There are fascinating academic discussions in the medtech space about what can be achieved using artificial intelligence, but not enough about what’s practical for medtech clients — physicians and other medical professionals.
When we talk about artificial intelligence — real, deep learning AI — the truth is most physicians aren’t using it. This reality can obscure what’s coming: widespread adoption of deep-learning AI at hospitals and health clinics.
What’s going to change the tide? Our greatest need is for developers and tech executives to stop designing theoretical solutions, and instead place physicians, patients, and specific clinical needs at the center of their solutions.
Even in specific fields like radiology and oncology, there will never be a one-size-fits-all AI solution. Each practice will have its own needs, habits and hardware to work around.
‘Utilization is the new ‘innovation’
Over and above everything else, we must design technology that is used. We have to prove how to put AI in the workflow and grow utilization, more than building products.
By collecting more data, organizations acquire more capacity to improve their products, which attracts more users, generating more data and so on in a virtuous circle.
Whether it’s blood testing, cancer detection or even more routine medical diagnosis, deep-learning systems will only be as smart as the amount of data it is able to reference.
In many respects, the AI that exists today isn’t what the physicians are looking for. It’s where we start, but we won’t know the destination until we create these human-AI usage loops. Too often, all this uncertainty can cause hospital administrators to shun these technologies. But, to end up in the right place we all have to get started today.
Look at one hospital in Florida, which deployed AI to address a somewhat mundane need — flagging adherence to “order sets,” groups of clinical orders meant to standardize care.
Executives at the hospital started with a narrow focus on pneumonia and began compiling patient data from electronic health records, billing, and other analytics into a single system. This allowed them to see what pathways were working best, and ultimately increased adherence with order sets from about 30% to 80%, and reduced the readmission rate from 2.9% to 0.4%.
How to Get to Our Destination: Real AI
As experts in the field of medical imaging have repeatedly observed, organizations tend to succeed when they enact changes incrementally. Big-bang, revolutionary digital projects appear ambitious, but they often fail, even when they involve the biggest names in medicine and technology.
Remember when big data was initially going to revolutionize cancer care and artificial intelligence was the key? In 2012 there was talk of cognitive computing systems being deployed at academic cancer centers to help eradicate cancer.
Humbling experiences followed that we all have learned from. For one, any big data enterprise must connect effectively with electronic medical record systems. The details of making that work are essential and undertaking too broad an initiative can miss the little things that allow for success.
A better approach is to start with one variable that needs attention, such as diagnostic protocols, hospital readmissions or patient satisfaction, and analyze it using specific metrics and as much data as possible. In addition to EMRs, this mass of information will enable enterprise AI to do what it does best: identify patterns that would be hard for humans to recognize.
In diagnostic imaging, the measurements that clinicians are making provide rich new diagnostic information that can help us better understand entire populations of patients and apply this to a specific patient in need of a particular high cost surgical procedures. At a clinical level, recognizing these patterns means earlier diagnosis, more efficient care and better outcomes. For physicians still unsure about the technology, it also builds trust required to expand into the next areas of AI utilization.
Photo: metamorworks, Getty Images