yellowbrickstats.com

The article addresses three main questions:

1. What does Home Health Care mean for readmission?
2. What does readmission mean for hospital outcomes?
3. What can we generally expect from statistical adjustments, such as were used here to control for patients’ severity of illness?

Background

We tracked readmission rates for 228,041 hospitalizations at seven US hospitals -- East, Midwest, and West, from 2011-2017, between 17,000 and 52,000 at any given hospital. This involved approximately 182,000 individual patients. After discharge from hospital, about a quarter of these patients had received Home Health Care (HHC) services such as from a visiting nurse or a physical or occupational therapist. We sought to evaluate the effect that these services had on readmission rates.

Findings

On the surface, patients with HHC had higher readmission rates. But this was no surprise, since these patients would normally have more severe illnesses and injuries than the others; that typically is why HHC would be provided in the first place. So we took a conventional route and assessed this relationship while taking into account, for any given hospital, about 30 control variables. These included Number of Prior Admissions; Length of Stay; and presence/absence of diagnoses that usually mattered for readmission rate. Statistically adjusting for these factors, using logistic regression, was expected to reverse the difference we were seeing: to show that HHC decreased readmission rates.

Our model did supply evidence that HHC was indeed provided to more needy patients. But even when we comprehensively adjusted for those needs via the control variables, the overall readmission rate was still 21% higher for patients who received HHC (weighting each hospital equally). The rate was higher by 10% to 42% per hospital. (These are relative, not absolute percentages--e.g., a rate of 11% vs. 10% would entail a 10% relative difference.)

We adopted many other methods to try to make sense of this finding. For example, we assessed the group difference using propensity score analysis. While the difference in readmission rates sometimes shrank, it never disappeared, let alone reversing itself. This made us question whether the type of regression we were using was even capable of showing such a reversal.

Much research and analysis later, we determined that logistic regression -- the type used most often to analyze binary outcomes such as ours -- was indeed capable of showing a reversal if the data justified it. Moreover, we introduced a structure for determining whether the inclusion of control variables will produce a "sign reversal." The data simply didn't justify such a finding here. HHC undeniably increased readmission rates. We believe this was because the additional care these patients received was often accompanied by additional monitoring. When recovery goes awry, home health providers seem to recommend a return to hospital more readily than the patient him/herself or the family would.

But does this necessarily point to a bad thing? For all we know, many of these "additional" or in government terms "excess" readmissions may have been beneficial to the patient's health, and may even have saved lives. Our findings imply that “Home Health Care is beneficial to the health of a patient and it increases the probability of hospital readmission” is not a contradiction in terms. Readmission rate, rather than being seen as a health outcome like mortality or hospital-acquired infection rate, perhaps is better seen as a "triage decision."¹ Thus we join with many other researchers and commentators^{2, 3, 4, 5} who have recommended, for various reasons, that readmission rates no longer be used by the US government's Centers for Medicare and Medicaid Services as a basis for penalizing hospitals or HHC agencies.

¹As our collaborator Kenneth S. Boockvar, M.D. aptly put it.
² Adding socioeconomic data to hospital readmissions calculations may produce more useful results, by Nagasako, Reidhead, Waterman, and Dunagan.
³ Thirty-Day Readmissions — Truth and Consequences, by Joynt and Jha.
⁴ A meta-analysis of hospital 30-day avoidable readmission rates, by van Walraven, Jennings, and Forster.
⁵ More considerations on the present study, by myself and and L. Courtney.

I welcome any comments you might like to add, publicly or privately.

copyright 2019 by roland b. stark.

yellow brick stats homepage

my statistical and research consulting work