Changes in Spending and Quality after ACO Contract Participation for Dually Eligible Beneficiaries with Mental Illness

Abstract

Background:

Fragmented care and misaligned payment across Medicare and Medicaid lower care quality for dually eligible beneficiaries with mental illness. Accountable care organizations aim to improve the quality and value of care.

Methods:

Using Medicare fee-for-service Part A and B claims data from 2009 – 2017 and a difference-in-differences design, we compared the spending and utilization of dually eligible beneficiaries with mental illness that were and were not attributed to Medicare ACO providers before and after ACO contract entry.

Results:

Dually eligible beneficiaries with mental illness (N=5,157,533, 70% depression, 22% bipolar, 27% schizophrenia and other psychotic disorders) had average annual Medicare spending of $17,899. ACO contract participation was generally not associated with spending or utilization changes. However, ACO contract participation was associated with higher rates of follow-up visits after mental health hospitalization: 1.17 and 1.30 percentage points within 7 and 30 days of discharge, respectively (p<0.001). ACO-attributed beneficiaries with schizophrenia, bipolar, or other psychotic disorders received more ambulatory visits (393.9 per 1000 person-years, p=0.002), while ACO-attributed beneficiaries with depression experienced fewer emergency department visits (−29.5 per 1000 person-years, p=0.003) after ACO participation.

Conclusions:

Dually eligible beneficiaries served by Medicare ACOs did not have lower spending, hospitalizations, or readmissions compared with other beneficiaries. However, ACO participation was associated with timely follow-up after mental health hospitalization, as well as more ambulatory care and fewer ED visits for certain diagnostic groups.

Implications:

ACOs that include dually eligible beneficiaries with mental illness should tailor their designs to address the distinct needs of this population.

Introduction

Although many efficacious treatments for mental illness exist, fragmented care and misaligned payment contribute to low care quality and poor patient outcomes, especially for populations with significant physical and mental health challenges. One such group is patients with mental illness who are dually eligible for Medicare and Medicaid; these individuals typically have complex clinical and social needs that are associated with high utilization and lower life expectancy.[1–3] Among people receiving treatment for behavioral health conditions, dually eligible beneficiaries have expenditures twice as high as other adults, and dually eligible beneficiaries with behavioral health diagnoses have nearly twice the spending as those without behavioral health diagnoses.[4] This population is thus ripe for interventions to improve quality and decrease potentially avoidable utilization.

Accountable care organization (ACO) contracts provide incentives to coordinate care and lower costs by limiting use of resource-intensive care while meeting quality benchmarks.[4] ACOs have spread rapidly in recent years; in 2021, there were approximately 1,000 ACO contracts across Medicare, Medicaid and private insurers.[5–8] There are multiple ways the ACO model may benefit patients with mental illness, including promotion of enhanced care management activities, which have been shown to improve outcomes in patients with severe mental illness and those with depression and at least one other chronic disease.[9, 10] Providers with ACO contracts may also choose to invest in multidisciplinary care management teams to reduce use of high-intensity services and promote better coordination. While not part of an ACO, a Pennsylvania program that deployed care management teams saw a subsequent drop in inpatient and emergency service use among patients with mental illness.[11] Additional ways ACOs may enhance care and reduce costs for dually eligible patients include: ‘whole person’ treatment (because the ACO is responsible for physical health measures and total cost of care), integrated primary care, cost-effective care settings and providers (e.g., preventing ED visits and inpatient hospitalizations), interoperability of electronic medical records, evidence-based treatment outside traditional medical care, and early identification and treatment.

However, scant evidence exists regarding whether and how ACO contracts may influence treatment of dually eligible beneficiaries.[12, 13] Our analysis considers the effects of Medicare ACO contracts on care for this population. Given the promise of ACO program incentives combined with unique opportunities to fund a broader array of services for people with Medicare and Medicaid coverage, we hypothesize that ACO participation would improve quality (e.g., result in more follow-up visits after hospitalization for mental illness) and lower emergency department visits and preventable admissions. Additionally, we hypothesize outpatient visits would increase as services shift from more to less intensive settings.

