How does being part of a pediatric accountable care organization impact health service use for children with disabilities?

Abstract

Objective

To examine the impact of a Medicaid‐serving pediatric accountable care organization (ACO) on health service use by children who qualify for Medicaid by virtue of a disability under the “aged, blind, and disabled” (ABD) eligibility criteria.

Data Sources/Study Setting

We evaluated a 2013 Ohio policy change that effectively moved ABD Medicaid children into an ACO model of care using Ohio Medicaid administrative claims data for years 2011‐2016.

Study Design

We used a difference‐in‐difference design to examine changes in patterns of health care service use by ABD‐enrolled children before and after enrolling in an ACO compared with ABD‐enrolled children enrolled in non‐ACO managed care plans.

Data Collection/Extraction Methods

We identified 17 356 children who resided in 34 of 88 counties as the ACO “intervention” group and 47 026 ABD‐enrolled children who resided outside of the ACO region as non‐ACO controls.

Principal Findings

Being part of the ACO increased adolescent preventative service and decreased use of ADHD medications as compared to similar children in non‐ACO capitated managed care plans. Relative home health service use decreased for children in the ACO.

Conclusions

Our overall results indicate that being part of an ACO may improve quality in certain areas, such as adolescent well‐child visits, though there may be room for improvement in other areas considered important by patients and their families such as home health service.

1. INTRODUCTION

Accountable care organizations (ACOs) are a new model of health care delivery designed to improve the quality and efficiency of care. ACO formation has proliferated, with an estimated 923 public and private ACOs covering over 32 million lives in the United States in 2017.1 ACOs are groups of providers that assume responsibility for the care of a defined population and share in any savings associated with improved quality and efficiency of the care they provide.2 One of the key assumptions of the ACO model is that the alignment of financial and quality incentives will drive improved coordination of health service, which will in turn result in improved patient outcomes and efficiency.3

Increasingly, ACOs are integrating high‐risk populations who have a higher than average probability of health care utilization or cost.4 High‐risk populations often have complex medical needs requiring care and service from multiple providers and therefore have the potential to benefit substantially from coordinated care provided by the ACO model.5 In contrast, traditional fee‐for‐service (FFS) arrangements lack financial incentives for providers to coordinate care.

The ACO model is viewed as a potential solution to the critical need for coordinated care, and many state policy makers are actively promoting the ACO as a way to improve quality and lower costs in Medicaid programs.6, 7, 8 Yet the existing evidence about ACO performance focuses heavily on Medicare and commercial ACOs with relatively little empirical evidence on ACO performance for high‐risk Medicaid populations.9 Moreover, we have no knowledge of health service use of high‐risk pediatric patients under the ACO model. In this study, we aim to fill that gap by examining the impact of a Medicaid‐serving pediatric ACO on health care service use for children with disabilities resulting from a policy change in Ohio.

1.1. Children with disabilities and medicaid

Children with disabilities are at higher risk for poor physical, psychological, and social health than children without disabilities making them among the most vulnerable populations.10 In addition to their primary disabling condition, children with disabilities are more likely to have multiple chronic conditions and comorbidities.11 As a result, they utilize health care service more frequently and intensively compared to children without disabilities.12, 13 In addition, children with disabilities face significant barriers to accessing health care, greater out‐of‐pocket health care costs, and poorer health outcomes than children without disabilities.14

In most states, children who met the Social Security Administration's disability definition are eligible for Medicaid under the “aged, blind and disabled” (ABD) category. In some states, the income eligibility level for Medicaid may be set lower than that required for Supplemental Security Income (SSI) eligibility.15 Children that qualify for Medicaid by virtue of a disability account for slightly <5 percent of all Medicaid‐eligible children; however, they account for 26 percent of all Medicaid spending.16

