to predict if a patient has non-alcoholic fatty liver disease which is the most common form of chronic liver disease [3]. This disease is often associated with metabolic syndrome, which is a list of multiple characteristics that raise risk for coronary heart disease, diabetes and stroke [4]. 

The fatty liver index is calculated using BMI, waist circumference and levels of γ-glutamyl transferase (an enzyme found in the liver that indicates high levels of damage to the liver [5]) and triglycerides (the fat in blood). This study found that machine learning systems were very good in predicting hypertension in individuals when looking at their systolic blood pressure (the pressure of the blood acting against the walls of the artery), age and fatty liver index (FLI). The model was able to predict the onset of hypertension with values from 76.5-82.5% of the time and the levels of accuracy were >0.8. 

Overall, this study demonstrated how useful machine learning and AI can be in predicting disease. These implications in medicine are important and will provide better treatment prevention options. 

Citations

1. IBM. "Artificial Intelligence." IBM Think, IBM, ⁠www.ibm.com/think/topics/artificial-intelligence⁠. Accessed 20 Jan. 2025.

2. Tanaka, M., et al. "Machine Learning-Based Analyses of Contributing Factors for the Development of Hypertension: A Comparative Study." Clinical and Experimental Hypertension, vol. 47, no. 1, 2025, pp. 2449613. doi:10.1080/10641963.2025.2449613. Epub 8 Jan. 2025. PMID: 39773295.

3. Khang, A. R., et al. "The Fatty Liver Index, a Simple and Useful Predictor of Metabolic Syndrome: Analysis of the Korea National Health and Nutrition Examination Survey 2010-2011." Diabetes, Metabolic Syndrome and Obesity: Targets and Therapy, vol. 12, 2019, pp. 181-190. doi:10.2147/DMSO.S189544. PMID: 30774403; PMCID: PMC6353218.

4. National Heart, Lung, and Blood Institute. "Metabolic Syndrome." National Heart, Lung, and Blood Institute, ⁠www.nhlbi.nih.gov/health/metabolic-syndrome⁠. Accessed 20 Jan. 2025.

5. Cleveland Clinic. "Gamma-Glutamyl Transferase (GGT) Test." Cleveland Clinic, ⁠www.my.clevelandclinic.org/health/diagnostics/22055-gamma-glutamyl-transferase-ggt-test⁠. Accessed 20 Jan. 2025.