One of the key issues with implementing AI solutions in healthcare settings is the fact that these algorithms are often in the cloud and the medical center data would need to leave the institution to be examined using the algorithm. Security and privacy of patient data that may be put at risk because of the movement of data in and out of the cloud where the algorithm analyzes the data. Many institutions are not comfortable with the routine traffic of patient data out of their firewalls due to security and privacy issues. As such, this can serve as a barrier for them in adopting AI technologies in the short-term
For decision support systems, this means that the AI output should be getting information and advice to physicians instantly while they’re in front of patients. If the data that the algorithm needs to generate recommendations in a real-world clinical setting is fragmented, not yet entered into the system, or initially unstructured, it would not be able to provide its output in a timely manner. These types of issues can make the algorithm obsolete.
Everyone should acknowledge that there will be a certain degree of resistance among clinicians when considering AI. Much of this may stem from the challenges clinicians faced during EHR implementations in the prior decade. Clinicians will be hesitant to adopt tools that they perceive as trying to replace them (though most experts believe clinicians have little cause for concern) or that will end up increasing their workload. Moreover, clinicians will also have deep concerns about the validation of AI recommendations and will need to evaluate the clinical use of AI in much the same way they would a new drug therapy or diagnostic tool. Approaching AI adoption as if it were a therapy will help create a model for clinician acceptance.
One of the most heated debates is about whether AI will be replacing large portions of the health care workforce. Estimates vary on the overall impact. The Royal College of Physicians projects that actual job loss could be less than 5 percent, while Gartner predicts that AI will lead to the creation of 2.3 million jobs in 2020, more than the 1.8 million it is estimated to replace.
Another workflow and implementation issue is the training of an algorithm developed on external data on local data to ensure that it will perform as expected on the local patient population. Selection of that data and additional training of the model will take work and create extra friction. Friction means less likely short- or medium-term adoption by the health systems and clinicians.