Currently Viewing:
The American Journal of Accountable Care March 2018
Currently Reading
Medicare Accountable Care Spending Patterns: Shifting Expenditures Associated With Savings
David B. Muhlestein, PhD, JD; Spencer Q. Morrison, BA; Robert S. Saunders, PhD; William K. Bleser, PhD, MSPH; Mark B. McClellan, MD, PhD; and Lia D. Winfield, PhD

Medicare Accountable Care Spending Patterns: Shifting Expenditures Associated With Savings

David B. Muhlestein, PhD, JD; Spencer Q. Morrison, BA; Robert S. Saunders, PhD; William K. Bleser, PhD, MSPH; Mark B. McClellan, MD, PhD; and Lia D. Winfield, PhD
From 2013 to 2016, successful Medicare Shared Savings Program accountable care organizations reduced spending by shifting expenditures from the inpatient and postacute care setting to the physician office setting.

Objectives: To examine whether Medicare Shared Savings Program (MSSP) accountable care organizations (ACOs) changed their expenditure patterns from 2013 to 2016 and whether those changes were associated with reduced spending.

Study Design: Fixed-effects regression model to assess changing expenditures over time and the effect of expenditures on savings rate. Multiple sensitivity analyses were conducted to ensure consistent results.

Methods: MSSP public use files were the source of the data.

Results: MSSP ACOs that achieved financial savings were associated with less spending on inpatient and skilled nursing facility care and more spending on physician services. On average, MSSP ACOs reduced expenditures on inpatient care, postacute care, ambulatory services, hospice care, and durable medical equipment, and they increased expenditures on care provided in the physician office setting.

Conclusions: Our findings suggest that ACOs may be able to achieve savings by reapportioning resources to different sites of care. Further research is needed to determine how ACOs are able to shift their expenditures.

The American Journal of Accountable Care. 2018;6(1):In Press
The Medicare Shared Savings Program (MSSP) seeks to improve care quality while reducing unnecessary costs in traditional Medicare fee-for-service by adopting an accountable care organization (ACO) model. MSSP participants receive additional Medicare payments for delivering care below an established benchmark price and for meeting certain population-focused quality measures. As of early 2018, CMS reported 561 MSSP ACOs covering 10.5 million lives, making it the largest of Medicare’s ACO programs.1 This paper examines whether, in their first 4 years, MSSP ACOs have changed spending patterns and if these changes are related to ACO savings.

The MSSP is a young program, and researchers have had limited data to assess both the quality outcomes and financial performance of the participating ACOs. In 2013, the first payment year of the program, MSSP ACOs collectively spent less than their benchmarks and therefore received bonus payments from CMS, although bonus payments exceeded savings and resulted in a net loss to Medicare of $73.5 million.2 However, Chernew and colleagues have criticized the benchmark as a metric for evaluating ACO savings and argued that quasi-experimental methods produce more accurate results that show more significant savings.3 In 2014 and 2015, MSSP ACOs demonstrated some savings, and the evidence suggests that ACOs that have been in the MSSP longer are more likely to generate savings.4,5

Research on how ACOs have changed their utilization patterns has focused on the relationship between reductions in postacute care (PAC) spending and cost savings. McWilliams and colleagues found that the 2012 MSSP cohort saved money by reducing spending on inpatient care, skilled nursing facility (SNF) care, and home health care.5 In a subsequent study, McWilliams and colleagues found that MSSP participation was associated with less PAC spending.6

Other studies have shown how Pioneer ACOs have saved money by spending less on inpatient care, hospital outpatient care, and PAC, and by spending more on outpatient care in the physician’s office.7,8 A study by Hsu and colleagues found that 1 Pioneer ACO’s care management program reduced rates of emergency department visits by 6% and reduced hospitalizations by 8%.9

Our study adds to this literature by using 4 years of data to examine within-ACO changes and how MSSP ACOs apportioned resources from 2013 to 2016. We ask 2 questions: First, are MSSP ACOs changing their spending patterns across different care settings over time? Second, are spending patterns different among ACOs that have improved their savings rate compared with ACOs that have not?

Many studies attempt to compare ACOs with non-ACOs; however, our research examines changes within MSSP ACOs themselves, which allows us to observe longitudinal changes in care delivery. Of course, relying on quantitative data and statistical methods only allows us to observe associations in these trends, not to determine why and how MSSP ACOs are able to make these changes. These changes are also not necessarily driven by the organization becoming an ACO and may reflect a broader trend in patient management. Still, understanding these trends in expenditures can provide insights into what kinds of changes in care are occurring in ACOs and whether they are associated with reduced spending.



