The increased viability of machine learning and AI for high accuracy applications has made it a major interest in the Healthcare field in years past. With growing population sizes and limited healthcare provides available, especially during times of crisis, researchers and medical networks are working on implementing AI-driven “doc-bots” for primary care.
Since the start of COVID-19, these chatbots have already been used to help millions of people diagnose their symptoms and get information on steps to take next. One provided by Sutter Health has diagnosed more than 30,000 people, determining if they have COVID-19 or not, and referring 11% of them to a physician. Whether the bots recommend patients to self-isolate or go see a doctor, or have a video consultation with a doctor, the strain on the healthcare system is being reduced.
Amwell survey data shows increase of video visits and general consumer satisfaction with remote care.
The benefits that AI could bring to healthcare are numerous, from faster round-the-clock assessments to differential diagnoses to more affordable primary care. But there are still several hurdles to overcome before we’ll likely see widespread adoption of AI bots.
Firstly, the doc-bots need to provide proven high-accuracy diagnoses that are consistent with diagnoses that physicians would provide. Both doctors and patients are hesitant to adopt AI if the algorithms can’t provide results that are at least as good as that of doctors. One AI system developed by researchers at the University of California and Google manages an accuracy of 75%, which is high but not particularly confidence inducing. Furthermore, the diversity of people could result in poor accuracy for minority groups that were underrepresented in the training of AI models.
Secondly, they need to be integrated into the whole healthcare system to ensure that the AI applications are a full part of the system and not just a hurdle for patients to overcome. When doctors perform diagnoses they consider a multitude of factors about the patient’s current condition and medical history. They then prescribe medication, tests, and treatments. In their current stages, medical bots have little to no access to medical backgrounds and the most they can prescribe is a consultation with a physician. This already helps but is really an under-utilization of their full capabilities.
Lastly, there needs to be a financial incentive to offer and use software-based diagnoses. At the current time, insurers don’t reimburse services rendered by medical bots. Even if AI bots can provide efficient and affordable service, health systems require the service to be financially viable to maintain revenue streams.
Overcoming these challenges will be a slow and iterative process but in doing so will provide massive improvements to the healthcare systems of growing countries around the world.