The involved care needs of children with disabilities, along with concerns about health plans’ provider network adequacy, experience, and willingness to serve this population, have long been considered great enough to exempt children with disabilities from the managed care enrollment required of other Medicaid‐eligible children and to leave them instead in the traditional FFS system.16 However, faced with increased pressure to improve care for beneficiaries with complex needs and address budgetary concerns, several states are implementing strategies such as mandated managed care enrollment and/or are investing in alternative payment models such as ACOs for the pediatric population eligible for Medicaid as a result of their disability.7

1.2. Medicaid‐serving ACOs

There are currently twelve states with Medicaid ACOs; however, there are several ACOs that largely serve Medicaid populations.17 Partners for Kids (PFK) is one such ACO and is the focus of this study. PFK is the nation's largest and oldest pediatric ACO focused primarily on Medicaid‐enrolled children. PFK assumes responsibility for both primary and specialty care to more than 300 000 Medicaid‐enrolled children in central and southeastern Ohio stretching from urban Columbus to rural Appalachia. PFK assumes full financial risk for the medical care needs of Medicaid‐enrolled children in 34 of 88 counties in Ohio through agreements with all five Medicaid managed care plans operating in the state. This study uses the recent policy change in Ohio as a natural experiment to assess claims‐based outcomes of children with disabilities in Medicaid before and after joining an ACO model of care compared to similar children enrolled in non‐ACO managed care plans.

2. METHODS

2.1. Study setting

Prior to a state policy change in July 2013, all Ohio children eligible for Medicaid by virtue of a disability were enrolled in a traditional Medicaid FFS plan (“ABD‐enrolled children”). After the policy change took effect, ABD‐enrolled children were required to enroll in one of five Medicaid capitated managed care plans operating in Ohio. The policy change effectively moved ABD‐enrolled children residing in 34 counties in central and southeast Ohio into PFK's defined population for which the ACO assumed full financial responsibility for medical care. Financial responsibility for all other ABD‐enrolled children residing outside of PFK's geographic area was retained by the Medicaid managed care plans.

The non‐ACO children received the standard Medicaid managed care package which included: inpatient; outpatient; physician; Early and Periodic Screening, Diagnostic and Treatment (EPSDT); prescription drugs; behavioral health; dental; vision; and other core service.18 They also received value‐added service that could vary by plan and included 24/7 nurse hotlines, health education materials, care management, transportation benefits, incentives, and other benefits. None of the plans operated as integrated delivery systems, and each plan had to cover service provided by any Medicaid provider in the state on a fee‐for‐service basis. While there was some use of bundled payment, few applied to the pediatric population. Narrow networks were not permitted.

Accountable care organization‐enrolled children received all of the standard and extended Medicaid benefits associated with their chosen managed care plan. In this aspect, benefits for intervention and control samples did not differ. Care within the ACO was differentiated by an active approach to pediatric‐specific quality improvement projects, disease‐ and process‐specific learning networks, and financial incentive programs that were open to Medicaid‐serving pediatric providers in the designated region. The ACO worked within the same strictures of the plans including payment of fee‐for‐service claims from any Ohio Medicaid provider; however, over 1000 providers participated in and were eligible to receive performance‐based incentives associated with a project or network. The ACO also provided pediatric‐focused care coordination beyond the standard care coordination offered by the child's managed care plan. Thus, while both the managed care plans and ACO have strong incentives for cost control, we hypothesize that the ACO's emphasis on quality improvement and care coordination alongside full financial risk will result in improved health process and utilization outcomes for children with disabilities in the ACO compared to similar children in non‐ACO managed care plans.

2.2. Data and study sample

Our analytic data come from Ohio Medicaid administrative claims data for years 2011‐2016. Administrative data included service provided through the ACO as well as those directly reimbursed through Medicaid. Our sample included children ages 18 or younger enrolled in Ohio Medicaid under the ABD eligibility criteria between August 2011 and June 2016.