In this paper, we aim to better understand the efficacy of ACO contracts to address the needs of dually eligible beneficiaries with mental illness. We compare the spending, utilization, and selected indicators of quality in fee-for-service Medicare beneficiaries that were and were not attributed to ACOs from 2009 to 2017, which spans the start of contracts in 2012 and later.

Methods

To assess the impact of ACO contract participation on dually eligible beneficiaries with mental illness, we compare Medicare spending and outcomes between beneficiaries with mental illness attributed to Medicare ACOs and a comparison population attributed to non-ACO provider organizations. We used a difference-in-differences design to account for time-invariant differences between Medicare ACO providers and non-ACO providers, and differences between the pre-intervention and post-intervention periods (i.e., before and after ACO contract start). A large literature has raised concerns about difference-in-differences study designs in the context of variation in treatment timing.[14] In order to test for heterogeneous effects due to variation in treatment timing, we evaluated results for each implementation cohort, limiting the pre- and post-periods.

Data and study population

We used 100% Medicare fee-for-service administrative Part A and B claims data from 2009 – 2017 to compare changes in spending and utilization for dually eligible beneficiaries age 21 years and older among beneficiaries who were or were not exposed to a Medicare ACO program (exposures defined further below). We restricted analysis to patients with at least one inpatient or two outpatient diagnoses seven days apart for mental illness during the study period. Applicable diagnoses included: anxiety, depression, bipolar disorder, schizophrenia or other psychotic disorders, phobias, and personality disorders (See Appendix Methods Supplement for codes included). We limited analyses to Medicare beneficiaries with at least one month of eligibility for full Medicaid benefits in that calendar year (including cost sharing and services covered by Medicaid but not Medicare). To ensure comparability between ACO and comparison populations (i.e., to compare groups with parallel trends before start of ACO contracts), we further excluded beneficiaries living in health service areas with Medicare ACO penetration below the 25^th percentile in 2017 (about 8.5%).

Exposure Measures

We attributed beneficiaries to Medicare ACOs (Pioneer ACOs, Medicare Shared Savings Program [MSSP], and Next Generation ACOs) based on where they received the plurality of qualifying evaluation and management (E&M) visits. The study includes ACOs that implemented contracts in each year between 2012 and 2017. ACO attribution was updated annually using a consistent algorithm over the study period. Claims were linked (via national provider identifier and tax identification number) to organizations in ACO provider files from the Medicare program.[15] For beneficiaries attributed to ACOs with contracts that began in April or July 2012, we excluded 2012 claims from analyses to avoid a mixed year of exposure to an ACO contract, a practice used in related ACO research.[12]

Outcome Measures

The primary outcomes of interest included measures of annual Medicare spending per beneficiary and utilization measures capturing primary care and hospital-based care. Outcomes for analysis included: spending on acute care, skilled nursing facilities (SNFs), and psychiatric facilities; number of acute care or critical access hospital inpatient admissions; and annual counts of utilization potentially related to quality (emergency department visits, ambulatory evaluation and management (E&M) visits, and potentially avoidable hospitalizations, as defined by the Agency for Healthcare Research and Quality).[16] Additional outcomes included the proportions of hospitalizations for mental illness that had a follow-up visit to a mental health provider within 7 or 30 days of hospital discharge, adapted with some modification from HEDIS.[17]

Statistical Methods

Among our cohort of dually eligible beneficiaries with mental illness, we tested the hypothesis that beneficiaries experienced lower spending and less hospital-based service utilization when served by providers participating in Medicare ACO contracts, compared with beneficiaries attributed to non-Medicare ACO provider groups. Using a difference-in-difference design, our main specification estimated beneficiary-year outcomes as a function of time (year indicators) and beneficiary state of residence indicators, with a control group comprised of dually eligible beneficiaries with mental illness attributed to non-ACO provider groups. Standard errors were clustered by state of beneficiary residence. For details on specification and control variables, see Appendix Supplementary Methods.

The pre-period parallel trends assumption was assessed by estimating separate slopes for outcomes from 2009–2011 for the ACO and non-ACO groups (using the same set of covariates included in the main models), and testing equality of these slopes, using a p<0.05 threshold. For 10 of our 11 outcomes, there were no significant differences in spending or utilization trends prior to 2012 comparing beneficiaries assigned to provider organizations that would later start an ACO contract and comparison beneficiaries. An exception was ambulatory E&M visits, for which the slope for ACO-attributed beneficiaries was higher than for comparison beneficiaries (p=0.040).

