The Potential Impact of AI on EHR Efficiency and Patient Care

In an interview with Integrated Healthcare Executive, Josh Budman, SVP of Research and Incubation at Net Health, provides valuable insights into integrating AI and machine learning into EHRs and its potential impact on efficiencies and clinical outcomes while honoring regulatory guidelines. 

My name is Josh Budman, and I am the SVP of Research and Incubation at Net Health. In 2014, after I had completed my undergrad and master's degrees in Biomedical Engineering at Johns Hopkins, a fellow engineering graduate and I started a company called Tissue Analytics. That colleague, Kevin Keenahan, is now the Chief Product Officer of Net Health. Tissue Analytics was the result of research we were conducting, and our goal was to make the act of quantifying the progress of skin conditions a lot more accurate, efficient, and objective. We did this by using rulers and pictures to measure the improvement or deterioration of skin conditions over time.

In the early days of our research, we involved advanced AI and machine learning to accurately quantify the progress of skin conditions. We built a full mobile app and web application in a way that our work could integrate with electronic medical records. Our company was acquired by Net Health in 2020.

Today, I conduct research around data and incorporating AI, machine learning, and business intelligence into our products. Any embodiment of advanced machine learning or data analytics used for our products goes through my team.

Could you share some examples of how AI has been integrated into EHR systems thus far?

I want to start by explaining that we use AI as an umbrella term. Technically, AI is any type of action that a non-human carries out. Or, in other words, any type of human-replicated action that a machine carries out. AI has existed in many industries, even health care, for quite a while. One example of basic AI that we've seen in health care is dictation software that has been incorporated into health care for almost 2 decades now. Dictation software is a good example of AI that isn't necessarily true machine learning, which is a subset of AI. Another example of AI in health care is the software that has been incorporated into EHRs to report on negative drug interactions. We refer to it as a rules engine, which is maybe more simple in the spectrum of simple to advanced AI.

The field of AI has continued to advance, and I do think that the large EHRs are starting to recognize the value of AI and machine learning applications because they've started allowing third-party vendors like Net Health to incorporate specific AI modules into their software. Net Health has also embedded some of our own AI and machine learning applications in our software such as the ability to estimate whether or not a wound is going to heal or whether or not a patient is at risk of missing a visit. We believe that embedding insights through advanced machine learning and artificial intelligence is really what defines the next generation of EHRs.

How do you think EHR will continue to change as AI becomes a more prominent part of health care delivery going forward?  

We've learned that the first area in which health care stands to benefit from AI is with efficiencies and operations, more so even than with improving clinical outcomes and diagnosis. A clear area of focus within efficiencies and operations is documentation. Clinicians probably spend more time in a single software platform, their EHR system, than individuals in other professions. Studies have indicated clinicians are dedicating around 2 to 3 hours a day to documenting and writing notes within their software.

Clinical outcomes benefit from documentation, but a lot of EHR use is driven by regulatory compliance. So, there is a massive amount of time clinicians need to spend in front of their computers documenting info in, to put it bluntly, glorified form fields. Hopefully, AI and machine learning can improve documentation efficiency so that these providers can spend more time treating patients. Many technologies are trying to solve this problem, and it's described globally as a digital ambient experience. Ideally, AI could take "voice to text" to a new level where clinical notes on a conversation between a clinician could be automatically filled out with high integrity and accuracy.

I do think we're also going to start to see major improvements in the way diagnoses are carried out using advanced machine learning and advanced data sets as effectively aided diagnostic tools. This would mean that providers can work in tandem with AI tools to start making more accurate diagnoses. A reason may not see this technology advance as quickly is simply because of the associated higher level of risk and regulation. When regulatory bodies get involved, the timeline from developing an algorithm to deployment can be longer. 

There is often initial distrust of new technological advancements. Could you address some of the common safety, cost, and data privacy concerns that can arise from integrating AI into health care technologies? 

Any time we talk about safety and health care, the Food and Drug Administration is involved. If AI and machine learning tools were to be involved in making a medical decision or documenting info in a medical record with little clinician oversight, of course, that would introduce a high degree of risk. Let's use a cancer diagnosis as an example. If we trusted an AI tool to make a cancer diagnosis without regulation, the risks of that diagnosis being incorrect are massive. The solution to troubleshoot this is strict adherence to FDA regulations. The FDA is a robust regulatory body, and its experts put together regulations and review submissions. I think if all parties follow those regulatory guidelines, many safety concerns are attenuated.

Another area of concern is data privacy. The data privacy concerns in these AI EHR systems aren't much different from the privacy concerns that already exist and are protected against in health care. HIPAA, the Health Insurance Portability and Accountability Act, and GDPR, Global Data Protection Regulations, are already applied very heavily to EHR technologies. I don't feel that the introduction of AI and machine learning changes the way that developers of EHRs or developers of health care software need to accommodate those regulations.

An exception that is very specific to AI is that we use data sets to train our algorithms. When data is fed into an algorithm, it makes the algorithm smarter. It's up to the developers of these tools to design data pipelines that honor their client contracts and the regulations set by the local and national governing bodies. These data sets also don't incorporate protected health information. When we train a dataset with patient data, the patient's name or exact date of birth is irrelevant. 

Do you have any advice for providers interested in integrating these new technologies into their health systems? 

My advice is to ensure that whatever technology providers or health systems choose to implement is very closely in line with the salient regulatory bodies such as the FDA.

Another piece of advice that is often overlooked is to evaluate how well the technology and existing business systems integrate. For instance, let's say I am a chief medical officer who uses Epic EHR systems, and I want to implement a new, advanced AI machine learning software in one of my health care practices. If that AI software doesn't integrate nicely with my Epic system, the risk of providers at my facility not using that software is extremely high. As a personal anecdote, when we first released our Tissue Analytics software, we wanted advanced AI and machine learning technology applied to the imaging of wounds. We assumed it would be used because it was "cool" even if it did not integrate well with existing EHRs. We couldn't have been more wrong.

Following that point, the main advice I always give clients implementing new technology is to do your very best to ensure it integrates nicely with clinical workflows. In health care, workflow truly is king.

Is there anything else you’d like to add? 

If you would like to learn more about what we're doing in the realm of AI machine learning visit nethealth.com. Our goal is to become "EHR 3.0". EHR 1.0 was paper documentation, our current phase of EHR 2.0 is the computerization of that paper documentation, and for EHR 3.0, we hope to use more advanced tools to create a better experience for providers and patients.