Currently Viewing:
The American Journal of Accountable Care March 2019
Safety Net Representation in Federal Payment and Care Delivery Reform Initiatives
J. Mac McCullough, PhD, MPH; Natasha Coult, MS; Michael Genau, MS; Ajay Raikhelkar, MS; Kailey Love, MBA, MS; and William Riley, PhD
ACO Use of Case Mix Index to Comprehensively Evaluate Postacute Care Partners
Mark E. Lewis, MPH; and Avery M. Day, MPH
Improvement of Outpatient Quality Metrics in a Limited-Resource Setting
Carolina dos Santos, BA; Torkom Garabedian, MD; Maria D. Hunt, LPN; Schawan Kunupakaphun, MS; and Pracha Eamranond, MD, MPH
Currently Reading
Improved Cost and Utilization Among Medicare Beneficiaries Dispositioned From the ED to Receive Home Health Care Compared With Inpatient Hospitalization
James Howard, MD; Tyler Kent, BS; Amy R. Stuck, PhD, RN; Christopher Crowley, PhD; and Feng Zeng, PhD
Making Sense of Changes in Healthcare: Lessons From the AcademyHealth National Health Policy Conference
Jaime Rosenberg

Improved Cost and Utilization Among Medicare Beneficiaries Dispositioned From the ED to Receive Home Health Care Compared With Inpatient Hospitalization

James Howard, MD; Tyler Kent, BS; Amy R. Stuck, PhD, RN; Christopher Crowley, PhD; and Feng Zeng, PhD
A retrospective analysis of Medicare claims was used to study emergency department (ED) dispositions, specifically evaluating inpatient admissions compared with home health referrals.
Outcome Variables

This research compares the costs and utilization for the 2 cohorts. Cost variables include the amount of out-of-pocket (OOP) dollars the patient paid and the amount of Medicare reimbursement, including total reimbursements for the episode of care and costs in the 90 days following the index visit. The cost for the index episode of care included all facility fees, physician fees, and durable medical equipment. The 90-day follow-up costs included all costs from inpatient, outpatient, skilled nursing facility, home health, and hospice claims from both institutional and carrier claims. We used 2 common hospital measures to track utilization within 90 days of the index visit: (1) percent of inpatient hospital readmissions and (2) percent of ED revisits.18

Analytical Models

Regression models were used to compare the costs and utilization of the ED/hospital for the 2 cohorts. Ideally, the characteristics of these cohorts should be comparable. To account for differences among member profiles of the 2 cohorts, we used propensity scoring to balance the characteristics of the cohorts. The propensity score indicates the probability of a patient being in a group after all observable characteristics are controlled for. Through this technique, researchers can conveniently control for the probability of belonging in the case group (home health) or the control group (inpatient) and create a quasi-experimental design.19,20

Typically, a logistic model is used to generate a propensity score. In this research, a simple logistic model was problematic in generating propensity scores because the sample sizes of the patient cohorts were imbalanced. The number of observations in the inpatient cohort was about 50 times the number of observations in the home health cohort. Ideally, the probability of being in a group should be higher than 0.1 because the logistic model may underestimate rare events.21,22 To address this issue, we first drew a 10% random sample from the inpatient cohort, then we combined this 10% random sample with the home health cohort and used a logistic model to calculate the probability of being in the home health cohort.23 Covariates for the propensity score model included age, gender, Medicaid eligibility, geographic locations, types of disease, a history of hospitalization, and Charlson Comorbidity Index (CCI) score.

We used a propensity score–weighted regression to evaluate costs and utilization of services.24-26 The propensity score weighting method was used for 2 reasons. First, there were relatively few observations in the home health cohort. Using propensity score matching further risked reducing the number of observations in the home health cohort. In addition, research has shown that propensity score weighting is more robust to misspecifications of regression models. It is more likely to generate an unbiased estimate of treatment effect, even when the model is not specified correctly.25


Descriptive Statistics

A summary of the descriptive statistics by cohort, weighted and unweighted (available in eAppendix), shows that before propensity score weighting, observations in the 2 cohorts were different in several ways, including demographics, comorbidities, Medicaid eligibility, and geographical locations. Upon weighting, cohorts were more evenly balanced. The differences in age groups, average age, gender, Medicaid status, principal diagnosis, CCI score, and history of hospitalization became statistically nonsignificant. The only remaining statistical differences were geographical location and having a comorbid cancer diagnosis. Most important, once these groups were weighted, they had similar propensity scores: 0.0198 for inpatient versus 0.0200 for home health (P = .812). This means that after weighting, patients in the inpatient cohort had near identical opportunities to be in the home health cohort.

The average actual OOP costs for the patient, average Medicare payment, and total average reimbursement for the index episode and the 90-day follow-up period are shown in Figure 1. The average reimbursements for the home health cohort were much lower than those for the inpatient cohort. The average Medicare reimbursements for the home health cohort were $986 for patient OOP costs, $12,025 for the episode, and $13,012 for the 90-day total. For the inpatient cohort, reimbursements were $1965, $18,248, and $20,325, respectively (P <.0001).

Figure 2 summarizes the 90-day inpatient hospital readmissions and ED revisits. Patients in the home health cohort had lower inpatient hospital readmissions at 23.7% (n = 84) compared with 32.8% (n = 5790) readmissions for patients in the inpatient cohort (odds ratio [OR], 3.796; P <.0001). There were no statistically significant differences in 90-day ED revisits between the home health cohort at 39.0% (n = 138) and the inpatient cohort at 42.7% (n = 7496) (P = .2349).

Figure 3 provides least squares mean linear regression modeled average cost differences for the index episode and the 90-day follow-up period: OOP, Medicare reimbursements, and total reimbursements comparing the inpatient and home health cohorts.

Regression Results

Propensity score–weighted regressions were used to model the impact of discharges to home health as opposed to inpatient admissions from the ED. The average OOP costs to the patient, Medicare payment, and total reimbursement difference for the index episode and 90-day follow-up were $975, $6338, and $7425, respectively, which means that the inpatient cohort costs were greater than those of the home health cohort with statistical significance (P <.0001).

The ORs for ED revisits and inpatient hospitalization within the 90-day period are reported in Table 2. Overall, compared with the home health cohort, patients in the inpatient cohort were more likely to be readmitted to the hospital (OR, 1.535; P <.0001). The OR for ED visits in the 90-day follow-up period was not statistically significant, at 1.806 (P = .2128).

Copyright AJMC 2006-2020 Clinical Care Targeted Communications Group, LLC. All Rights Reserved.
Welcome the the new and improved, the premier managed market network. Tell us about yourself so that we can serve you better.
Sign Up