Expanding community-based service options for children with serious emotional disturbance is a priority (
1,
2). Efforts to decrease the institutional placement of these children have focused on Medicaid Home and Community-Based Services (HCBS) Waivers for the children and their families. Recent findings provide evidence that these services are cost-effective and beneficial in terms of child and family mental health outcomes (
3–
7).
The New York State Office of Mental Health (NYSOMH) provides targeted case management (TCM) service options, with service intensities targeted to the emotional needs of the children served. NYSOMH also provides a Medicaid HCBS Waiver for children. The TCM and HCBS Waiver programs serve similar children. However, to qualify for the HCBS Waiver program the child must be at risk of placement in a psychiatric inpatient setting. This is determined by each New York State local government–designated single point of access (SPOA) for children and families.
HCBS Waiver costs are higher ($2,502 per child per month) than TCM costs ($525 per child per month). The HCBS Waiver program includes six wraparound-type services not otherwise funded through Medicaid: individualized care coordination, crisis response services, intensive in-home services, respite care, family support services, and skill-building services (
8). TCM individualized care includes accessibility to needed medical, community, and social services; educational services; and other services such as assessments, referrals, family supports, crisis intervention, monitoring, and follow-up. The HCBS Waiver and TCM programs are similar in terms of service approach, populations served, and duration. On average, children placed in either program remain for about a year. Children receiving services from the programs have similar profiles in terms of clinical diagnoses, behavior and symptoms, and family characteristics. In addition, both programs use a person-centered approach to service planning, delivery, and evaluation, with the goal of successfully keeping the child at home and in the community (
9–
11).
The need for individualized (specific to a particular child and family), comprehensive, community-based programs for children with serious emotional disturbance has received considerable discussion in the literature over the past few decades (
12–
15); however, only a handful of studies have examined the cost utility of these comprehensive programs (
16–
18). Several studies have explored the intensity of service use and cost-efficiency within HCBS Waiver programs, but no study has compared the HCBS Waiver program with TCM (
19,
20). Only a limited number of studies used either a self-control or matched-control design in examining the intensity of service use and clinical outcomes or cost (
6,
21,
22). The general cost savings of community-based programs compared with institutional care are well documented (
23); however, the literature is very scant with regard to cost comparisons among community-based programs. To our knowledge, no studies have compared service use patterns and cost across these community-based programs. Identifying service use patterns and comparing costs by program type will benefit these children, their families, and other stakeholders.
The objective of this study was to compare per member per month (PMPM) Medicaid cost for children who received HCBS Waiver services and those who received TCM. Specifically, service use patterns were determined for similar children served in the HCBS Waiver or TCM programs, and the programs’ relative impact on Medicaid cost was examined.
Methods
Study Setting
This was a pre-post quasi-experimental study with propensity score–matched comparison groups. The study population was selected from the NYSOMH administrative data systems, Child and Adult Integrated Reporting System (CAIRS), and Medicaid.
The study population consisted of 5,695 Medicaid-eligible children ages four to 18 admitted to TCM (N=3,223) or the HCBS Waiver program (N=2,472) from January 2009 to June 2010 and discharged between July 2010 and June 2012. The total data period spans five years from January 2008 to June 2013. The 12 months preceding admission and the 12 months postdischarge were considered as the pre and post periods. Children who were ever enrolled in both programs or at a residential facility were excluded. A propensity score–matched cohort was created consisting of children with Medicaid eligibility for at least 92% of the time before, during, and after TCM or HCBS Waiver episodes.
Measures
Characteristics of the child and family at admission that were used for propensity score matching were extracted from CAIRS, including demographic characteristics, behavioral health diagnoses, custody status, symptoms or behaviors, child strengths, family strengths, and custodian education level. Demographic characteristics were categorized as follows: age at admission (four to 18), gender (male or female), race-ethnicity (white non-Hispanic, black non-Hispanic, Hispanic, and multiracial), and residential region (New York City or rest of state). Custody status was characterized as living independently or with family or in foster care. Custodian education level was characterized as college or not. Psychiatric diagnoses were characterized as three variables: psychosis or not, depression or not, and attention-deficit hyperactivity disorder (ADHD) or not. Other behavior or symptom indicators were characterized as having the following behaviors ever or not: suicidal, dangerous, aggressive, or sexually improper. CAIRS scales related to child strengths, family strengths, and family needs were also utilized. Questions about family strengths, characteristics, and needs used Likert scales regarding the existence of conditions such as unemployment, domestic violence, housing, severe chronic illness, alcohol and drug abuse, and poverty. Likewise, questions on child strengths asked about regular school attendance, positive friends and family relations, anger management skills, enjoyment of a hobby, and respect for the rights of others. Total scale scores were calculated as the sum of items categorized as positive (present) or not.
