We asked our clients to share their thoughts on the solutions and approaches that they evaluated against Jvion's Cognitive Clinical Success Machine. C-level executives from administration, quality, IT, finance, and clinical operations representing hospitals ranging in size from multi-facility integrated delivery networks to 150-bed organizations participated. And these systems use multiple platforms including Epic, Cerner, MEDITECH, and Allscripts. Three general flavors of comparison solution categories emerged during the discussion: enterprise data warehouses (EDWs)/data lakes, scoring methods, and electronic health/medical record (EH/MR) models. The following provides a summary of key findings.
A strong theme around patient-specific action crossed all three comparison solution categories. For EDWs, the biggest concern was the retrospective nature of the approach. Penny Burlingame Deal, CEO from Onslow Memorial Hospital, explained the challenge this way:
"(T)he greatest flaw in (the EDW) model is that (patient suffering) … has already occurred. Unfortunately, the patient has already suffered from whatever disease process or illness— what you're trying to prevent from happening—and that's not the best practice."
Scoring methods fell short on delivering the granularity and specificity needed to enable effective and targeted action. The relative effectiveness value of these solutions is low; they require that clinicians treat a large number of people to impact the small percentage that will suffer from an adverse event. For example, LACE—a common scoring method used to identify patients at risk of a readmission—requires that clinicians evaluate 1400 patients marked as low to moderate risk to find the 100 patients who will be readmitted. With such a high number of patients in scope, it is nearly impossible to effectively intervene and reduce risk for a large portion of the patient population.
In the case of EH/MRs, advanced statistical models are used to target at-risk individuals. The problem is that these models require clean, complete data—something that is nearly non-existent in healthcare. Even though these models may come as what appear to be easy "add-ons" to existing functionality, the resources and time required to make them work add up to a huge cost.
The second theme that came into focus was speed to value. Across all three solutions and approaches, speed to value was described as slow and limited. EDWs had the longest time to delivering returns. Much of the delay was due to heavy provider resource requirements and the time needed to aggregate and cleans large volumes of inconsistent, disparate, and incomplete data. Scoring methods remained stagnant in potential ROI. And while speed to value could be realized for high-risk segments—which represents a small volume in the overall patient population— the moderate and low risk segments were too large and generalized to drive meaningful returns.
EH/MRs posed a unique challenge. Many offer some kind of "predictive" model targeted to specific adverse events. But, as mentioned previously, for these models to work, clean, complete, and consistent data must be fed into the solution. This process requires resources—most often clinical staff—to make sense of notes, patient forms, and patient records that can train the model. This high-level of expert intervention adds up to delays in value realization and extended time to clinical and financial returns. Moreover, these point solutions can only be applied to one kind of illness or event. They cannot extend to other areas of risk because like scoring methods they are fixed in the attributes consumed and outputs that they can deliver.
As cognitive machines like Jvion's Cognitive Clinical Success Machine gain traction and market share within healthcare, understanding the distinctions between available approaches will become more important. Clarity into the value of each solution is oftentimes obfuscated by marketing spin and buzz. While we wait for third party research firms to bring some verity into the way we categorize the market, seeing how these solutions work within real, live hospitals is our best tool for evaluating and estimating the potential for this next wave of technology.