Predictive analytics as a solution category has become so empty that it does little to help providers understand or estimate the value of a predictive analytic tool. There are two broad trends that have led us here: the flood of vendors claiming to have a predictive solution and the loose standards used to evaluate predictive analytics.
In a recent industry poll, more than 300 vendors claimed to have some kind of clinical predictive analytic solution. They ranged from more advanced neural net models to simple statistical algorithms executed using spreadsheets. While the solutions varied in how they produced "predictions," they shared some commonalities:
- They were all limited in the number of data elements that they could accommodate
- They were static and only effective/relevant for a small, specific patient population
- They delivered broad categories of risk stratification
- They didn't enable effective action or deliver recommendations
Predictive analytics has become a commodity that anyone can package and sell using marketing buzz and sales speak.
More sinister is the way that predictive analytic vendors have used the lack of clear standards to evaluate and compare solutions. Accuracy, positive predictive value, specificity, sensitivity. These terms are thrown around with an air of data science expertise to lure potential buyers and obfuscate the real effectiveness of a predictive solution.
There are some predictive analytic solutions that deliver value. Many provide useful methods to stratify risk across a population. But the accuracy of these solutions relies on clean and complete data. Moreover, these solutions lack the flexibility to accommodate new and different populations, account for external factors impacting risk, and deliver recommendations on the most effective actions.
Cognitive science driven solutions, in comparison, are built specifically to address gaps in data and quickly adapt to changes in patient populations and needs. These machines provide prioritizations, predictions, recommendations, and interventions that can stop risk and improve health. They do this using sophisticated data science techniques like Eigen spheres to make sense of “bad” data and deliver a highly precise, ultra-definition view into a patient's future health. This view enables the actions that will have the greatest positive impact for the patient. And the machine can do this for any population across any illness or condition because it is designed to continuously learn and advance effective interventions.
Where predictive capabilities fall short, cognitive science starts. These solutions drive the kind of insights that account for the things that we can't see - the exogenous factors impacting a patient's life. And they engender action that leads to more precise and effective interventions. Cognitive science does much more than predict risk; it helps providers by making sense of the data that we can't see and enables a view into the future state of the patient that is holistic, actionable, and precise.