Cost analysis of a disease management programme for adult Medicaid clients with schizophrenia

Summary

This investigation assessed changes in direct medical costs, from the perspective of a public payer, associated with a comprehensive, field-based disease management programme for adult Medicaid clients with schizophrenia in the US State of Colorado.

A propensity score-matching algorithm was employed in this retrospective analysis owing to the inherent non-randomisation of enrollees. Of the 126 clients initially enrolled, 73 (58%) remained within the programme continuously for 6–12 months.

These participants were associated with 30% lower overall per member per month medical costs (p<0.001), although no differences were noted for overall pharmacy costs. Provision of the disease management programme was through an external vendor and cost $31,250 per month regardless of the number enrolled.

Future research should seek to assess long-term clinical, humanistic and economic outcomes in this population and to develop methods that increase programme participation.

Keywords: :

• disease management
• schizophrenia
• United States
• Medicaid
• pharmacoeconomics

Introduction

Relatively few studies within the scientific literature have evaluated schizophrenia disease management (DM) programmes, particularly within Medicaid client populations. Broad cost of illness investigations have reported that the economic burden of schizophrenia is substantial, with approximately 70% of total costs from productivity losses being attributed to impairments produced by the disease1. Furthermore, individuals with schizophrenia represent approximately 10% of the totally and permanently disabled population and utilise an estimated 2.5% of all healthcare expenditures per year2. Substantial gaps in care have also been identified, with an estimated 62% of inpatients and 29% of outpatients not receiving adequate dosages of antipsychotic medication alone. Provision of psychosocial services has been reported as being inadequate3,4. The cost of atypical antipsychotics has also resulted in a renewed interest in assessing economic outcomes of schizophrenia and differences in total direct medical costs between treatments.

The National Pharmaceutical Council (2004) conveyed that DM efforts are less established in schizophrenia than in other disease states, despite large clinical and economic burdens of illness5. Large-scale programmes targeting the management of schizophrenia have historically included the Program for Assertive Community Treatment (PACT), the Patient Outcomes Research Team and the Texas Medication Algorithm Project (TMAP)3–9. The National Institute of Mental Health's funding of the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE), which compare the effectiveness of antipsychotic medications, evaluated the outcomes associated with four typical treatment settings and represented the largest government study to compare the safety and efficacy of major second-generation antipsychotics10.

Given both the importance of managing schizophrenia and the coinciding lack of research, the overall purpose of the current research endeavour was to evaluate the changes in healthcare costs of a comprehensive DM programme for Medicaid clients with schizophrenia plus other medical co-morbidities within the US State of Colorado. More specifically, the objectives focused on assessing medical and pharmacy costs before and after DM programme implementation relative to propensity score-matched comparators. As noted in the National Pharmaceutical Council report (2004), this study seeks to provide a formal ex-post analysis of the impact of comprehensive interventions in schizophrenia and is one of the few studies present within the scientific literature5,11.

Patients and methods

This retrospective, observational study was established from the perspective of a public payer to evaluate Medicaid schizophrenia patients who were clients of a DM programme. This quasi-experimental, non-randomised research design employed a pre-post analysis for DM clients relative to a propensity score-matched comparator group. Comprehensive, longitudinal, computerised administrative medical and pharmacy claims data were used to evaluate changes in medical and pharmacy costs across co-morbidities. Data analysed included patient demographics, medical and pharmacy administrative resource utilisation claims and costs, diagnostic information (International Classification of Disorders 9th Revision (ICD-9-CM)), eligibility and enrolment information, and programme-related information from an independent, private, external DM vendor. Mental health carve-out encounters were not available and, therefore, could not be included in the analysis (i.e. behavioural health services such as group therapy, individual therapy, day treatment, group homes, inpatient hospitalisations and appointments with psychiatrists for medication-related issues). Carve-outs are broadly defined as health coverage mechanisms wherein both the management and reimbursement of specific benefits (e.g. behavioural or mental health) are administered separately from general programmes12. These carve-out plans often utilise specialty provider networks and protocols as a method to contain costs and improve quality.

Disease management group

The DM intervention examined within this investigation was implemented and conducted by an independent vendor with long-term national experience in community, field-based programmes to manage chronic diseases and associated co-morbidities across several disease states. Using a combination of group and individual interventional approaches, this vendor conducted meetings to interact directly with clients and with those impacting these individuals' lives. Importantly, the programme focused strongly on the management of medical co-morbidities associated with schizophrenia. According to McDonald et al, co-morbidities are present in substantially higher proportions in individuals with schizophrenia than in the general population13. Thus, although specifically developed for individuals with schizophrenia, these interventions targeted the management of other co-morbid disease states and also included activities designed to improve the clients' overall health. Examples of these approaches included, but are not limited to, the following:

• • educational sessions with the food preparation staff at group and boarding homes to target diabetic patients' specific dietary needs;

• • weekly walking group sessions to increase physical activity, reduce excess weight and promote both socialisation and psychological well-being (incentivised by rewarding participants with a trip to a coffee shop for small low-fat food treats and a celebration after the group members consistently participated for a specific number of walks);

• • group educational sessions regarding various topics including diabetes, dyslipidaemia, heart disease and other health-related topics;

• • face-to-face meetings with clients at cafeterias or soup kitchens to discuss how to make better choices for healthier diets, taking into account the client's co-morbid conditions (occasionally meeting clients at the grocery store to teach them how to make healthy choices and to stay within their monetary budgets);

• • attendance of physician visits (when invited) to support clients or to remind them of questions they wanted to discuss with their primary care providers and to be able to reinforce or explain the clinician's instructions to the client regarding their care management plan;

• • encouragement and facilitation of preventive activities that included weight management, vaccinations and increased exercise.