The data used for this analysis were primarily based on the MSSP ACO Public Use Files (PUFs) released annually by CMS.10-13 The PUFs contain more than 100 variables based on expenditures, beneficiary demographics, admissions, quality, and personnel information for all ACOs participating in the MSSP program.

The primary variables of interest for this study were the savings rate, total expenditures, and an indicator variable about whether an ACO saved in each year. The savings rate was defined as the percentage of savings relative to expected benchmark expenditures. This was the same as benchmark expenditures minus the total expenditures divided by the benchmark expenditures.

CMS made minor changes to the format of the PUFs after the first year of the MSSP. Consequently, certain variables are only available for 2014 to 2016 and were not included in the original 2013 PUF. This constrains some model specifications to only include the last years of data. However, certain variables from the 2013 PUF, such as the savings rate, could be successfully reconstructed based on other available variables. The 2013 PUF also included data on ACOs that entered in 3 separate rounds.14 In the original data, the expenditures were not annualized because some ACOs were in the program for more than 12 months. We annualized the data based on the start dates of the participating ACOs.

The benchmark expenditures were established by the historic performance of the ACOs, based on the defined population of Medicare beneficiaries for which that ACO is responsible. The benchmark was established prior to the 3 years of enrolling in the ACO and adjusted forward with the national growth rate. ACOs were rated relative to their own performance. Modifications to this benchmark rule have been implemented, but they were not in place at the time of this research.15


To investigate whether ACOs have reduced costs and how they have reallocated money to different sites of care, we performed 2 primary analyses. For these analyses, we used a fixed-effects regression model to evaluate changing expenditures over time and the effect of expenditures on savings rate. We used a fixed-effects model for our primary analyses because we believed that certain ACOs are more likely to achieve better results based on unobserved characteristics. Every ACO’s unique characteristics may impact the savings rate or other outcomes we measured. To control for this, the fixed-effects model removed all entity-specific characteristics that were time-invariant by assuming that an individual ACO’s error term is not correlated with other error terms, but is unique to that ACO (confirmed using the Hausman test). This allowed us to focus on the effects of changing expenditures on the savings rate while excluding the entity-specific effects.

Multiple sensitivity analyses were conducted to ensure consistent results. We also examined the size of the ACO, the number of admissions, the rate of readmissions, and other ACO characteristics; however, none of these factors influenced the results, nor were they significant.

Analysis of Changing Expenditures

To analyze whether expenditure patterns are changing over time, we utilized an unadjusted year fixed-effects regression model. Each expenditure was used as a dependent variable with time as the independent variable. This method isolated the changes over time to each type of expenditure to create an unadjusted year fixed-effects model as described in the model below, where αi represents the ACO fixed effects and δt represents year fixed effects.
Expenditure Categoryit = δt + αi

Analysis of Expenditures on Savings Rate

We also used a fixed-effects regression analysis to evaluate how shifting expenditures affect performance and savings.16 Fixed-effects regression analysis compares an ACO with itself over time to examine how an ACO changes. It examines the within-unit change in the dependent variable (ie, savings rate) based on within-unit changes of the independent variables (ie, inpatient expenditures). Doing this implicitly controls for all time-invariant characteristics of ACOs, which may be difficult or impossible to measure. Mechanically, this is done as follows in the model below, where αi represents the ACO fixed effects, δt represents year fixed effects, and βn represents the expenditure categories for each ACO within each time period.
Savings Rateit = βnExpenditure Categoriesitδt + αi

This type of analysis allowed us to focus on ACO characteristics that were changing over time, over which ACOs have a larger measure of control. We also used a beneficiary weighted fixed-effects model to test that our results were consistent. We used the latest year of beneficiary equivalent years, because the weights could not vary over time, and we found no major changes to our findings.

We extended the fixed-effects model to a logistic regression with fixed effects. This allowed us to use a binary variable as the dependent variable. We used the logistic regression with fixed effects to look at differences between ACOs that saved money compared with ACOs that did not achieve savings. We presented this analysis in addition to the analysis of how expenditures affect the savings rate. The independent variables are consistent with our previous models, but the dependent variable now takes the value of 1 if the ACO saved money in a given year and 0 otherwise.
Savedit = βnExpenditure Categoriesitδt + αi

Because many of the independent variables were collinear in this analysis and the coefficients were difficult to interpret for the logistic fixed effects, we verified the results using alternative specifications where the expenditure is the dependent variable as a sensitivity analysis. Whether an ACO saved or did not save is included as an independent variable. This showed which expenditures were statistically significant for ACOs that achieved savings compared with ACOs that did not save after controlling for other ACO expenditures.

Copyright AJMC 2006-2018 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

Sign In

Not a member? Sign up now!