Medicaid claims and encounters were used to derive the following variables: Medicaid eligibility for children in the HCBS Waiver program or TCM program; prior psychiatric hospitalization days, calculated as the total number of days paid by Medicaid for inpatient psychiatric claims before program enrollment; PMPM costs, estimated as the total monthly cost contributed by Medicaid-eligible children in the program that month. This person-month approach allowed for 12 member-months with different individual admission and discharge dates. Each child could contribute up to 12 member-months (series) in each period. The PMPM estimates for the period before program admission used only claims from the pre period. Likewise, the PMPM estimates for the period during enrollment in the HCBS Waiver or TCM program used claims during the enrollment period, and the PMPM estimates for the period after discharge from the HCBS Waiver or TCM program used only claims after discharge.
The outcomes of interest were PMPM Medicaid cost for all services (total cost) and for mental health services before and after program enrollment. Also of interest were the PMPM Medicaid cost for children with psychosis (schizophrenia diagnosis or psychotic symptoms) and those with ADHD (ADHD diagnosis or hyperactive or impulsive symptoms) at admission, typically identified as high needs. This study examined outcomes in terms of changes in cost, because the provision of services and health outcomes are generally tied to dollars spent.
Statistical Methods
Propensity score matching was used to create a comparison group. This approach provides an empirical method to identify similar children in the HCBS Waiver and TCM programs (
24–
26). Propensity scores were estimated by using logistic regression, and the outcome was admission to the HCBS Waiver program (scored as 1) versus TCM (scored as 0). Predictors were derived from an array of demographic, clinical, and symptom or behavior indicators prior to program admission. This included a rich set of child characteristics available in CAIRS, including demographic factors, psychiatric diagnoses, custody status, living situation, symptoms or behaviors, child strengths, family strengths, and custodian education level. A greedy matching algorithm with a caliper of .1 in the probability scale was used to create a 1:1 matched analytic cohort. Children admitted to the HCBS Waiver and TCM programs were matched on the estimated propensity scores and the number of psychiatric hospitalization days prior to admission (
27,
28). Balance in covariate distribution between children admitted to the HCBS Waiver or TCM programs in the matched analytic cohort was assessed with weighted standardized difference (
29,
30).
Two separate analyses, exponential smoothing (
31,
32) and difference in difference (DID), were considered. Exponential smoothing fits a time trend model that weights recent observed series (PMPM estimates) more heavily than remote observations. This is accomplished by specifying smoothing parameters for the time trend. It was used to remove random noise from the 12 member-month contributions, allowing for a better identification of the PMPM cost trends. Also, the PMPM estimates for the period before program enrollment (12 months) were used to forecast the future PMPM costs (two months) beyond program admission.
DID techniques were used to estimate the average annual difference in cost before and after TCM services subtracted from the average annual difference in cost before and after HCBS Waiver services (
33). A negative difference indicates a reduction in cost for children admitted to the HCBS Waiver program compared with TCM.
To examine whether the change in PMPM average cost significantly differed from before to after participation in the programs, unadjusted and adjusted DID analyses were calculated. The unadjusted DID analysis used the unmatched sample. The adjusted DID models used the matched cohort and generalized estimation equations with a log link and a gamma distribution for the error term. Bootstrapping techniques were used to construct 95% confidence intervals for the DID estimates (
34). Children from each program were separately sampled with replacement, estimates of average person-month cost for each child before admission to and after program discharge were obtained by using the model described above, and the DID model for each sample was estimated. This process of sampling, model estimation, and DID calculations was repeated 1,000 times. With 1,000 estimated DIDs, the 2.5th and the 97.5th percentile were obtained to construct a 95% confidence interval. All statistical analyses were performed with SAS software, version 9.4.
