A Beautiful Theory: Care Interventions

This article series will introduce and then subsequently discuss how to implement a framework in order to build interventions for complex populations that are more likely to reach their intended outcomes.

I recently gave a lecture at a leadership summit for health plan and medical group executives, and I recognized the value in sharing my experiences with a broader audience. Thus, here I will share my framework (in a two part series) in addition to some of the observations and research that informed the steps that comprise the framework. This framework was developed from a combination of experiences spanning both my academic research and my role as Chief Population Health Officer in a full-risk environment.

First, why does it matter? Risk-bearing organization are increasing in number as value-based care gains traction. For example, it is projected that 90% of Medicare is going to be tied to some value-based payment mechanism by the end of 2018, while Medi-Cal (in California) now comprises 80% of managed care and growing. As an effect, duals populations aligned with risk-bearing organizations are increasing at an unprecedented rate. It is clear we need more reliable, cost-effective and informed strategies for sustainably managing these populations in a value-based environment.

Why can’t we rely on what we already have available?

-Traditional or “gold-standard” models are often evaluated when they are first implemented (perhaps even measured over a few years), but generally are not continually re-evaluated over time due to costs. This means we don’t have a reliable measure of whether they are effective or as effective today.

Before adopting a model it is important to question when it was last evaluated.

 In the event interventions are continuously evaluated and evidence shows they are ineffective, it usually doesn’t matter much; defunding in the face of contradictory evidence is surprisingly and unfortunately rare. Many dynamic changes in the healthcare system render once effective models ineffective—a good case study is the Health Quality Partners case management intervention. When the Congressional Budget Office (CBO) produced a report showing no cost savings after ten years of program operations, they hypothesized this had to do with the increasing standard of care-- not the model itself (which was originally extremely effective in reducing inpatient utilization).

Thus, even if the intervention continues to be implemented with perfect fidelity, external context may render interventions duplicative.

Careful consideration is required now more than ever when selecting which models to emulate.

Lastly, often models are evaluated as effective, but only in one geography and in one population, yet they are considered fit to be replicable without taking into consideration that they may be point-specific (only effective in a particular context). Because a model was effective in one setting does not guarantee the strategies are able to be generalized beyond the original environment. How do we tell? There are questions (outlined in the framework) that we can ask to bring us closer to understanding what will really work as we survey the landscape of existing care models.

My research shows that only 2% of care models targeting Medicare or dual beneficiaries with co-morbid chronic disease are effective at statistically significantly reducing preventable inpatient utilization. To illustrate, if I go to a conference with 100 interventions/organizations claiming to reduce inpatient utilization in a Medicare or dual populations with co-morbidities, I know that approximately only 2 of these will really work. This changed the way I look at the entire system and it underscores the exigent need to be more discerning regarding which strategies to adopt. The following framework is intended to be a tool in your toolkit which helps you navigate this complex environment. 

This framework consists of four steps: 1.) Problem precision, 2.) evidence-based, 3.) adaptation, and 4.) actionable measurements. Please see the next article for a description of each step. (I didn't want to make this too long, but it needed the introduction).