We identified 17 356 ABD‐enrolled children who resided within the PFK counties as the ACO “intervention” group and 47 026 ABD‐enrolled children who resided outside of the PFK counties as non‐ACO controls. All children enrolled in Medicaid under the ABD eligibility criteria before July 1, 2013, were covered by Medicaid FFS; as of July 1, 2013, ABD eligibility resulted in ACO enrollment (intervention) or other managed care (control). The cohort was open and thus included new entrants to ABD eligibility during the study period. All months of Medicaid eligibility during the study period for ABD‐enrolled children were included, even if the child was Medicaid‐eligible through a different (ie, non‐ABD) program for some of the study period. In sensitivity analysis, we confirmed that results were robust to the restriction of only months of Medicaid coverage in the ABD program. Observations from children who were enrolled in managed care plans prior to the policy change (n = 11 137, nT = 22 563) or those who were not enrolled in managed care plans after the policy change (ie, children who were granted waivers to remain in Medicaid FFS) (n = 41 714; nT = 67 899) were not included in the study sample. The final sample included 47 358 annual observations from 17 356 children in the ACO, and 142 579 annual observations from 47 026 children not in the ACO.

2.3. Analytic approach

We used a difference‐in‐difference (DID) design based on the policy change in Ohio related to ABD‐eligible children. We examined differences in patterns of health care service use by children in the ABD eligibility category before and after enrolling in an ACO, compared with children who were in the same category who enrolled in non‐ACO managed care plans after the policy change. This study design employed a contemporaneous control group to reduce the threats to validity from other unobserved changes that may have occurred disproportionately during the ACO implementation period. In addition, this study design separates the effect of the FFS to capitation changes from the FFS to ACO change, focusing on the difference between switching into an ACO vs a non‐ACO managed care plan.

We ran person‐year linear models with person‐specific fixed effects to control for any time‐invariant differences that may exist between ACO and non‐ACO enrollees, possibly due to regional differences. The model controlled for chronic condition diagnoses recorded during the year and the number of months a child was enrolled in the Medicaid program in the year. The interaction term between residence in the PFK region and a postperiod indicator captures the effects of enrolling into an ACO after implementation. That is, the model was specified as follows:

$$y_{\text{it}}={\beta }_1{\text{ACO}}_{\text{it}}+{\beta }_2{\text{Post}}_t+{\beta }_3{\text{ACO}}_i\times {\text{Post}}_{\text{it}}+{\beta }_4{X}_{\text{it}}++{\mu }_i+{\epsilon }_{\text{it}}$$

where i indexes children and t indexes years, X reflects time‐varying covariates, μ are the child fixed effects and ε are model errors. Standard errors were adjusted for clustering at the county level using non‐nested or multiway clustering.

We examined the parallel paths assumption through similar linear models using only the 2 years prior to implementation. We found no evidence of different time trends between the PFK and control regions for any of the outcomes examined here. We also used the propensity‐weighted technique proposed by Stuart and colleagues where control group and postperiod observations are weighted using a first‐stage propensity model for four groups based on timing and intervention status.19 Results were largely similar to those reported here and are provided in an Table S1.

2.4. Outcomes

We selected claims‐based process and utilization outcome measures expected to be influenced by the ACO model based on a review of the literature, existing recommendations by the Healthcare Effectiveness Data and Information Set (HEDIS), the Medicaid Child Core Set, and recommendations from our Patient Advisory Panel.20 These included annual health care utilization outcomes related to primary care, emergency department (ED) use, inpatient hospitalization use, and medication use based on the most prevalent diagnosis using the Chronic Illness and Disability Payment System algorithm.21 Our Patient Advisory Panel, whose members consisted of family caregivers and other stakeholders of children with disabilities, ensured that all outcome measures we used were meaningful to children with disabilities and their family caregivers. Per our Patient Advisory Panel recommendations, we added additional outcome measures, including utilization of commonly used classes of prescription medications among children with disabilities, home health service use, and use of physical, occupational, and speech language therapy (PT/OT/SLT). The definitions of outcome measures included in this analysis are given in Table 1. Outcomes were measured on an annual basis, except posthospital outcome measures (ie, follow‐up care after inpatient hospitalization) which were defined only after a hospital stay. All outcomes were binary except for the number of hospital days.

