Predictive AI used as diagnostic tool, generative AI still finding a role
Much of the recent news about the current capabilities of artificial intelligence, or AI, spurred by the introduction of platforms such as ChatGPT, Scribe and Jasper, have focused on how they will change communications and affect present-day work. When assessing the recent AI advancements’ impact on the healthcare industry, though, it’s important to understand how ChatGPT and other generative AI tools differ from the more commonly used predictive AI tools.
AI is broadly defined as any technique which enables computers to mimic human behavior. Since its early development, AI has branched into what is now called machine learning and natural language processing. Through these branches, computer systems are able to perform pattern recognition and anomaly detection and are also able to perform human like tasks such as visual perception, speech recognition, and analysis based on data patterns.
While the creative possibilities of AI are grabbing attention now, the healthcare industry has been successfully using AI for more than a decade, and the revolutions in the workspace anticipated in other industries are already happening in healthcare. AI often works by detecting data anomalies in numbers and images sooner than is possible for humans.
Grant Thornton Healthcare Advisory Managing Director Mandeep Maini said the first point in understanding AI use in healthcare is to distinguish “generative AI” from “predictive AI.”
“Generative AI is, essentially, a word prompter,” Maini said, “so it's talking in human-like language. But as you go deeper into it, you learn that it cannot quantify or validate answers, or why it came to a certain conclusion.” The buzz created by generative AI has put the technology front and center, highlighted opportunities as well as risks, and even prompted Congress to draft legislation to regulate the use of AI. While generative AI is rapidly becoming more powerful, it’s use in medicine is still limited, Maini. Predictive AI on the other hand uses statistical algorithms to analyze data and make predictions about future events. It is a valued tool in the medical world because of its consistency and ability to validate why it arrived at a particular diagnosis.
Futurist Jim Carroll on AI in healthcare
4:47 | Transcript
Predictive AI use in medicine
“Generative AI is, essentially, a word prompter. As you go deeper into it, you learn that it cannot quantify or validate answers, or why it came to a certain conclusion”
The benefits that AI offers to the healthcare industry include predictive capabilities, assistance in administrative tasks, expertise in pattern recognition and learning and explanation of outcomes. Each of these areas has been leveraged to create AI applications for use by patients, by clinicians, by administrative staff, and by the pharmaceutical industry.
For non-medical people, the most well-known and accessible use of predictive AI are over-the-counter wearables such as Freestyle Libre 2, FitBit and Apple Watch and various blood pressure monitors. These devices collect data on bodily functions such as heart rate, temperature, blood sugar and mobility, then compile the data to deliver health diagnoses directly to the wearer.
But medical personnel have long used a wide variety of predictive AI applications as an essential diagnostic aid. There are many common uses, including:
Blood samples taken at birth which are used to identify genes associated with possible genetic disorders
Early detection of major afflictions such a cancers, respiratory and circulatory system diseases using algorithmic analysis of imaging from CT scans, MRIs and X-rays
Research with large population datasets that can identify health trends in geographies, demographics and income levels
Earlier identification of neurodegenerative diseases, where early diagnoses are difficult because symptoms of these problems are common to many different afflictions
The premise of using AI to augment diagnoses is to be able to diagnose issues earlier and better. If you can diagnose a disease sooner and with more accuracy, patient outcomes will likely be better because it helps augment and accelerate treatments, such as having patients take medication earlier. And though it’s not really possible now, in the near future there may be ways to directly attribute improvements in patient outcomes within a system to particular AI uses.
AI is well positioned to improve patient care and potentially save lives, save healthcare costs by early detection, targeted treatments and automation, but it cannot replace human interaction, Maini said. According to the American Medical Association, a combination of man plus machine results in enhanced human capability.
