Too many imaging AI vendors make big promises that their technology can’t keep when it’s deployed in the real world. Here are the big questions to ask to see if your vendor will deliver meaningful outcomes, not just aspirational promises.
Many imaging AI vendors claim to provide solutions that offer simple, seamless integration with other systems, and lightning-fast implementation times of just a few days—or even a few hours.
For example, if you’re looking for solutions to track incidental findings and manage patient follow-up, you’ll find many bold claims that vendors can capture these findings with 99% accuracy, and prevent leakage with seamless patient tracking.
It’s easy to make claims like these, but to actually achieve those kinds of results, you need a dedicated technology partner with experience and expertise in breaking down data silos to drive information awareness, access, and availability.
So how do you figure out if the vendor you’re considering can really deliver the results you need? Here are five searching questions to ask, and the responses that should raise alarm bells for any buyer.
1: How is your EHR integration superior to your competitors’?
Some companies say they have a “special” relationship with an EHR vendor that enables “deeper” integration and faster implementation times. Be wary of these claims; sending and retrieving patient data follows a standardized process where the data must map to a specific format for the EHR. There’s no secret handshake that allows special access.
When vendors claim they have superior EHR integration, be sure you understand the full workflow and how information is being exchanged from one system to another.
2: How do you achieve a fast implementation?
Beware of vendors’ claims that they can complete an implementation in days—that speed usually means you’ll get a basic installation with no real-time, bi-directional data exchange.
A patient’s healthcare experience and medical history extend far beyond the confines of any single provider or specialty. Without taking the time to complete robust systems integration that enables instant data exchange, you’ll never have the complete picture of the patient needed to deliver high-quality, timely care.
Make sure you understand how your vendor handles real-time patient updates and events like address changes, or mortality status. You wouldn’t want to send a follow-up reminder to the spouse of a deceased patient. Nor would you want your patient coming in for their scheduled low-dose lung cancer screening exam just a week after having chest CT while in the ED.
3: How do you ensure appropriate alerts get to the right clinicians and patients?
When vendors claim their entire alert and notification system is automated, it sounds great—but automatic alerts don’t always lead to better care. Clinicians are already experiencing more burnout, in part due to alert fatigue from a constant stream of inaccurate notifications.
What’s worse is that without a complete view of the patient and real-time information on their care, there’s a risk that automated systems will send inappropriate and insensitive notifications to patients (alive and deceased).
So, when vendors focus on automation, get more details on how it will impact clinicians and patients. You wouldn’t want to send a follow-up reminder to a sick patient or their primary care physician if the patient is already undergoing related treatment.
4: How do your precision and accuracy claims account for AI model drift?
Some imaging AI vendors claim they have unparalleled precision and accuracy for identifying and tracking nodules. But as your datasets get larger and more complex, it becomes increasingly more difficult to achieve precision and accuracy without producing lots of false positives.
Algorithms always have an implicit bias based on the datasets they were trained on, and each organization will want to look for different things, based on their people, processes, and patients. So if a vendor also says their model has the flexibility to meet every organization’s need, their precision numbers may not be as high as they claim.
Ask your vendor if its AI model can still achieve high degrees of precision and accuracy in the variable and unpredictable world of healthcare, rather than just in a controlled experiment. Be sure you understand how their model will perform on your unique data and what flexibility you have to modify search criteria.
5: What’s the total cost of ownership?
A cheaper price point may look attractive, but it can be a short-term saving with long-term consequences. It always pays to consider the total cost of ownership and what customization, integration, and flexibility your organization will need to get the most out of the technology.
And if you’re looking at a point solution for tracking and follow-up in a specific specialty, beware of deploying a solution that’s siloed from the rest of the organization, limiting its ability to deliver long-term value.
Take the time to do some more detailed forecasting about the real costs involved. Consider what different pricing tiers actually offer—and how much you will need to spend on add-ons to get the outcomes you’re seeking. There may be additional long-term software and support costs as your program grows.
Choose a vendor that delivers outcomes-focused AI
Healthcare organizations face increased staffing shortages and budget cuts, and successful outcomes require a collaborative approach to planning, implementing, and adopting AI solutions. But that collaborative approach is just one part of the solution.
It’s even more important to work with a partner that offers a single, data-driven platform that enables you to drive an enterprise AI strategy at scale. At Nuance, that’s what we do. We help healthcare organizations use powerful AI solutions to connect the point-of-read to the point-of-care—and deliver better outcomes for more patients.