To assess whether results varied by type of mental illness, we repeated analysis separately for two diagnosis groups: a) beneficiaries with bipolar disorder, schizophrenia, or other psychotic disorders at any time during the study period; and b) beneficiaries with depression at any time during the study period, but without any diagnosis of schizophrenia, bipolar disorder, or other psychotic disorders during the study period.

Sensitivity Tests

Due to potential confounding created by differences in the provider groups that participate in ACO models or beneficiaries served by ACO providers in a given year, we also implemented two additional specifications following Ouyagodé et al.: a) fixing attribution to ACO or control group based on the pre-ACO 2011 attribution, while requiring beneficiaries to have positive spending in either 2009 or 2010 to qualify for this cohort, and; b) same as (a), but also dropping the first observation year for each beneficiary, since attribution requires that beneficiaries must have positive spending (at least one eligible visit) while beneficiaries in good health managing chronic conditions without a visit are excluded from attribution.[18]

Limitations

The primary limitation of our study is the use of solely Medicare data. For the dually eligible population, Medicare pays for medical services, while Medicaid pays for cost sharing, long-term services and supports such as case management, residential or nursing home care, and psychosocial rehabilitation services. However, Medicaid does not provide reimbursement for care at Institutions for Mental Diseases, so Medicare would be the primary payer for those services. In this population, wraparound services like personal care, transportation, or other services typically excluded from Medicare’s covered benefits could be important. For any services where Medicaid pays for cost sharing, utilization and most spending will be captured in Medicare claims, so our research design would capture changes in utilization. In dually eligible beneficiaries younger than 65 with mental illness, Medicare paid more than 80% of inpatient, outpatient and drug spending, while Medicaid spending was concentrated in residential, case management, and partial hospital spending (median Medicaid spending was less than $1,000).[1] Because our design compares trends over time in beneficiaries who are and are not attributed to ACOs, these Medicaid-only services will affect our estimates only if they differ over time between the beneficiaries in and not in ACOs. Additionally, our analysis does not include Part D data, so we are unable to use changes in prescribing patterns as a marker of follow-up care.

Many programs were initiated by the Medicare and Medicaid programs in the past decade, including the creation of a federal coordination office. It is plausible that successful efforts under programs such as Duals Special Needs Plans (D-SNPs) in Medicare Advantage or the Financial Alignment Initiative (FAI) could have spillover effects into fee-for-service.[19] To date, there is little evidence of such effects. We calculated the share of beneficiaries who switched enrollment from traditional fee-for-service to Medicare Advantage (which includes D-SNPs and FAI plans) and found the rates were similar in magnitude for ACO-attributed and non-ACO-attributed beneficiaries (Table A1).

Finally, we did not test the effect of Medicaid ACO programs on our study population’s outcomes, though some participants were attributed to Medicaid ACOs. Medicaid ACOs differ dramatically across states, with highly variable benefit designs, incentive structures, and quality metrics. There is also a lack of consensus about which programs should be classified as Medicaid ACOs. We therefore felt that inclusion of such programs in our models would not yield useful or valid results.[20]

Results

Over the period 2009 to 2017, our final sample included 5,157,533 unique beneficiaries and 21,571,635 beneficiary-year observations (Table A2). From 2009–2011, a majority had disability as the original reason for entitlement (62%), and 51% were below age 65. Compared to the average fee-for-service Medicare beneficiary, this population was more likely to live in areas of poverty (35%), to be of Black race or Hispanic ethnicity (29%), to live in a nursing home (19%), and to have 3 or more comorbid conditions (Tables 1 and 2). Across all study beneficiaries, 38% were first attributed to any ACO provider between 2009 and 2011, which preceded the start of the first ACO contract. In the subsequent six years, the percentage of beneficiaries first attributed to any ACO provider in that year ranged from 2.8% to 5.8%. The remaining 36.6% of beneficiaries in our study were never attributed to an ACO provider. The average duration of beneficiaries’ first ACO attribution was 2.3 years (Table 2). Additional information on demographics and outcomes by pre/post status is shown in Table A2. In this large sample, many differences between groups in the pre-period were statistically significant, but few were clinically or economically meaningful in magnitude.