Table 1.

Study outcomes and definitions

Outcomes	Definition	Type of outcome	Hypothesized direction
Primary care
Any well‐child visit, age ≤6	An indicator of at least one well‐child visit for children under the age of 6	Binary	↑
Any well‐child visit, age ≥12	An indicator of at least one well‐child visit for children age 12 or older	Binary	↑
Any primary care provider visit	Outpatient visits with one or more evaluation and management code	Binary	↑
Immediate care use
Any emergency department visit	Visits to an emergency department that resulted in discharge without hospital admission	Binary	↓
Acute and postacute care
Hospitalizations	Admissions into hospital inpatient department	Binary	↓
Follow‐up after hospitalization within 7 d	Had any outpatient visit within 7 d after being discharged from a hospitalization (discharge level analysis, among patients who had any hospitalization)	Binary	↑
Follow‐up after hospitalization within 30 d	Had any outpatient visit within 30 d after being discharged from a hospitalization (discharge level analysis, among patients who had any hospitalization)	Binary	↑
Readmission within 30 d	Readmitted to the hospital within 30 d after being discharged (discharge level analysis, among patients who had any hospitalization)	Binary	↓
Hospital days	Days spent hospitalized in the inpatient department	Count	↓
Medication use
Any antidepressant	Use of any antidepressant prescription medication in a year	Binary	↑
Any anticonvulsant	Use of any anticonvulsant prescription medication in a year	Binary	—
Any antianxiety	Use of any antianxiety prescription medication in a year	Binary	↑
Any antipsychotic	Use of any antipsychotic prescription medication in a year	Binary	↓
Any asthma medication	Use of any asthma rescue prescription medication in a year	Binary	↑
Any ADHD medication	Use of any attention deficit/hyperactivity disorder (ADHD) prescription medication in a year	Binary	—
Other
Home health visit	Home‐based nursing, nurse aid, or other therapy service	Binary	↑
Home‐based PT/OT/SLT	Home care visits by physical therapy (PT), speech therapy (ST), or occupational therapy (OT) providers	Binary	↑
Office‐based PT/OT/SLT	Outpatient visits to these provider types in an office‐based setting	Binary	↑

—, no directional hypothesis based on existing literature.

3. RESULTS

Demographic characteristics of our samples are depicted in Table 2. Two‐thirds of the study sample were male, and the average age was approximately 12 years old in both the ACO and non‐ACO samples. White children represented over 65 percent of the ACO sample compared to 49 percent in the non‐ACO sample. African American children represented a greater proportion of the non‐ACO sample compared to the ACO sample (46 vs 29 percent).

Table 2.

Demographic characteristics of ACO sample compared with controls

Demographics	ACO sample mean (n = 17 356)	Control sample mean (n = 47 026)
Female	34.0%	34.5%
Age (y)	12.3*	12.7
Percent age ≤6	24.9%*	23.9%
Percent age ≥12	59.3%*	61.8%
Race
African American	29.4%*	46.0%
Native American	0.2%	0.1%
Asian/Pacific Islander	0.8%*	0.4%
Native Hawaiian	~0%*	0.1%
White (referent)	65.1%*	49.0%
Race unknown	4.5%	4.3%
Rural	31.0%*	11.4%
IDD	33.1%*	22.1%
Complex conditions	56.2%*	52.4%
Number of months on medicaid	32.3*	33.7

^*Difference between ACO and control sample mean is statistically significant at P < .05.

Unadjusted descriptive statistics on the outcomes used in our analysis are presented in Table 3. Prior to ACO implementation, the rate of well‐child visits is lower in the ACO region than the control region, although both groups increased in the postperiod. The rate of PCP visits was similar across groups. ED and inpatient use were lower among in the ACO region compared to controls, but were more similar in the postperiod. Follow‐up after hospitalization and readmissions were higher in the ACO region, but both groups decreased in the postperiod. Utilization between the ACO sample and the controls was comparable with respect to most measures of medication use, home health, and PT/OT/SLT service.

