Artificial Intelligence Has Come to Medicine. Are Patients Being Put at Risk?
Mar 11, 2020
Artificial intelligence has been readily adopted by industries including technology, automobile, and even the entertainment industry. Although AI has grown steadily in those industries, it has not been as easily accepted in the healthcare industry, due to numerous false alarms. For example, an AI system at a London hospital produced two false alarms, for every correct result, leading to unnecessary test for some patients and the withholding of treatments for others. Still, AI has more to offer than what its critics envision. According to the American Cancer Society, a high proportion of mammograms yield false results, leading to 1 in 2 healthy women being told they have cancer. The use of AI is enabling review and translation of mammograms 30 times faster and with 99% accuracy reducing the need for unnecessary biopsies. While much concern revolves around AI’s predictive accuracy, AI has incredible potential in the healthcare industry, and its accuracy is expected to increase gradually as algorithms get trained with more data over time. In addition, AI can be even more precise and beneficial when used in tandem with electronic health records.
The level of attention and human interaction that a doctor requires before diagnosing their patients is tremendous. As of yet, computerized systems have not been able to treat patients without input from doctors and physicians. This may be due to a lack of randomized clinical trials where AI healthcare technology has been deployed, as AI developers have little to no interest in performing costly trials. Even amongst the trials that have been conducted, the computer lab-tested results differ from real life results as there are too many false positives in the computer output. As opposed to other industries, where AI can be utilized even with a moderate accuracy metric, medicine and healthcare have a great deal of repercussions with false positives. In addition, there are legal concerns with the implementation of AI. For example, if there is an unanticipated, false, or damaging outcome, it is unclear whether the responsibility falls on the physician or the manufacturer of the AI technology. Almost every AI system requires a plethora of information about the patients, leading to HIPAA privacy concerns with the sharing and storing of such data. Medicine is regulated at the state level through licensure by the boards of medicine, so the question remains as to whether additional licensure or regulatory scrutiny would be required if the devices supplant all or part of the physician’s practice of medicine.
On the other hand, many experts have an optimistic view on the potential of AI in the field of medicine. According to a research article from Scientific American, patients with a high risk of death within 48 to 72 hours of hospital admission can be identified with a high level of predictive accuracy, thus enabling clinicians to take proactive measures to treat them in ways that mitigate further risk. On top of saving a patient’s life, this can add to significant cost savings for the hospital. This positions AI as a complementary technology that can help improve the performance of doctors and physicians to do their job in their field. In addition to that, AI is likely to be a beneficiary from collecting Electronic Health Records (EHR) with patient’s data. By accumulating data on patient history, current condition and previously prescribed medication, AI systems can classify patients based on similar characteristics without having to go through a longer, more costly diagnosis. A hospital’s EHR can connect with the EHR of other hospitals to create a “health information exchange” (HIE). This can help providers get access to all the patient’s medical information regardless of where he or she may have been treated. The purpose of providing access to a patient’s comprehensive medical information is to decrease medical errors as well as eliminate the provision of duplicate tests or procedures that may occur when a specialist may not have access to another provider’s records. Once enough high-quality data from HIEs is present, it will likely increase the accuracy of an AI model’s predictions. The more we digitize and integrate our medical data, the more we can use AI to help us find insightful patterns – patterns we can use to make precise, cost-effective decisions in complicated analytical processes.
The largest presence of AI within the medical domain lies in diagnosing diseases, accelerating the process of drug development, personalizing treatments and improving gene editing. This, however, is just the beginning, as there are still many areas where AI has not placed a foothold. When artificial intelligence is created along with thorough training and careful testing to ensure pinpoint accuracy, there is incredible potential for its utilization in the health field.