AI applications are becoming increasingly common as the current AI wave shows no signs of slowing. There are many aspects of work that an AI system, tool or application can improve, whilst retaining the human aspect required - the primary example being using AI as an assistant rather than allowing it to take the wheel, so to speak. This article aims to identify potential benefits of AI in healthcare systems today, where the challenges and pitfalls arise, and wraps up by taking a look into the future.
Currently, there are many AI programs in use in the medical field to assist healthcare professionals with their work. Machine learning tools are being used to assist in more accurate cancer diagnoses, as well as early diagnosis of fatal blood diseases via AI-enhanced microscopes. On a larger scale, production of medicines and new developments has become much easier with the help of artificial intelligence. On the customer side, AI has revolutionized the first phases of medical care - early patient interaction and office visits. Two tools are responsible for this; the first being customer service chatbots. These chatbots are important in offloading extra work that healthcare professionals would otherwise be burdened by, as the bot can suggest courses of action to the potential patient and attempt to resolve their issue before professional help is required, therefore reducing potential workload immediately whilst reducing the chance of a needless appointment.
A step up from a basic chatbot, Virtual Health Assistants are another useful AI technology for assisting patients, both current and potential. They are responsible for responding to the queries of routine patients via calls and emails, sending follow-ups and clinical appointment reminders to the patients, managing medical information of the patients, scheduling appointments with doctors, etc. These technologies, alongside telemedicine, are at the forefront of digital health today. Whereas this use of AI may not impact healthcare professionals’ workloads as much as the customer service bots, it is still an extremely useful way to assist the patients.
Having discussed the current landscape and the individual scenarios, what are the general benefits of AI systems?
By using AI tools, the accuracy and efficiency of certain processes improves. Humans are prone to bias, especially in healthcare - it is entirely plausible that things go undetected. They will prioritize examining areas that line up with their previous cases; known weak points in the human anatomy. While this is a good way to go about finding what they are looking for, AI can help to examine everything else at faster speeds. A duo of a human and AI is much more powerful than a human alone, combining their expertise with an AI's almost perfect analytical skills. AI can help with precision medicine also, taking into account many other factors from the information available about the patients everyday life to assist with their general health. Applying AI to any medical device can help produce more accurate and efficient yields of information.
Laborious tasks are another hurdle AI can help phase out. Using AI as a teammate can eliminate tasks that are time-sinks; for example, Kristin Yakimow, MSc, MBA explains on the Segmed 'Bytes of Innovation' webinar that this AI teammate can assist processes and creation of important documentation to aid healthcare professionals in reducing time spent on writing - and boost efficiency by enabling them to work on more pressing matters. This can also include administrative tasks as well as reporting, data security and electronic health records - anything that involves long texts and/or sensitive data can be streamlined relatively easily, boosting general and clinical workflows. The healthcare industry is constantly looking for ways to refine their processes, and AI, machine learning and deep learning can be an extremely promising way to achieve that goal, whilst maintaining or even improving public health.
A general improvement that comes with integrating artificial intelligence is the health care provided to patients becoming an easier, smoother and quicker process. Centralizing a person's electronic health record on the same platform as the AI tools and implementing a neural network could help to minimize disruption and prevent delays usually incurred in current systems. Introducing AI to perform data analysis and providing information based on historic records that can help with diagnosis.
Clinical care can improve with the application of AI, and in turn the health care system would massively benefit from the implementation, not only by reducing the time spent in clinical care, but by reducing time spent on billing, documentation and scheduling thanks to machine learning.
Not only is the care experience improved but also accessibility to said care. Healthcare providers can use tools such as the aforementioned chatbots, but also Telehealth, Telemedicine and mobile applications to vastly improve the feedback rate on many of the aspects of care. Using AI in a centralized platform, such as a healthcare application, allows the healthcare providers to send critical information, results, and copies of notes directly to the patients. By moving this aspect of healthcare to mobile phones and powering them with AI, it can greatly reduce the need for face-to-face communication, whilst still allowing it when necessary.
One thing that is important to remember is that there is always a need for lower costs - whether this is for the hospital, healthcare provider or the patients, it is always beneficial. Using AI can aid this; for example, drug administration can be automated based on an AI program that orders what is necessary from pharmaceutical companies based on a patient's needs. This would allow the individual institution (hospital, clinic etc.) to order the necessary amount without overspending and being left with a surplus - although this can obviously be tweaked to allow a certain number of doses to be available at all times as a buffer. Boosting the turnover of patients also benefits institutions, as this will increase the number of patients they can see. In fact, Harvard researchers estimate AI could save the healthcare industry around $200 billion to $360 billion a year (5-10%) if implemented into healthcare sufficiently.
A primary pull of the integration of AI technology into the healthcare system is the data analysis and data science potential. The multi-tool nature of artificial intelligence means the application is very broad - from image analysis to text generation, to predictive analytics - meaning various aspects can be enhanced, such as data analysis. Institutions house significant volumes of data and finding ways to make that data useful without skyrocketing storage costs is imperative. An AI system dedicated to data analysis can help to retain data sets necessary for analysis whilst discarding data that cannot be used. The most probable use for AI in data analysis is in medical imaging; using image data and image recognition to train AI and students, analyzing images to find discrepancies a radiologist may have missed, and formatting the data to remove PHI and allow the data sets to be available for use.