Table 3.

Annual descriptive statistics on health service use in the ACO sample compared with controls

Outcome	Pre‐ACO period (SFY2012‐2013)		ACO period (SFY2014‐2016)
	ACO sample mean (nT = 15 771)	Control sample mean (nT = 39 089)	ACO sample mean (nT = 33 587)	Control sample mean (nT = 130 490)
Primary care
Any well‐child visit, age ≤6	53.2%*	56.6%	58.3%	59.5%
Any well‐child visit, age ≥12	29.2%*	34.0%	35.9%*	38.8%
Any primary care provider visit	75.3%	71.4%	77.9%	76.7%
Immediate care use
Any emergency department use	31.0%*	38.4%	36.1%	40.6%
Inpatient use
Any hospitalization	8.1%*	8.9%	6.6%	7.5%
Hospital days [mean (SD)]	1.4* (14.0)	2.1 (21.8)	0.8* (8.4)	1.0 (9.7)
Posthospital indicators (obs)	5351	15 158	5251	16 810
Follow‐up after hospitalization in 7 d	36.0%*	31.4%	33.7%*	28.6%
Follow‐up after hospitalization in 30 d	71.2%*	63.2%	67.8%*	59.9%
Readmission within 30 d	22.4%*	20.7%	19.1%	18.3%
Medication use
Any antidepressant	12.7%	11.8%	11.5%	11.5%
Any anticonvulsant	18.5%*	15.6%	11.7%	9.3%
Any antianxiety	7.3%*	6.2%	5.2%	4.3%
Any antipsychotic	16.9%	16.0%	12.9%	13.7%
Any asthma medication	15.1%	14.9%	13.4%	13.1%
Any ADHD medication	25.6%*	25.2%	27.4%	29.0%
Other
Home health	30.4%*	18.4%	17.2%	16.5%
Home‐based PT/OT/SLT	0.3%	0.4%	0.01%	0.1%
Office‐based PT/OT/SLT	37.0%	35.7%	34.3%	34.6%

Abbreviations: nT, number of person‐year observations; SD, standard deviation.

^*Difference between ACO and control sample mean is statistically significant at P < .05.

Average marginal effects from our difference‐in‐difference analyses are presented in Table 4. Although we found no difference in well‐child visits among young children (age 6 and under) after the implementation of the ACO, we did observe a greater probability of adolescents receiving a well‐child visit in the ACO; the average marginal effect reflects a 5.1 percentage‐point higher rate of well‐child visits in the ACO region as compared to the non‐ACO region of the state (P < .05). This reflects a 14 percent relative increase in the rate of well‐child visits among adolescents in contrast to the control group mean (5.1/37.5 percent = 14 percent). No differences were observed in the broader measure of use of primary care visits after adjusting for covariates, possibly due to the relatively high level of primary care use in both groups.

Table 4.

ACO marginal effects on health service use

Outcome	Obs	Coefficient	95% Confidence interval		P < .05
Primary care
Any well‐child visit, age ≤6	34 341	−0.0064	−0.045	0.032
Any well‐child visit, age ≥12	97 145	0.0506	0.017	0.084	*
Any primary care provider visit	191 937	−0.0005	−0.0171	0.0161
Immediate care use
Any emergency department use	191 937	0.0163	−0.0062	0.0388
Inpatient use
Any hospitalization	191 937	0.0037	−0.0053	0.0145
Hospital days	191 937	−0.0207	−0.192	0.151
Posthospital indicators (obs)
Follow‐up after hospitalization in 7 d	42 570	0.0151	−0.0129	0.0431
Follow‐up after hospitalization in 30 d	42 570	−0.0066	−0.0328	0.0197
Readmission within 30 d	42 570	−0.0153	−0.0421	0.0116
Medication use
Any antidepressant	191 937	0.0072	−0.0034	0.0179
Any anticonvulsant	191 937	0.0058	−0.0049	0.0165
Any antianxiety	191 937	0.0054	−0.0020	0.0128
Any antipsychotic	191 937	−0.0039	−0.0111	0.0337
Any asthma medication	191 937	0.0026	−0.0096	0.0148
Any ADHD medication	191 937	−0.0110	−0.0191	−0.0030	*
Other
Home health	191 937	−0.0450	−0.067	−0.023	*
Home‐based PT/OT/SLT	191 937	0.0019	−0.0013	0.0052
Office‐based PT/OT/SLT	191 937	0.0002	−0.036	0.037

