Predictive analytic companies have been latching on to the idea of “impactability” and it is masking the intent of the term and overstating the real capabilities of predictive analytic solutions. The idea behind “impactability” is this: there are patients whose outcomes can be changed with the right intervention and there are patients whose outcomes can’t be changed no matter what you do. This isn’t about caregiving. Clinicians should provide the best care to all patients regardless of the outcome. This is about focusing the right actions to the exact patients who can benefit.
For example, if a patient is at high-risk for Congestive Heart Failure (CHF) and is on the transplant list, he or she will be back in and out of the hospital pretty frequently. In fact, the best care for this patient may actually involve frequent admissions. Targeting activities to reduce a readmission for this patient is misguided at best and might actually create harm, as suggested in recent research. A provider would still provide the best care for this patient. But deploying readmission staff and interventions as part of the care plan is not going to change the all but certain likelihood that this patient will—and maybe even should—return to the hospital.
Why are predictive analytic companies trying to own the idea of “impactability”? Because at their core, these solutions are not designed to increase patient impact. What they are doing amounts to a statistical shell game.
Predictive analytic solutions are modeled, tuned, and measured by their ability to identify high-risk patients. These are the patients who, like our CHF example, are highly likely to experience some adverse event (e.g. readmission, pressure injury, sepsis). The challenge is that the majority of these patients will have the adverse event regardless of clinical action. Our CHF patient is going to be readmitted to the hospital; we cannot impact our CHF patient’s readmission.
Predictive analytic solutions cannot identify those patients who are on a trajectory toward becoming high-risk. These are the patients who, under the static projections delivered by predictive analytic risk stratifications, show up as medium or even low risk. These are also the patients where we have the best opportunity to change the trajectory of the individual toward an adverse event. These patients are “impactable.”
But just knowing you can impact someone isn’t enough. You need to know the best intervention that will knock that patient off of his or her trajectory. The other great failing of predictive analytics is its inability to identify the exact, individualized interventions that will not only change the outcome but will also ensure patient engagement. The effectiveness of interventions is highly reliant on an understanding of the clinical, socioeconomic, and exogenous drivers behind the risk and how those factors influence intervention effectiveness. Predictive analytics, with its limited understanding of risk, adverse events, and outcomes, is not equipped to extrapolate the actions that will ensure the best possible quality outcome for a patient.
The only demonstrated way to identify “impactable” patients and the patient-level actions that will lower the risk of an adverse event is through the use of Eigenspace.
Eigenspace is a highly complex, intertwined mapping technique that has proven itself a critical tool in making sense of extremely difficult concepts including quantum physics, facial recognition, consumer behavior, and (now) patient health. By establishing an Eigenspace that comprises millions of patients and includes thousands of clinical and non-clinical attributes for each patient, we can transpose any new patient or patient population and answer dozens if not hundreds of questions related to risk trajectory and effective clinical action.
This approach is completely different from predictive analytics. Within the Eigenspace, we can see those patients who are currently at high-risk of an event and the patients who are on a trajectory toward that event. We can extrapolate all of the factors behind the risk. And, perhaps most importantly for the patient and this conversation, we can determine if we can impact the outcome and—if so—HOW.
Instead of building models for each new adverse event, the Eigenspace patient topography can be used to determine risk, impactability, and clinical action across dozens if not hundreds of clinical applications from avoidable ER admissions, to sepsis, to readmissions, and no-shows. And because the topography is already built, realizing value from this approach takes weeks (not months or years of building, training, testing, and deploying predictive models).
When a predictive analytic company starts talking about patient impact, stop them. Ask them the right questions: How do you identify those patients who not high-risk yet but are on a trajectory toward becoming high-risk? Are you able to determine the clinical and non-clinical factors driving that trajectory? Can you determine who is impactable and who isn’t? Can you determine the BEST personalized intervention that will not only change a patient’s trajectory but also ensure engagement? If they tell you they can do these things, they aren’t being honest. And the best thing you can do for your patients is walk away.