Table 1:

Demographic Characteristics of Study Beneficiaries by Attribution Status, 2009–2011

Characteristic	All Beneficiaries	Attributed to a Medicare ACO?
		Yes	No
A. Any Mental Illness
Beneficiary-years (N)	7,210,176	3,847,630	3,362,546
Age, mean (SD)	62.2 (18.1)	61.7 (18.2)	62.9 (18.0)
Female (%)	65.2	65.5	64.8
Black (%)	16.8	16.0	17.6
Hispanic (%)	12.2	11.8	12.6
Non-Hispanic white/other (%)	71.1	72.2	69.8
High poverty (by census tract of residence) (%)	35.4	33.4	37.7
Nursing home resident (%)	19.0	18.2	19.8
Disabled (original reason for entitlement) (%)	61.8	63.2	60.2
Partial dual eligibility (any months) (%)	6.0	6.1	5.9
Age under 65 (%)	50.7	52.3	48.9
MA penetration for >=65 for HSA of residence, mean (SD)	19.6 (14.4)	20.3 (15.0)	18.8 (13.7)
MA penetration for <65 for HSA of residence, mean (SD)	12.1 (11.5)	11.7 (11.3)	12.5 (11.6)
B. Bipolar, Schizophrenia, or other Psychotic Disorders
Beneficiary-years (N)	2,948,579	1,575,711	1,372,868
Age, mean (SD)	58.1 (18.3)	57.5 (18.3)	58.7 (18.3)
Female (%)	60.1	60.5	59.6
Black (%)	18.1	17.0	19.4
Hispanic (%)	9.2	8.8	9.6
Non-Hispanic white/other (%)	72.7	74.2	71.0
High poverty (by census tract of residence) (%)	35.3	33.4	37.6
Nursing home resident (%)	21.7	20.4	23.2
Disabled (original reason for entitlement) (%)	73.0	74.2	71.6
Partial dual eligibility (any months) (%)	6.1	6.1	6.1
MA penetration for >=65 for HSA of residence, mean (SD)	19.9 (14.7)	20.6 (15.4)	19.0 (13.9)
MA penetration for <65 for HSA of residence, mean (SD)	11.9 (11.4)	11.6 (11.3)	12.4 (11.6)
C. Depression w/o Bipolar, Schizophrenia, or other Psychotic Disorders
Beneficiary-years (N)	2,998,347	1,613,855	1,384,492
Age, mean (SD)	65.4 (17.1)	64.9 (17.3)	66.1 (16.9)
Female (%)	71.2	71.5	70.8
Black (%)	14.8	14.5	15.2
Hispanic (%)	14.2	13.9	14.6
Non-Hispanic white/other (%)	70.9	71.6	70.2
High poverty (by census tract of residence) (%)	35.1	33.2	37.3
Nursing home resident (%)	19.2	18.7	19.8
Disabled (original reason for entitlement) (%)	54.3	55.9	52.5
Partial dual eligibility (any months) (%)	6.1	6.3	5.9
MA penetration for >=65 for HSA of residence, mean (SD)	19.5 (14.3)	20.1 (14.8)	18.7 (13.7)
MA penetration for <65 for HSA of residence, mean (SD)	12.1 (11.5)	11.7 (11.3)	12.5 (11.7)

SOURCE: Authors’ analysis of Medicare claims and enrollment data for 2009–2011. NOTES: Acute care includes critical access hospitals. SD = standard deviation, MA=Medicare Advantage. Observations in Hospital Service Areas below 25^th percentile of ACO penetration excluded. Attribution to Medicare ACOs based on plurality of E&M visits. MA penetration is calculated among dually eligible beneficiaries with full Medicaid coverage. High poverty is an indicator for being in a census tract with 20% or more of the population below the 2010 federal poverty level. Partial dual eligibility refers to beneficiaries who qualify for Medicaid coverage of some expenses incurred under Medicare (including premiums and cost-sharing), but who do not have coverage for additional Medicaid benefits such as long-term services and supports.