Standard errors are adjusted for clustering at the county basis.

^*Difference between ACO and control sample mean is statistically significant at P < 0.05.

We observed no statistically significant difference in any of the measures of ED visits, hospital utilization, or follow‐up with outpatient providers after a hospitalization. The implementation of the ACO also did not significantly change the number of hospital days per year.

We observed a decline in the use of attention deficit/hyperactivity disorder (ADHD) medication in the ACO group after joining the ACO as compared with the non‐ACO managed care controls. The remaining five medication classes (antidepressants, anticonvulsants, antianxiety, antipsychotics, and asthma rescue medications) showed no differences in the probability of use between ACO and non‐ACO children.

Finally, our model estimates that there was a large proportionate decline in access to Medicaid‐paid home health service after ACO implementation, in contrast to trends in the comparison group. The size of this effect was a 4.5 percentage‐point decrease in use of home health service (P < .05), which is a 26 percent relative decline over the rate in the non‐ACO population. There was no statistically significant change in the use of either PT/OT/SLT service delivered in the home or through office‐based PT/OT/SLT service.

Robustness analyses using propensity score weighting found very similar results (Table S1), although three additional outcomes (primary care use, the use of anticonvulsant medications, and the use of medications for asthma) became significantly higher with ACO enrollment. We conducted several subgroup analyses to test the robustness of our main findings to different subsamples. Subgroups of ABD‐eligible children examined included those with residence in rural and urban counties consistent with the National Center for Health Statistics Urban‐Rural Classification Scheme for Counties, intellectual and developmental disabilities diagnoses (IDD) following the Chronic Illness and Disability Payment System Definition and complex chronic conditions based on the Pediatric Medical Complexity Algorithm (PMCA).20, 21, 22, 23 Our subgroup analyses (reported in an Table S1) were generally similar to the findings from the main analytic model, although a few differences were notable. We found that the increase in well‐child visits for adolescents was not significant in rural children in the ACO, but did occur in urban children and children with complex chronic conditions. We also saw no difference in the use of inpatient service in the general population, but did see an increase in the use of hospital care among urban ACO enrollees. Finally, the reduction in the use of home health service in the ACO was similar in magnitude in subgroups with IDD, complex chronic conditions, or in urban areas.

4. CONCLUSION AND DISCUSSION

Alternative payment models continue to gain momentum, and new ACOs are developing across the country. While Medicare ACOs have largely dominated the ACO market, there is increasing interest and growth in Medicaid and pediatric ACOs.24, 25 To date, twelve states have sponsored Medicaid ACOs and 10 other states are pursuing or exploring Medicaid ACO programs.26 The results from this study can offer guidance to these states and others considering alternative payment models to help them better understand the likely impact of ACOs on health service utilization for children with disabilities.

Our findings suggest that being part of the ACO was particularly influential on preventative service, particularly for adolescents, as compared to similar children in non‐ACO capitated managed care plans. This may reflect specific efforts by the ACO to prioritize adolescent health well‐child visits as reflected in their physician incentive program.27 Although primary care visits were similar between the two groups, access to primary care for children both in the ACO and non‐ACO regions is relatively high suggesting that access to primary care may not be problematic for this population and thus not a focus of the ACO's efforts.