Unfortunately, the benefits of introducing AI into healthcare aren't as cut and dry as they might seem. Here's some of the potential difficulties:
Including investment and its cycle, and the warnings and pitfalls associated with implementation; this can be best summarized by Yakimow herself - “is the juice worth the squeeze?”. Currently, there is pushback from many healthcare professionals, as is the case with any new changes to one's work systems due to the need to relearn how they operate, with a slight tinge of personal bias surrounding the potential risks of AI impossible to disprove, and probable in its existence. Not only that, but the investment is a serious risk this early in the adoption process. Emerging technologies often suffer from a lack of trust, and as such the implementation will be slow at first.
As it stands, the human element is still present in any task an AI is designed to assist with; today's technology has not advanced enough to reach the point of full, unflawed automation - and its results should never be the sole medical diagnosis. However, the inclusion of any artificial intelligence is sure to upset some people, who may feel more comfort in putting their care in another human over a machine. Chatbots and VHA (Virtual Health Assistants) being the front-end of people's experiences in healthcare is a concern due to the amount of interaction with a non-human entity, such as an AI - some people will not embrace the change at first, but its implementation will take time. AI technology is not perfected, especially for human conversation; GPT models have been examined creating false information, such as the recent case involving lawyers who quoted a court case that didn't exist. This may not be a problem for other roles, however. Tools purely to assist a human in their work, such as image analysis in radiology or speech recognition via notation, only serve to improve accuracy whilst still being monitored by the healthcare professional.
Although more of a contract issue in this situation, data privacy rules are a risk for any potential implementations. Tools are going to be 3rd-party the majority of the time; institutions will have to make sure their privacy rules, state laws, federal laws, security standards and the program itself all align. This headache is more noticeable if the applications or tools aren't from American companies as country data security laws vary wildly. Also, the use of image data, and any other data on humans for that matter, has to include no PHI (Protected Health Information) by law. These problems may incur extra headaches at the contract stage.
Currently, there is still an immediate need for a human element in many of the areas AI can assist. As it stands, there's no replacement for an actual human in many roles - however, this may become an unfortunate consequence as AI evolves. Potentially, virtual assistants powered by AI could be an early replacement for those on the front-desk, but the future is unclear. In terms of patient information, if image data or data with PHI is not treated right, there is a chance that PHI becomes available to the public. This is predominantly an issue if the data is exported outside of the institution, perhaps for training or selling anonymous data sets, and would cause serious repercussions if it were to happen. Ethical considerations are also a sticking point for many in the argument against implementation, adding to the problems with introducing these technologies.
For the benefits and challenges to be properly observed, they actually have to be used; meaning implementation has its own pressures and requirements to keep it running.
The chasm of adoption in new implementations is a tough hurdle to overcome. Designated as the gap between the early adopters and early majority of users, the chasm of adoption is wider the less trust, documentation, and proof there is.
If the tool or application isn't built with solid documentation and communication to iron out the required needs, companies and institutions will not want to risk using it in their systems. This new technology will have to justify its implementation, and the down-time associated with not only implementing it, but having employees learn and understand it.
In the case of machine learning and deep learning AI technologies, training them based on the needs of the institution is crucial. If the format of the data fed to the system is different than the data sets it will be analysing, the system will struggle to analyse it. Uniformity and similarity are tantamount to any other variable in creating new AI technology; and if it does not work in its new environment, it is not worth the implementation time. Clinical trials with new systems are a good final step to ascertaining whether it is ready for general use, as patient safety should always be the priority.
Even before the AI is implemented, a strategy must be laid out. It is important to question whether the move is necessary, whether it will aid the institution and if it is worth the down-time and learning time. A clear goal for what the technology should achieve and keeping a close eye on the statistical data coming out of its use is imperative to ascertaining the usefulness, accuracy and efficiency of the AI systems. This may not be quantifiable for certain situations; however, user experience is just as important as pure statistics. Regular checks with users of the programs and survey work will help to document and observe the impact that AI has on workflow and processes.
With the current furore around AI, it has proven that the concept is still alien to most of the population. By informing and educating people on the applications, risks and potential of AI technology, distrust in AI will surely lower. Tackling misinformation, addressing common concerns and information workshops all help to allay fears of the tools that can massively benefit the healthcare industry. Open acceptance will help to push implementation at a higher rate; the future of medicine is always unknown until it is in the present, and naturally the medicines of today were not always accepted when they were introduced.
The priority of AI technology in healthcare is to revolutionize the processes in order to make people's lives easier and make sure diagnoses are accurate and fast. AI has the ability to massively reduce workloads, but only if implemented and used correctly. Trust must be earned - it cannot simply be given - in order to make many of these changes, not just with patients but also boards and chairpeople. With time, and successful integrations from those early adopters, the rest of the world will one day benefit from the enhancements brought about by AI technology. However, the pitfalls are numerous; data privacy must remain airtight, and the price has to be right for any of these systems to be introduced. Although it is a hard path to follow, AI has the ability to change many aspects of healthcare for the better, and that remains the goal for the foreseeable future.
This article was a summaty of an episode of the Bytes of Innovation webinar series, hosted by Segmed. If you wish to learn more on the topic, please take the time to watch Kristin Yakimow MSc, MBA speak on the topic in Episode 22 of Bytes of Innovation: The AI Fellow.
.svg)
