Evidence Supports Scalability of Effective Models: Enormous Possibility

In the June 2012 article in Health Affairs by Brown et al., “Six Features of Medicare Coordinated Care Demonstration Programs That Cut Hospital Admissions of High Risk Patients” a subgroup of patients was defined using criteria available in Medicare claims data;

[(HF, CAD, or COPD) AND ≥1 hospitalization in prior year]
OR [(diabetes, cancer (not skin), stroke, depression, dementia, atrial fibrillation, osteoporosis, rheumatoid arthritis/osteoarthritis, or chronic kidney disease)
AND ≥2 hospitalizations in the prior 2 years]

Members of this subgroup participating in the Health Quality Partners (HQP, http://www.hqp.org) program had -33% fewer hospitalizations (p=0.02), -30% lower Part A & B Medicare expenditures (with program fees excluded) (p=0.045) and -21.5% lower net costs (program fees included) (p=0.15).  All terrific stuff and since the emphasis of this particular analysis was to identify common elements of successful programs, using complex subgroup definitions for that purpose is fine.  However, there are significant real-world challenges in trying to use such a complex eligibility criteria for program implementation and scalability.

In the HQP experience, it remains hugely challenging to cobble together a patchwork of collaborative data sharing agreements with hospitals and primary care practices in order to serve a geographic region.  Complex criteria sets such as these make that job harder.  Having worked many years with the authors of this article I know that they too are fully aware of and appreciate this concern, but the inexperienced reader might confuse or meld these two separate issues: finding common elements of successful programs vs. defining the “best” target population for scaling effective care management interventions.In tables in the Appendix to the article another, simpler subgroup is defined as;


Just having one or more of these 3 conditions meets this subgroup criteria; no other prior hospitalization usage, other co-morbidities, etc.  This group is a lot easier to “find” prospectively with data readily available in primary care practices (their billing data).  In the demonstration, HQP randomized 695 individuals meeting these criteria (43% of all those in the study) vs. just 273 (17% of those enrolled) of the more complex subgroup above.  Results for this simpler, more easily identified subgroup?  For HQP’s program, not bad; -25% fewer hospitalizations (p=0.005), -20% lower Parts A & B Medicare expenditures (-$220 per person per month) (p=0.02), and -10% net savings when program fees were included (-$116 per person per month) (p=0.22).

There are plenty of challenges to scaling highly-effective care management programs like HQP’s.  One challenge we can and should avoid is making the criteria set for eligibility needlessly restrictive and difficult to implement – especially when the evidence supports a wider population of people who can benefit.  With each larger scale cycle of testing, the criteria can be further refined (and coned down, if necessary), but in the meantime, we should encourage the use of target group criteria that are feasible to implement and support system redesigns with the greatest possible chance of successfully transforming our health care system for the better.

This same blog article is also posted on the HQP blog at http://www.hqp.org/blog/

Leave a Reply