Hotspotting Lessons—Part 1

Last month, Amy Finkelstein, Annetta Zhou, Sarah Taubman, and Joseph Doyle presented their analysis of a randomized clinical trial in the New England Journal of Medicine.  They examined the effect of intensive case management for medically and socially complex patients—‘hotspotting.’  

The authors worked with Dr. Jeffrey Brenner who developed intensive interventions for his patients in Camden, New Jersey. Atul Gawande described Brenner’s approach in a 2011 New Yorker article.   In 2013 Brenner received a MacArthur Fellowship, for “creating a comprehensive health care delivery model that addresses the medical and social service needs of high-risk patients in impoverished communities.” 

The structure of the NEJM study

The studied tested the Camden Coalition’s program: “In the months after [an index] hospital discharge, a team of nurses, social workers, and community health workers visits enrolled patients to coordinate outpatient care and link them with social services.” 

By randomization, patients assigned to the Camden program should differ only on treatment experience compared to the control group of patients who received usual care.  ‘On average’ there should be no other factors or patient characteristics that could confound the effect of the Camden program and prevent cause-and-effect inference. 

Of course, prudent investigators always confirm that a randomized assignment doesn’t inadvertently assign patients with a specific characteristic to one group but not the other. The authors show the similarity of their two groups on a range of characteristics (Table 1, p. 156).  So far, so good. 

The conclusion 

“In this randomized controlled trial involving patients with very high use of health care services, readmission rates were not lower among patients randomly assigned to the Coalition’s program than among those who received usual care.” 

The authors suggest previous studies that demonstrated an impact of hotspotting might be fooled by a strong ‘regression to the mean’ effect:   patients who are hospitalized in a six-month period tend to have a lower rate of hospitalization in the subsequent six months. 

They demonstrate this phenomenon for the patients in their study, Figure 2, p 160: 

The graph shows that both the treatment and control groups had relatively high rates of hospitalization in the two quarters prior to the index hospitalization and roughly constant rates earlier than six months before the index hospitalization. 

The support for ‘regression to the mean’ interpretation comes from the two sets of bars at Quarter 1 and Quarter 2 after the index hospitalization:  the hospitalization rates are lower after the index hospitalization for both groups, with no big differences in rates. 

Porpoising 

My colleague Chris Crowley at West Health introduced me to the term ‘porpoising’ to describe the regression to the mean pattern shown in the hospitalization graph.  In cost management projects, focus on patients with an index hospitalization is like observing a porpoise or dolphin at the top of a breaching jump—the highest costs occur near the index hospitalization.  Subsequent admissions and cost of care per patient will tend to drop in subsequent months for many patients. 

I’ll continue discussion of the NEJM study in Part 2 of this post.