Artificial intelligence, or AI, has become a frequently used term in the news and popular media, sometimes with approval and sometimes with fear, depending on the prejudice of the author. But AI is simply tool or machine to aid us in the accomplishment of some task or solving of some problem. The purpose of this article is to introduce you to what AI is, how it is currently being utilized in health care, what might expect to see in the future, and the limitations of AI.
What is AI?
In its simplest expression, AI is technology that simulates human intelligence in decision-making and problem-solving situations. An easily recognized form of this is an internet or retail search engine (think Google or Amazon) which will make suggestions to the user based on a handful of typed letters or words, sometimes even if misspelled! The outputs are not just based on one user’s typed search but a collection of many other searches, and of advertiser’s paid suggestions that produce a particular set of results. Another form of AI is used with GPS devices, whether on a computer, phone, or car, which will make suggested travel routes based on starting and ending points, and other limiting criteria like “no highway travel.”
It is important to note here that AI’s “suggestions” are not the product of actual, human thinking but are rather the accumulation of common data inputs that are then presented as reasonable outputs based on specific paraments and the selected criteria. In the two examples above, AI is very helpful is directing users and consumers to the most likely result based on millions of similar searches made by many users in many places. It is also important to note that AI is not a single device or machine but rather is a class or type of “machine learning” being used in many different areas and disciplines across the globe.
What AI is currently doing in healthcare
Because AI is not a single entity, it is being utilized in the healthcare industry in a number of different areas. Its primary usefulness is doing time-consuming and tedious analysis, thus freeing medical personnel to work more personally with their patients, utilizing AI’s outputs as part of their care. For example, AI can be employed to fill out forms or take clinical notes, even if these documents need to be double-checked by a human medical worker for refinement and/or correction. Virtual, on-line assistants, medical chatbots, and telemedicine programs can also be AI-driven to ensure patients are directed to right types of actual human healthcare providers.
In the area of preventative care, AI can analyze multiple radiology images in order to identify potential problems. This would include mammograms, kidney, and lung screenings. Normally this kind of work would take healthcare providers minutes or even hours to do, carefully reviewing multiple images to assess the situation and make a correct diagnosis. AI can analyze such images in a matter of seconds, allowing the healthcare provider to focus on patient diagnosis and best treatment options.
In the area of risk assessment, AI is being used to identify heart problems even when there are no noticeable symptoms. One AI program is being used to predict heart attack risks based on coronary arterial calcium, and another has successfully identified colon polyps and risk of colon cancer better than an endoscopist could by themselves.
In the area of research, AI programs can analyze medical journals, published academic papers, laboratory reports, and even anonymized medical records to find the most pertinent and helpful information to medical professionals trying to decide on the best course of action for their patient.
What AI might be doing in the future of healthcare
As we’ve seen, AI is already being used in a number of areas in the healthcare industry but the potential for growth is very promising. Here a few areas where AI might be even more helpful in the future of healthcare.
Selecting and matching patients for the most promising clinical trials: AI-powered analysis can quickly and easily find candidates for clinical trials of new drugs and/or procedures. Rather than doctors submitting patients’ names into a large database, which must be reviewed and approved by clinicians, AI could select qualified patients, allowing the selection and approval process much faster.
Monitoring of health devices: AI never sleeps so devices like pacemakers and AID (Automated Insulin Delivery) devices could alert medical professionals in real-time when there are emergencies or equipment/mechanical failures with the devices themselves.
More advanced disease detection and diagnosis: Similar to monitoring of health devices, AI could not only alert medical providers of primary problems in a patient, but also be “trained” to look for underlying causes/reasons for those problems, especially in higher risk patients where multiple comorbidities complicate healthcare.
Personalized disease treatment: Because AI systems can “learn” and “remember”, disease treatment need not be limited to generally known behaviors and characteristics of a given disease, but can be custom-tailored to a patient’s medical history, personal preferences, and other physical or mental needs.
Enhanced imaging: We have already seen AI’s uses in the analysis of medical imaging but this is being further utilized by clinicians in preparation for certain types of surgery, which in turn reduces the overall time between diagnosis and treatment.
Clinical trial efficiency: Quoting from an IBM article on the uses of AI in medicine; “A lot of time is spent during clinical trials assigning medical codes to patient outcomes and updating the relevant datasets. AI can help speed this process up by providing a quicker and more intelligent search for medical codes. Two IBM Watson Health clients recently found that with AI, they could reduce their number of medical code searches by more than 70%.”
There are many more potential uses AI could have in the area of healthcare and no doubt new uses will be found as medicine and technology continue to improve patient care.
The Limits of AI
As a helpful tool for data analysis and even prediction, AI will very clearly be part of the healthcare industry for the foreseeable future. However, we should also be aware that AI is not going to replace doctors, nurses, and other human healthcare workers anytime soon. AI is a very effective analytical tool but data analysis is not the same as diagnosis and treatment. Trained medical staff cannot replaced by AI, as patients are individuals each with a unique set of medical conditions which requires the attention of a human healthcare provider who can address the whole person and not as a dataset.
The primary limitation of AI is that of available data; AI requires data in order to make good recommendations. Less data means a narrower sample size, and therefore potentially less accurate or biased results. As an example, it is possible that an AI chatbot or Teledoc application could successfully rule out certain medical diagnoses but could not accurately identify a given medical problem. The patient would be required to physically go to a facility to be correctly diagnosed. A very similar situation exists when individuals search for medical advice online and self-treat, yet do not follow up with an actual healthcare provider. This is not a fault of AI or the internet, but is a real limitation.
Medical record digitization, or rather its absence, is another potential limitation. The digitization of these records will likely continue for some time, which will improve datasets and therefore AI’s effectiveness, but the accuracy and completeness of such records is still a factor. In addition, privacy issues will remain extremely important and how to make the data available while maintaining privacy will no doubt occupy thought-leaders and those developing AI tools in the healthcare field for some time.
Having access to healthcare is another corollary of the data limitations since there are segments of the population that either do not have use of, or do not make use of, the healthcare system in a regular, systematic way, so data limitations will prevent AI from being as useful as it could be for certain segments of the population when they do require medical care.
The Risks of AI
The limitations of AI can also become real risks. In reference to already-mentioned medical record digitization, a NIH article points out that; “2018 saw Google acquire DeepMind, a leader in healthcare AI. When it was discovered that the NHS had uploaded data on 1.6 million patients to DeepMind servers without the patients’ consent to construct its algorithm, Streams, an app with an algorithm for treating patients with acute renal impairment, came under criticism. A patient data privacy investigation on Google’s Project Nightingale was carried out in the USA. Data privacy is now much more of a problem since the app is now formally hosted on Google’s servers.” The potential for such information, though necessary for AI to be of help, to be exploited is very real.
There have also been class action lawsuits filed against healthcare companies for using AI algorithms to wrongly deny health insurance claims. Appeals to claim denials are not new but those involving AI are, and will likely require a combination of new legislation and technical specialists that can properly adjudicate such claims.
Conclusion
AI is currently a very helpful tool for the healthcare industry, enabling medical professionals to “outsource” some of their more tedious work to a machine, allowing them to be more productive, and more focused on whole person care. AI itself is able to “learn” and the potential uses of AI for the future are varied, even more than the few topics discussed in this article, and despite the limitations and risk of such technology. But for all its usefulness, AI will not replace healthcare workers and the care and expertise that they can and do bring to their patients.
Looking for help with insurance? Schedule a phone or video call with us today! https://calendly.com/ga-health-insurance/daniel-meeting
Sources:
https://www.ibm.com/topics/artificial-intelligence-medicine