Results
The study population consisted of Medicaid-eligible children admitted to TCM (N=3,223) or the HCBS Waiver program (N=2,472). Of these, 1,602 children in the HCBS Waiver and 2,740 children in TCM were Medicaid eligible for at least 80% of the study period. The final propensity score–matched cohort included 1,307 HCBS Waiver and 1,307 TCM children with Medicaid eligibility for at least 92% of the time before, during, and after TCM or HCBS Waiver episodes. The median (25th and 75th percentiles) length of stay for children in TCM and the HCBS Waiver program were 12 (seven and 18 months) and 11 (six and 16 months), respectively. Detailed descriptive characteristics of the propensity score–matched study sample are presented in
Table 1. The standardized weighted differences for most of the variables included in the propensity score model were less than .1, indicating a good covariate balance in the matched analytic cohort (
Table 1).
The trends of total Medicaid cost and total mental health cost are shown in
Figure 1 and
Figure 2. The estimated annual PMPM total Medicaid cost and total mental health cost for HCBS Waiver children postdischarge decreased by 25% and 29%, respectively. In contrast, the PMPM total Medicaid cost and mental health cost for TCM children postdischarge increased by 15% and 12%, respectively. Similar trends were identified for total Medicaid cost when the sample was stratified by gender, age, and region (New York City versus rest of state) (data not shown). The exponential smoothing forecast for the PMPM expenditure for the next two months, which used estimates of the PMPM for the 12 months before program admission, were $6,009 and $6,233 for the HCBS Waiver program and $4,040 and $4,196 for the TCM program, indicating a continuous upward trend in both.
Table 2 summarizes the results from the unadjusted DID models. Average PMPM total Medicaid cost for children admitted to the HCBS Waiver program decreased by $1,190 from before to after the program. For children admitted to TCM, the average PMPM total Medicaid cost increased by $471 from before to after the program. Thus the unadjusted DID for children with similar characteristics indicated a relative decline of $1,661 in average PMPM total Medicaid cost attributable to the HCBS Waiver program.
Table 2 also shows that for children with psychosis (primary schizophrenia diagnosis or psychotic symptoms), the average PMPM total Medicaid cost for children in the HCBS Waiver program was $1,490 less than for those in the TCM program. Similarly, for children with a primary ADHD diagnosis or hyperactive or impulsive symptoms, the average PMPM total Medicaid cost for those in the HCBS Waiver programs was $1,590 less than for those in the TCM program.
Table 3 displays results from the adjusted DID models. Adjusted results indicated that the average PMPM total Medicaid cost for children admitted to the HCBS Waiver program decreased substantially from before to after the program by $498, whereas the cost for children admitted to the TCM program increased by $448 after discharge. Thus for these children, all of whom had a similar propensity to receive services from the HCBS Waiver program on the basis of the measured baseline covariates, the adjusted DID model indicated a statistically significant decline of $946 in average PMPM Medicaid cost attributable to program type. In the analyses restricted to children with primary psychosis and to children with primary ADHD, the adjusted DID models indicated a statistically significant decline of $1,026 and $962, respectively.
Discussion
This study is the first to compare preliminary Medicaid cost for children who received HCBS Waiver services and children who received TCM by using a quasi-experimental pre-post design. The study examined the trend in Medicaid cost for similar groups of children prior to admission to, during participation in, and after discharge from these programs, as well as the annual average PMPM cost from before to after each program. In this effort, a technique for creating matched pairs based on propensity score methods was used to create similar groups of children, and DID generalized linear models were used to evaluate the differences in Medicaid cost. Previous research has demonstrated the power of incorporating both DID and propensity score matching (
35–
38).