Table 2:

Study Cohort Attribution and Clinical Characteristics, 2009–2017

	Any Mental Illness	Bipolar, Schizophrenia, or other Psychotic Disorders	Depression w/o Bipolar, Schizophrenia, or other Psychotic Disorders
N	5,157,533	1,993,853	2,215,837
Year first attributed to ACO provider, %
2009 to 2011 (prior to first ACO contract)	38.0	40.3	37.1
2012	4.8	4.9	4.8
2013	5.8	5.9	5.9
2014	4.6	4.4	4.7
2015	4.0	3.7	4.2
2016	3.5	3.0	3.7
2017	2.8	2.3	2.9
Average duration of first ACO attribution (for those ever in ACO), years	2.3	2.3	2.4
Clinical condition, % (ever in study period)
Behavioral health conditions
Anxiety	54.4	55.2	48.9
Bipolar Disorder	21.7	56.2	0.0
Depression	69.7	69.2	100.0
Schizophrenia and other psychotic disorders	26.5	68.5	0.0
Other Mental Illness	16.8	10.4	11.5
Substance Use Disorder	10.1	14.9	7.9
Physical health conditions
Coronary Artery Disease	15.5	13.2	17.6
Vascular Disease	35.9	35.0	38.8
Polyneuropathy	2.0	1.9	2.3
Cardio-Respiratory Failure and Shock	23.1	22.4	24.8
End Stage Renal Disease	2.3	1.9	2.6
Renal Failure (non-acute)	3.8	2.7	4.6
Diabetes	40.8	39.3	44.0
Cancer	13.9	11.6	15.6
Chronic Obstructive Pulmonary Disease	34.0	33.6	36.0
Congestive Heart Failure	31.5	27.5	35.8
3 or more HCCs	64.6	66.8	67.4
Died during study period	26.3	24.6	28.4

SOURCE: Authors’ analysis of Medicare claims and enrollment data for 2009–2017. NOTES: HCC=Hierarchical Condition Category. Mental Illness conditions required 1 inpatient or 2 outpatient diagnoses 7 days apart, within year. “Other Mental Illness” includes phobias, personality disorders (see appendix for specific codes), as well as combinations of single outpatient diagnoses of anxiety, depression, bipolar disorder and schizophrenia that total two or more. Non-mental illness conditions are based on HCC flags, version 22. “Any Mental Illness” column includes some beneficiaries who are represented in neither the “Bipolar, Schizophrenia, or other Psychotic Disorders” nor the “Depression w/o Bipolar, Schizophrenia, or other Psychotic Disorders” columns. Average duration refers to consecutive years attributed to same ACO as first ACO to which first attributed (rather than across all ACOs).

Medical care utilization in this patient population was high. Average annual spending per beneficiary in the pre-period was $17,899 and was similar across the ACO and non-ACO attributed beneficiaries (Table 3). Below we report difference-in-difference estimates comparing the pre- versus post- ACO contract start outcomes in beneficiaries that were and were not attributed to ACO providers. For dually eligible beneficiaries with mental illness, there were no total, acute care, or SNF savings and no reductions in ED visits, potentially avoidable hospitalizations (Prevention Quality Indicators), or changes in ambulatory E&M visits associated with Medicare ACO participation (Table 4). There was a small increase in SNF payments associated with Medicare ACO participation ($56, p=0.011). There were no significant changes in the proportion of hospitalizations that resulted in readmission related to ACO attribution.

Table 3:

Spending and Utilization by Attribution Status, 2009–2011

Spending/Utilization Measure	All Beneficiaries	Attributed to a Medicare ACO?
		Yes	No
A. Any Mental Illness
Beneficiary-years	7,210,176	3,847,630	3,362,546
Annual spending ($), Mean (SD)
Total	17,899 (34,221)	17,797 (34,066)	18,017 (34,398)
Acute care/critical access	6,356 (18,165)	6,430 (18,279)	6,272 (18,033)
SNF	1,590 (6,554)	1,592 (6,559)	1,589 (6,547)
Psychiatric hospital	209 (2,312)	203 (2,289)	216 (2,338)
Annual utilization (/1000 beneficiaries), Mean (SD)
Emergency Dept. visits	1,610 (3,212)	1,617 (3,229)	1,603 (3,194)
Acute/critical access admissions	680 (1,422)	679 (1,428)	682 (1,416)
Prevention Quality Indicator overall composite admissions	143 (554)	140 (551)	146 (557)
Ambulatory E&M visits	12,469 (10,409)	12,385 (10,350)	12,564 (10,475)
Annual follow-up/readmission measure, Mean (SD)
Mental Health 7-day follow-up (proportion)	0.38 (0.47)	0.39 (0.47)	0.36 (0.46)
Mental Health 30-day follow-up (proportion)	0.68 (0.45)	0.70 (0.45)	0.65 (0.46)
30 day readmission proportion	0.11 (0.22)	0.11 (0.22)	0.11 (0.22)
B. Bipolar, Schizophrenia, or other Psychotic Disorders
Beneficiary-years	2,948,579	1,575,711	1,372,868
Annual spending ($), Mean (SD)
Total	18,095 (31,820)	17,880 (31,948)	18,342 (31,671)
Acute care/critical access	6,153 (16,932)	6,172 (17,108)	6,131 (16,728)
SNF	1,685 (6,832)	1,660 (6,777)	1,713 (6,895)
Psychiatric hospital	479 (3,511)	465 (3,477)	494 (3,549)
Annual utilization (/1000 beneficiaries), Mean (SD)
Emergency Dept. visits	1,940 (3,836)	1,947 (3,852)	1,933 (3,819)
Acute/critical access admissions	698 (1,505)	692 (1,505)	705 (1,505)
Prevention Quality Indicator overall composite admissions	132 (542)	128 (537)	136 (549)
Ambulatory E&M visits	12,012 (10,419)	11,996 (10,436)	12,030 (10,401)
Annual follow-up/readmission measure, Mean (SD)
Mental Health 7-day follow-up (proportion)	0.38 (0.47)	0.40 (0.47)	0.36 (0.46)
Mental Health 30-day follow-up (proportion)	0.68 (0.45)	0.70(0.45)	0.65 (0.46)
30 day readmission proportion	0.12 (0.24)	0.12 (0.24)	0.12 (0.24)
C. Depression w/o Bipolar, Schizophrenia, or other Psychotic Disorders
Beneficiary-years	2,998,347	1,613,855	1,384,492
Annual spending ($), Mean (SD)
Total	18,924 (37,428)	18,840 (36,889)	19,022 (38,047)
Acute care/critical access	6,883 (19,534)	6,984 (19,725)	6,764 (19,308)
SNF	1,685 (6,693)	1,694 (6,721)	1,674 (6,659)
Psychiatric hospital	28 (724)	25 (697)	31 (755)
Annual utilization (/1000 beneficiaries), Mean (SD)
Emergency Dept. visits	1,462 (2,769)	1,469 (2,780)	1,454 (2,757)
Acute/critical access admissions	717 (1,414)	718 (1,424)	716 (1,403)
Prevention Quality Indicator overall composite admissions	160 (583)	158 (583)	163 (582)
Ambulatory E&M visits	13,276 (10,617)	13,164 (10,511)	13,408 (10,738)
Annual follow-up/readmission measure, Mean (SD)
Mental Health 7-day follow-up (proportion)	0.36 (0.48)	0.38 (0.48)	0.34 (0.47)
Mental Health 30-day follow-up (proportion)	0.69 (0.46)	0.71 (0.45)	0.67 (0.47)
30 day readmission proportion	0.10 (0.21)	0.10 (0.21)	0.10 (0.21)

SOURCE: Authors’ analysis of Medicare claims and enrollment data for 2009–2017. NOTES: Acute care includes critical access hospitals. SD = standard deviation, SNF = skilled nursing facility, E&M = evaluation and management visits. Observations in Hospital Service Areas below 25^th percentile of ACO penetration excluded. Mental Health follow-up measures reflect proportion of index hospitalizations that result in mental health visit with 7 or 30 days of discharge. Attribution to Medicare ACOs based on plurality of E&M visits.