Healthcare providers are still the last word on medical diagnoses no matter how sophisticated the AI applications are — at present, AI is only a tool. Physicians must be “on record” to look at the data and interpret it, not merely parrot what an AI tool says. This physician authority is particularly important in situations that happen frequently when patients visit doctors citing their own “diagnoses” based on evaluating their symptoms and performing an inexpert search on a browser such as Google.
Industry series stories
Organizing to succeed
“We're not replacing nurses, we're not replacing doctors if they are simply either talking into an application to transcribe notes to be written or taken from a patient consultation.”
Grant Thornton Healthcare National Managing Principal David Tyler that similar to other healthcare concerns, bringing AI tools into the business must be the result of a sound decision-making process.
“If you are at a business and or clinical leader in healthcare and you are not evaluating the efficacy and ethical use of AI in whatever processes you perform — whether it be patient communication, business office, billing, collection, outreach to new patients, evaluating physicians on quality metrics or staffing on the nursing floor — that would be an oversight and you need to begin to do that.”
Maini said using AI tools should not require extensive retraining to use, however, there may be changes in workflows due to the introduction of AI tools, and user training will be required for these new workflows.
The advent and use of AI in healthcare is inevitable. Here is what health systems should do to prepare their organizations for incorporating AI and governing its use.
- Create an AI strategy: Assemble a team of resources from all functional areas, including technology and clinical personnel, to identify which AI tools to implement and which processes will be automated; identify the workflows that will need to be adjusted due to the automation, and what training plans will need to be put in place; create a roadmap for AI Strategy and implementation.
- Create or refine a data strategy: Create an analytics team to define what data elements the organization wants to collect by process area; assign owners and stewards of this data; create a data model and relationships; and create controls and standards for the management and control of the data. The collection of data used to train any AI / ML model must be done carefully to ensure that the data is complete and comprehensive and does not contain any inherent bias in the models.
- Invest in a cloud strategy: Given the volumes of data AI models are able to collect and process, investing in infrastructure and cloud strategy which is subscription based is advised. This way, an organization will not continue to add application and database servers, IT staff and data storage. This strategy will also increase the interoperability as well as security of data.
- Refine the privacy and security strategy: Since digitized data is easier to share across an organization(s), adequate privacy and security policies should be created to keep data secure, prevent unauthorized access to protected health information, and reduce the risk of data breaches.
- Change management: Make sure that staff members understand that AI applications are being implemented to improve quality of care and productivity and are not intended to replace clinical and non-clinical judgement and staff. AI technology requires human oversight, review and decision-making.
- ROI: clearly articulate the time period over which any ROI is expected since the technology is still evolving.
More stories on artificial intelligence
Generative AI possibilities
Though predictive AI has been the primary focus of AI, the growth of generative AI recently indicates it should have some place in medical providers’ plans. One main focus of generative AI is to more efficiently produce internal company communications, a need that certainly exists in healthcare. Maini said healthcare providers could explore generative AI use for writing press releases, handling patient data across platform, and even to transcribe doctors’ notes.
“We're not replacing nurses, we're not replacing doctors if they are simply either talking into an application to transcribe notes to be written or taken from a patient consultation,” Tyler said.
In that vein, Microsoft and the company Epic announced the formation of a partnership to use generative AI in Epic’s electronic health records (EHR) platform. Like other generative AI uses, this one is not directed at patient care but at improving the accuracy and efficiency of health records – allowing staff to more quickly access and update patient information. Nonetheless, these generative AI uses can free up staff time for patient care by eliminating human copying errors and making the process quicker.
Whatever the uses, no one disputes that the use of AI in healthcare, both in preventative and generative applications, is still rapidly increasing. Making decisions on how to obtain and use these powerful tools is best done by a well-informed healthcare board that hears and acts on the advice of its physicians, clinicians and nurses.
“AI is a new and exciting and has high potential for good,” Tyler said. “And if you're not even evaluating the tool, I think that's an oversight.”
Get the outlook for other key industries
More healthcare insights
No Results Found. Please search again using different keywords and/or filters.