Author: Mike Brandofino, President and COO, Caregility
As in almost every aspect of life, Artificial Intelligence (AI) has entered the healthcare space, driving healthcare leaders to simultaneously be optimistic and concerned. The potential for AI to modernize care delivery and deliver on the promise of improving patient outcomes, increasing caregiver efficiency, and allowing practitioners to work at the top of their certifications is garnering understandable attention. However, like an iceberg, there is much more beneath the surface to consider before diving into AI adoption.
The Hidden Layers of Health AI
As we approach the AI iceberg, it’s imperative to consider the unseen, equally important, and often challenging aspects of effective implementation. The healthcare AI market is nascent and in flux, as evidenced by the spate of recent acquisitions, company failures, and startups entering the market. With regulatory frameworks still pending, healthcare AI remains a moving target. A measured approach to adoption is crucial, particularly in the high-stakes world of patient care, where poor execution will cost money and can put programs and patient lives at risk.
On the surface is the shiny object called Artificial Intelligence, but right out of the gate, a great deal of technology that is touted as AI is nothing more than automated data gathering with some logic around what to do when certain data is captured. While this can still prove to be valuable and in some cases, even more impactful than true AI, it should be considered as a different potential tool.
True AI will certainly capture data, but the difference is what is being done with that data and what inferences are deduced from comparing that data against a proven historical model to provide predictive analysis. That predictive analysis can be in the form of a diagnosis, a recommended care plan, or an alert indicating patient decline. The permutations are endless but what all AI engines have in common is the need to compare current data against known models. AI designed to learn will add new data to the model to continue to learn and enhance the accuracy.
Responsible health AI implementation warrants a look below the surface to expose the hidden challenges and considerations when evaluating technology for use in healthcare.
Data Model Considerations
As mentioned above, the success of AI in healthcare depends largely on the quality of the training data used to develop models. Data quality, representation across diverse patient populations, and model accuracy are vital to ensuring that AI systems can be trusted to make sound clinical decisions without bias. Health systems should require transparency from AI vendors and rigorous testing to ensure that models are accurate and support reliable outcomes. It is important to ask critical questions including: How was the model built? Where did the data come from? Will you be using my data? Is the data anonymized? Where is the data stored? Can you opt out of having your data used? The bottom line is, the data model is what drives the accuracy of an AI engine, and the answers to these questions will provide valuable insight on the viability of the engine.
For AI to be effective, it must be embraced by staff. There is often apprehension that AI will replace human workers. It’s important to implement AI in a way that is clinically impactful, enhances workflows, and minimizes disruption. A key question to ask clinical leadership when evaluating AI technology is, “Does it add value?” This value can come in the form of decision support, productivity gains, or actionable information. The question to ask the business is whether this is worth paying for, and if so, how much.
As an example, there is AI technology that can count the number of patient coughs per hour and provide an alert on the frequency increasing and the probable cause of the cough. Is this valuable? Who will get this information? Will it create false alerts from visitors coughing? This is one example, and one you may find value in, but ask the challenging questions and think through the impact on the clinical team, the volume of potential data and alerts, and how this will fit into the operational flow. And also, is it valuable enough to pay for the service?
AI solutions can generate large volumes of protected health information (PHI), adding to patient data vulnerability. Safeguarding the PHI generated and processed by health AI solutions is paramount. Solutions that support local edge processing can enhance security by keeping PHI within the confines of the healthcare facility, minimizing the transmission of sensitive information over the internet.
Additionally, the data captured by AI should be used responsibly. The data required for models to support machine learning AI means the solution provider will want to leverage data from your patients. You should have the option to opt out of this data participation and if you do opt in, you must be confident in how this data is being handled and protected.
This plays into another critical component of the use of AI in healthcare: patient acceptance.
Implementing AI raises questions about patient rights, awareness, and consent. Health systems should consider provisions for offering patients insight into the AI solutions being used by the care facility and offer clear patient opt-out options when applicable.
I have had numerous conversations with healthcare executives and innovation teams across the country, and there is an often glaring disconnect between them and the clinical team on the floor about what is feasible. There is an overwhelming desire to expand nurse-to-patient ratios and AI is seen as the panacea for the staffing shortages and high cost of care. However, what many fail to realize is that there are a number of potential logistical challenges that can derail the intended benefits of AI. The potential for a massive amount of information and alerts that need to go somewhere can lead to caregivers becoming desensitized and potentially missing critical warnings. Evaluation of any AI solutions must include understanding the impact on the clinical team and the logistics of fitting it into the workflows as an augmentation.
AI implementation in healthcare is not a one-size-fits-all proposition. Scalable, multi-solution setups require flexible foundational infrastructure that can support a variety of technologies, both native and third-party, on-premises and cloud-based. As the market evolves, health systems will want to avoid getting locked into siloed solutions that may become obsolete as technology advances. Agile infrastructure that allows for adaptability and growth is key, enabling organizations to integrate new AI tools as they emerge and extend use cases where it makes sense.
Due to the processing power required and the massive amount of data required by AI engines, many providers utilize cloud-based solutions. Understanding the impact on your network and the amount of data being transmitted can be an important aspect of determining if the solution can scale.
AI is undeniably the shiny new object in healthcare, but it’s really not about technology. It’s about our ability to create solutions that solve problems for caregivers. We’re standing at a technological pivot point in healthcare, and leaders must approach AI implementation with open eyes, looking beyond the hype to understand the full scope of challenges and opportunities that come with implementation.
Field testing and clinical feedback are essential to ensure that AI tools meet the real-world needs of healthcare professionals. This is not a race to adopt the latest technology. It is an intentional move toward more modern, future-proof care delivery models that better serve patients and healthcare organizations. By taking a measured, thoughtful approach to AI implementation, health systems can navigate the hidden challenges of the AI iceberg and chart a course toward intelligent, truly transformative care.
This article was first published in Becker’s Hospital Review.