Table 4:

Changes in Care for Dually Eligible Beneficiaries Associated with Medicare ACO Participation, 2009–2017

Annual Outcomes	Any Mental Illness	Bipolar, Schizophrenia, or other Psychotic Disorders	Depression w/o Bipolar, Schizophrenia, or other Psychotic Disorders
Annual spending ($) (95% CI)
Total	30 (−215 – 275)	184 (−98 – 465)	−63 (−323 – 197)
Acute care	−11 (−118 – 96)	52 (−63 – 66)	−38 (−156 – 80)
SNF	56 (13 – 98)*	76 (17 – 134)*	45 (−3 – 93)
Psychiatric hospital	−4 (−17 – 10)	−15 (−44 – 15)	1 (−3 – 5)
Utilization (/1000 beneficiaries) (95% CI)
Emergency department visits	−15.5 (−37.9 – 6.9)	−0.1 (−29.9 – 29.7)	−29.5 (−48.6 – −10.4)**
Acute care/critical access admissions	3.3 (−5.2 – 11.8)	8.6 (−1.0 – 18.2)	0.2 (−9.9 – 10.2)
Ambulatory E&M visits	191.1 (−2.2 – 384.5)	393.9 (155.2 – 632.6)**	60.0 (−145.6 – 265.5)
Prevention Quality Indicator	1.1 (−1.7 – 3.9)	2.4 (−0.5 – 5.4)	0.7 (−2.6 – 3.9)
7 day follow-up (%) (95% CI)	1.2 (0.6 – 1.8)***	1.2 (0.6 – 1.9)***	0.7 (−0.3 – 1.8)
30 day follow-up (%)(95% CI)	1.3 (0.8 – 1.8)***	1.3 (0.7 – 1.9)***	1.6 (0.4 – 2.9)**
Readmission proportion (%) (95% CI)	−0.1 (−0.2 – 0.1)	0.0 (−0.2 – 0.2)	−0.1 (−0.2 – 0.0)
Person-years	21,571,635	8,861,127	8,920,924

SOURCE: Authors’ analysis of Medicare claims and enrollment data for 2009–2017. NOTES: Acute care includes critical access hospitals. CI = confidence interval, SNF = skilled nursing facility, E&M = evaluation and management. Observations in Hospital Service Areas below 25^th percentile of ACO penetration excluded. Year 2012 observations for beneficiaries attributed to Medicare Shared Savings Program ACO contracts that were implemented in April or July of 2012 excluded. Prevention Quality Indicators are rates of potentially preventable hospitalizations (indicator 90) defined by the Agency for Healthcare Research and Quality. Except where otherwise noted, estimates are per beneficiary-year. “Any Mental Illness” column includes some beneficiaries who are represented in neither the “Bipolar, Schizophrenia, or other Psychotic Disorders” nor the “Depression w/o Bipolar, Schizophrenia, or other Psychotic Disorders” columns. Person-years for follow-up (readmission) were 900,209 (5,520,211) for any mental illness, 820,986 (2,205,144) for bipolar schizophrenia or other psychotic disorders and 72,953 (2,491,328) for depression without bipolar, schizophrenia or other psychotic disorders.

^*p<0.05

^**p<0.01

^***p<0.001

In analyses restricted to people with depression (without other serious mental illness), ACO attribution was associated with a decline in ED visits (−29.5 per 1000 person-years, p=0.003); this decline in ED visits was not observed in analysis restricted to those with bipolar disorder, schizophrenia, or other psychotic disorders (−0.1 per 1000 person-years, p=0.996). In the bipolar disorder, schizophrenia, or other psychotic disorders cohort, there was an increase in ambulatory E&M visits associated with ACO participation (393.9 per 1000 person-years, p=0.002), while no significant change occurred in the depression cohort (60.0, p=0.560, Table 4).

In the pre-period, the percentage of mental health hospitalizations that resulted in 7-day and 30-day follow-up for the whole cohort was 38% and 68%, respectively. Medicare ACO participation was associated with increases in follow-up with a mental health provider within 7 days of hospital discharge (1.17 percentage points, p<0.001) and within 30 days of discharge (1.30, p<0.001).