The study found that total Medicaid PMPM cost for both child groups increased rapidly in the 12 months leading to program admission. Based on the PMPM cost model projections, the upward trend in cost would likely continue over time if not for program intervention. As mentioned above, program intervention is determined by SPOA. Children identified as having significant mental health needs are referred to the SPOA Committee for review. On the basis of clinical information, child and youth needs and strengths assessment, and planning efforts with family, the committee determines the level of care to be TCM, HCBS Waiver, or another intensive program. The SPOA Committee then forwards the referral to the Pre-Admission Certification Committee specialist for consideration and placement.
Because of the nature of the clinical eligibility determined by SPOA for these programs, it is expected that some children may need short-term psychiatric hospitalization to stabilize a crisis prior to enrollment. For children preadmitted to the HCBS Waiver program who require hospitalization, their program slot may remain active during any hospitalization for up to 60 days of a 75-day period. However, if the child’s hospital stay is longer (or if it has been predetermined that the child will need long-term hospitalization), the child is disenrolled from the HCBS Waiver program. The hospitalization cost in this period may explain the difference in PMPM cost before program entry and may also explain the spike in cost right before enrollment.
Importantly, both interventions changed the trajectory after program placement; however, only the HCBS Waiver program appeared to have a lasting impact postdischarge, with a declining cost trajectory. The postdischarge trends are particularly important because states continue to transform children's mental health systems. This is particularly useful information as NYSOMH amends its current 1115 Medicaid Waiver.
This analysis showed that the higher-cost HCBS Waiver program appears to be cost-effective, stabilizing the cost of care for children with relatively higher costs during the program and changing the trajectory after the program. The findings are consistent with those of other studies that examined the effects of community-based programs and total Medicaid service cost (
16,
18,
20,
22). Research suggests that participation in the HCBS Waiver program is associated with potentially lower Medicaid cost (
23). This finding supports the evidence that to reduce costs associated with the care of children with serious emotional disturbance, an intensive community-based HCBS Waiver program or a wraparound program may be ideal. Although the cost of these wraparound programs is relatively higher, the cost savings postdischarge and potentially long term support this investment. In addition, this study demonstrated that the upward trend in cost prior to program admission would likely have been sustained over time if not for the intervention. This finding appears to underline a need to review the evaluation processes leading to a child’s being waitlisted and admitted or not admitted.
The study had some limitations. The PMPM cost included services paid for under a fee-for-service arrangement in the HCBS Waiver program, as well as encounter claims for services provided under Medicaid managed care, which are paid for on a capitated basis and may not reflect the direct care provided to the child and family. Second, the study cohort was restricted to children with at least 80% Medicaid eligibility before, during, and after the HCBS Waiver and TCM episodes prior to propensity score matching; however, eligibility was 92% for the final propensity score–matched cohort used in the analysis. Third, the study attempted to alleviate bias due to systematic differences between the two programs by using propensity score matching with a robust set of CAIRS indicators, but the method, although statistically sound, may not fully explain program placement. Unobserved or unmeasured variables could not be accounted for in the propensity score matching analyses and differences may have remained that could not be controlled because of the nature of the programs. Also, costs for children in the HCBS Waiver program were relatively higher before and during the program. This was not controlled in the propensity score matching because it was examined as an outcome. The child groups may have been similar in terms of adjusted covariates; however, identifying the reason that a child was placed in the TCM program and not in the HCBS Waiver program was beyond the scope of this study. Finally, although changes in cost from before to after the program are an important proxy for care and health improvement, this study did not focus on objective clinical indicators. Therefore, changes in dollars may not reflect changes in clinical outcomes.
Conclusions
Efficient identification of service use patterns and cost comparisons for community-based programs will benefit children with complex mental health needs, their families, and health care providers. The results were consistent with those of other studies and strongly support the need to increase investment in HCBS Waiver programs or similarly designed community-based programs for these children.
Perhaps most important, the study advances knowledge about the cost of community-based services and helps identify Medicaid cost patterns. This is important as NYSOMH remodels its current 1115 Waiver program. The upward cost trend prior to program enrollment also calls for a review of the health care system, especially during the few months leading to admission. Overall, the resource-intensive HCBS Waiver program appears to be effective, stabilizing Medicaid cost during the program for children whose care was relatively more expensive and resulting in a decline in Medicaid cost after the program. It is important to continue to investigate the community-based program qualification rules and their impact on Medicaid cost.