Artificial Intelligence (AI) is healthcare’s latest hot topic. Get the top five questions you should ask when evaluating if an AI solution is right for your organization.
Artificial Intelligence (AI) is healthcare’s latest hot topic. The noise around AI’s potential industry impact has risen from a constant buzz to a loud din. As a consequence, it is hard for providers to discern AI marketing hype from real, actionable information. To help guide AI solution evaluation, we compiled the top five questions that every provider should ask potential vendors. These questions are designed to separate those solutions that are making a real, positive impact on patient lives from those that are simply claiming the AI hallmark to drive sales.
Artificial Intelligence (AI) is healthcare’s latest hot topic. Get the top five questions you should ask when evaluating if an AI solution is right for your organization.
1. How does your solution address a patient’s total cost of care?
AI offers a path to driving clinical meaning out of data. Electronic Health Records, which have become commonplace in large part because of government mandates, aren’t seen as powerful clinical tools; their primary value, according to Primary Care Physicians, is data storage (44%). AI solutions offer a path to extracting clinical value from the large amounts of patient data collected. It is estimated that AI has the potential to improve health outcomes by 30% to 40% and will help reduce healthcare costs by as much as 50%.
An effective AI solution should have a demonstrated breadth of clinical application that supports primary, secondary, and tertiary prevention. You should get more than just risk scores. You should receive patient-level risk propensities, the clinical and non-clinical factors driving that risk, and the patient-specific recommendations that will most effectively reduce the likelihood of an adverse event. The solution should deliver these outputs for more than one clinical use case and have the capability to quickly expand to new areas of application with few technology or data science resource demands. If a solution isn’t addressing multiple aspects of patient care—from primary prevention within the community setting to avoidable inpatient events and discharge—then the tool isn’t a true AI asset.
2. How does your solution adapt to new patient populations?
One of the big challenges providers face when implementing any kind of predictive tool is the inability to extend the models to new patient populations. Once tuned to a specific group, it can be hard to replicate performance in new community settings.
A true AI asset is capable of quickly adjusting to highly diverse patient populations with little to no impact on solution performance. These adjustments should require very little in the way of testing and validation. And you should be able to see the impacts of localization on the recommendations and outputs delivered by the solution.
3. What is your approach to measuring and communicating return on investment (ROI)?
ROI can be tricky when we talk about prevention. It can be hard to tease apart those patients who were helped by an intervention from those who avoided an adverse event all together. In the same way that we are able to evaluate a vaccine’s effectiveness by looking at rates of infection and disease, we can measure ROI impact by looking at rates of the target clinical area.
An effective AI solution must have a robust way of capturing and communicating this ROI. And this communication should account for expected variations in the number of adverse events as well as provider usage. Without visibility into the value that a solution is driving, there is no way to capture the true impact to patient lives.
4. How do you ensure the performance of your solution across areas of clinical application, populations, and technology environments?
A true AI asset is designed to maintain and improve performance across multiple areas of application. This includes different populations and data sets. This ability—to understand a problem and then extend that understanding to other areas—is one of the attributes that truly sets AI assets apart from simple predictive analytic or machine learning solutions. If you have to train, test, and validate each new use case, you aren’t looking at an AI asset that will work for healthcare.
5. What are the workflow adjustments required to drive value from your solution?
Lastly, it is important to remember just what an AI solution is supposed to do. Within healthcare, AI is supposed to reduce the cognitive demands placed on caregivers. This means that it should not require additional time, drastic workflow changes, or effort to understand its outputs. The risk data and interventions provided by the solution should be delivered in a way that is easy to understand and action. If they aren’t, you risk a lack of adoption and the failure of your AI program.