Sensitivity tests of our results using alternate specifications or stratification by age yielded qualitatively similar results. Coefficients varied between the main and alternate fixed attribution specifications accounting for selection, but the direction and magnitude of effects were generally similar (Tables A3.a and A3.b). Results stratified by age under 65 status (as of 2009) are shown in Tables A4.a and A4.b. In testing for heterogenous effects by treatment timing, we found that effects were qualitatively similar considering all start-date cohorts compared with results that averaged across cohorts. In cohort-specific analyses, there was no significant change in spending (and large confidence intervals around spending estimates), fairly consistent increases in follow-up after a hospitalization for mental illness, and an increase in E&M visits after ACO contract participation across most ACO cohorts (Table A5). A notable difference was a change in sign for number of E&M ambulatory visits for the 2015 entry cohort, though the result was not statistically significant.

Discussion

Dually eligible beneficiaries with mental illness have high levels of fragmented care, comorbid illness, and high spending (63% above the average Medicare beneficiary) [21], making them a natural target for interventions intended to improve quality and reduce high-cost utilization. Evidence suggests effective interventions, such as behavioral health integration, could be deployed by ACOs, given their incentives to deliver high-quality care at lower costs.[22] However, after six years of Medicare ACO contract participation, no savings were achieved. Medicare ACO contracts were associated with greater follow-up after hospitalization and more ambulatory visits for patients with bipolar disorder, schizophrenia, or other psychotic disorders, suggesting a shift to lower-cost outpatient settings that might be expected to be cost saving, yet this additional management did not substitute for readmissions, or lower ED or inpatient utilization overall. Higher use of outpatient services, however, does suggest greater access to care, which may result in better outcomes for more complex patients and those at risk for hospitalization in the long run.[23, 24]

The intractable nature of high spending among dually eligible beneficiaries with mental illness is disappointing after early evidence from Medicare’s Physician Group Practice Demonstration, the precursor to ACO contracts, suggested substantial savings for dually eligible beneficiaries.[25] However, our results are consistent with a Government Accountability Office report warning that integrating benefits or coordinating care for disabled dually eligible beneficiaries may not lead to savings.[26] Because most spending among people with mental illness addresses physical health conditions, there is a need for a more complete assessment of total medical and mental health spending. ACOs also have less ability to control care for nursing home patients, who make up almost one-fifth of dually eligible beneficiaries with mental illness. While many ACOs are now forming relationships with short- and long-term nursing facilities, they face challenges related to aligning incentives and sharing information.[27]

Our results complement earlier studies of how alternative payment models impacted mental health outcomes, which tended to find limited improvements in outcomes, but some improvements in mental health process-of-care measures and modest evidence of lower spending and utilization.[28, 29] Our work augments a study of early Medicare Shared Savings Program results by Busch and colleagues, which found no evidence of savings specifically on mental health-related spending for ACO-attributed beneficiaries with mental illness.[12] Whereas this study lacked substance use disorder claims for key phases of ACO contract participation and stopped with 2013 data, our study includes substance use disorder claims (CMS recently reversed its practice of redacting these data) and uses more recent data, permitting a longer implementation period for activities to benefit patients with mental illness.

While ACOs focus significant effort on reducing costs and improving care for complex patients through strategies like patient segmentation, care coordination, and patient activation[30], they may not be best positioned to affect mental health care. Based on data from the National Survey of ACOs, 84% of ACOs included responsibility for behavioral health treatment in the total cost of care for at least one of their ACO contracts, yet only 14% reported significant integration of behavioral health treatment into primary care.[31] With few exceptions, ACO programs lack robust mental health quality benchmarks.[32] Among the 33 quality measures used for assessing early Medicare ACOs, only one addressed mental health (screening for depression). Although CMS has added measures like Depression Remission at 12 Months, such quality measures are not scored, and ACOs report difficulty incorporating mental health quality measurement into clinical work flows.[33] To address concerns that ACO quality benchmarks might increase the diagnosis of depression over time by ACO relative to non-ACO providers, we analyzed the percent of beneficiaries (in each year) who had depression, stratified by ACO attribution. We found the proportion of our cohort diagnosed with depression grew slightly more in the control cohort (33% to 43% during the study period) than the ACO cohort (35% to 43%).

Conclusion

Costs and quality changed little for dually eligible Medicare beneficiaries with mental illness following entry into Medicare ACO contracts, yet follow-up after hospitalization for mental illness improved. The distinct needs within this population present unique challenges, and evidence from many innovations over the past decade targeting dually eligible beneficiaries will help improve the design of payment and delivery models for this challenging group.