If you are reading any healthcare related article, it is difficult to escape the exponential rise of articles about the application of Artifical Intelligence (AI) in healthcare. AI’s positioning is a panacea for what ails the fragmented healthcare system in the USA and many other countries. Does the use of AI live up to the hype? Should we proceed the with caution? The answer to both questions would appear to be maybe.
A new JAMA viewpoint by Ezekial Emmanuel from the University of Pennsylvania calls for caution around the potential use of AI in healthcare. Early use cases of AI in pattern recognition show promise with algorithms often outperforming human readings of MRI and Xray films for diagnosing cancer and Alzheimers. Emmanuel argues that healthcare is not suffering from a lack of data rather the levers to really change health trajectories from one of chronic disease to health lies in the science of behavior change. These big levers include physical activity, nutrition, stress management, sleep, and smoking cessation. Will AI be able to impact those issues? An additional gear would be medication adherence, but data is clear there is a myriad of reasons as to why people don’t take medications which in the USA today often includes the cost of the drugs. This complexity is reflected in the movement in the USA for healthcare systems to lean into the Social Determinants of Health as what happens outside of healthcare is often influencing how healthy someone is. The authors point out that only 50% of Americans are taking medications as prescribed, that’s a considerable margin of error and what is behind that other 50% is complex issues as referenced above. Behavior change is also challenging, we have many theories, frameworks, and techniques to show the DNA of how to impact behaviors that can positively or negatively impact health but again many studies don’t show the complexity of our day to day lives that make it easier or harder to be successful in changing habits. Behavior also doesn’t account for physiology that may be in the driver seat for some of the choices we make daily. AI can shed light on patterns and provide timely insights, but in itself will be insufficient.
While I agree with much of this viewpoint, traditional healthcare has often struggled to support people in managing their health, are programs that teach people the necessary behavior change skills offered frequently? In modalities that map to a person’s schedule? At a dose that is likely to yield proximal and distal outcomes? The answer is often no. A lot of what we know from clinical studies is at a population level, the promise of AI is analytics at the individual level- does our clinical knowledge at a population level translate well to a personal level? Again the answer is often no. The foundation of our clinical knowledge is about to experience tectonic shifts in data accumulation and analytics but are we in danger of a “garbage in-out” scenario if we don’t have reasonable hypotheses governing data frameworks?
My personal take is the next ten years will see massive shifts in our knowledge of disease states and the sequelae and data from genomics and the microbiome in addition to new discoveries in neuroscience are likely to impact treatment protocols positively.
Thanks for reading – Trina
(Opinions are my own)
Artificial Intelligence in Healthcare