Our results show that the ACO had very similar rates of five categories of medications among children with disabilities as compared to other capitated managed care plans. Given that child and adolescent depression and anxiety are substantially under‐detected and under‐treated in the United States, particularly among children with special health care needs, thus it is somewhat surprising that we do not find differences in these underutilized medication classes.28 We did find a comparative decrease in the use of ADHD medications in the ACO enrollees. Table 3 reports that the rates of ADHD medication use are increasing in both groups over the study period, but the growth is smaller in the ACO region, yielding a negative difference. Because we do not focus in this analysis on children who may have ADHD diagnoses, it is difficult to form a value judgment on this finding.

Finally, our analysis revealed a relative reduction in access to home health use among children in the ACO as compared to similar children in non‐ACO capitated managed care plans. The existing literature is scant on home health utilization by children with disabilities, and home health service are determined by the managed care plans’ benefit design, not the ACO. However, we offer a few possible reasons to explain our findings. One possible explanation is that our results may detect a change in the supply of home health providers in the ACO region during the ACO implementation period. Although our current data do not have information on the actual contractual arrangements made between home health agencies and managed care plans, the number of home health agencies in the ACO region did not decline relative to that in the non‐ACO region during the observation interval. Alternatively, we cannot rule out that this difference may reflect a change in billing practices by ACO providers as compared to other managed care providers.

This study has several important limitations that should be noted. First, it is based on a specific population involving one large ACO in one state, thus this may limit the generalizability of our findings. For example, our study sample had a greater proportion of white children in the ACO sample compared to the non‐ACO sample likely reflecting the inclusion of Appalachia Ohio counties in the ACO region. African Americans make up 22.9 percent of Ohio's population residing in non‐Appalachia counties but only 8.9 percent in Appalachia.29 However, our study adds to the limited literature on ACO performance for pediatric patients and provides a basis upon which future studies can build. Second, although our analytic approach is among the strongest of observational study design, if any other factors disproportionately occurred in ACO‐enrolled children in the postperiod as compared with controls, the difference‐in‐differences analysis is at risk of attributing their effect to the implementation of the ACO. In addition, claims data do not contain complete information on all conditions experienced by insured populations, but rather those diagnoses noted by providers in billing data, nor do claims data capture service that were not reimbursed by Ohio Medicaid (eg, paid entirely out‐of‐pocket). Our study encompasses the initial 3 years following ACO implementation for the population included in this study. It may be the case that investments in health care service made by the ACO in its early years do not pay off in terms of utilization differences for several years postimplementation. Finally, our data do not allow us to adequately evaluate impact of the ACO on health care costs. While the impact on costs could be approximated using publicly available data sources, any resulting estimates would have inherent limited interpretability due to measurement error and the inability to reflect the specific provider markets included in this study. Moreover, while our study found changes in utilization for certain outcomes, the magnitude of these changes suggests the impact of the ACO on total costs of care for this population would likely be insignificant.

Despite these limitations, our study provides important insights into ACOs and policy makers on what impact being part of an ACO may have on the health service utilization by children with disabilities. For example, longer attribution in the ACO may be necessary to have impacts on primary care or other service. However, other dimensions of utilization such as medication use may be observable in the near term after joining the ACO.30 Given the importance of home health service cited by our Patient Advisory Panel, ACOs, policy makers, managed care companies, and other decision makers may want to focus on improving access to home health service. While our study does not allow us to evaluate cost of care under the ACO model for children with disabilities, our overall results indicate that being part of an ACO may improve quality in certain areas, such as adolescent well‐child visits, though there may be room for improvement in other areas considered important by patients and their families such as home health service. Our findings are consistent with a previous study of this ACO that evaluated the impact on quality for its broader population.31 Future studies should include additional outcomes that are meaningful to this population, such as missed school days, as well as understanding how changes in preventive service impact future health outcomes.

Supporting information

 

 

Song PH, Xu WY, Chisolm DJ, et al. How does being part of a pediatric accountable care organization impact health service use for children with disabilities? Health Serv Res. 2019;54:1007–1015. 10.1111/1475-6773.13199