Initially, 350 total Medicaid clients with a diagnosis of schizophrenia were identified for potential DM participation via a query of the Medicaid administrative claims database. Telephone contact was made with 155 clients (44%), of whom 126 persons (81%) ultimately enrolled. Those 29 (19%) who did not enrol at the study onset either chose not to partake in the programme, denied healthcare problems or did not qualify for the programme based on other inclusion criteria (e.g. individuals <18 years of age).

To ensure an adequate duration for the DM programme assessment, the 73 clients (58%) who completed 6 months or more of interventions were included in an empirical analysis. The 53 clients not completing at least 6 months of interventions predominantly had changes in Medicaid eligibility (n=42; 79%), with the remaining 11 (21%) involving other issues such as an unknown reason for dropout or death. Based on the aforementioned, the inclusion criterion for the DM group (as opposed to the comparator group) was the documentation of schizophrenia and continuous enrolment in the DM programme from 6 to 12 months. These criteria yielded 73 clients who received long-term interventions in the DM group. The enrolment date of the programme was defined as the study start (i.e. index date) for these clients.

Propensity score-matched comparator group

As the selection of clients to participate in the DM programme was non-random in nature, a propensity score-matching algorithm was used to select a relevant comparator group14,15. Essentially, propensity scores are bias-removing procedures that involve the conditional probabilities of assignment to a treatment versus comparator group given a set of observed covariates16. The choice of covariates is established for each investigation and is based on theoretical and practical justifications. Ultimately, a set of covariates is selected to yield a single scalar variable wherein the algorithm tends to produce unbiased estimates of treatment effects17. Thus, the propensity score methodology seeks either to reduce or to eliminate any inherent biases between non-randomly selected groups as a function of the factors included in deriving the score itself and is frequently used for pharmacoeconomic analyses14,15.

Within the current investigation, a 1:3 ratio of programme enrollees to potential matches was initially used to identify comparators for the DM group (i.e. individuals who did not enrol in the programme but had similar demographic and health-related characteristics). The propensity scores were computed using a logistic regression that included the following variables: age; gender; total baseline overall medical claim costs; total baseline overall pharmacy claim costs; and the RxRisk case-mix risk adjustor18. For baseline cost assessment, the time period was defined as 3 months prior to an initial assessment date. For comparators that did not have a date of initial assessment available, an index date of the 1st July 2002 was used to determine the end of the baseline period (i.e. coinciding with the official date that the DM programme was launched). The case-mix risk adjustment method used in this study was the RxRisk, formerly the Chronic Disease Score, which is recognised as the most extensively described pharmacy-based case-mix adjustment measure in the scientific literature18–20. The RxRisk is a validated metric that predicts future health services resource use and costs based on the severity of disease states.

Importantly, generated propensity score values have been noted occasionally to provide imperfect one-to-one matches (i.e. several potential matched comparators may have been generated for each DM client)14,15. Thus, to reduce any potential for bias in how the matches were established, eligible comparators were placed in a random order via a random number generator. After this random sort order, the propensity scores were generated and matched. Furthermore, although the rationale behind propensity scores asserts that matching should reduce or eliminate numerous forms of bias, the need to ensure that valid pairings were maintained was considered important enough to pre-empt using Medicare and Medicaid dual eligibility issues within the algorithm. Therefore, three separate propensity score matches were performed, one for each of the following groups: 1. Medicaid clients without Medicare; 2. Medicaid clients with Medicare <65 years of age; 3. Medicaid clients with Medicare ≥65 years of age. Once these clients were matched with comparators, data for all three strata were combined. Any subsequent results did not differentiate by category of payer.

Concerning the propensity score matching for dual eligibles, it is important to highlight that certain Medicaid clients may also qualify for Medicare. There are two conditions that qualify Medicaid clients to receive Medicare benefits: having a severe illness or disability; or reaching 65 years of age. Although these categories are not mutually exclusive, the current analysis included all over-65 dual-eligible Medicaid beneficiaries in the ≥65 year old category. Both groups of Medicare and Medicaid dual-eligible clients may require a higher level of care than the rest of the Medicaid population. Therefore, to incorporate the distinctions between these three groups into the DM client/comparator matching scheme and to ensure that study participants in each group of dual eligibility were matched with similar comparators, the matching algorithm divided the three groups based on dual eligibility status before the propensity score matching.

Overall, the broad selection of comparators began by identifying an initial pool of 33,419 clients diagnosed with schizophrenia from the Medicaid administrative claims database using diagnosis codes on all healthcare claims paid between the 1st July 2000 and the 31st May 2004 with ICD-9-CM codes ranging from 295.00 to 295.90 (schizophrenic disorders), or with schizophrenia and a chronic condition diagnosis code on medical claims. Beyond the propensity-score algorithm, the specific inclusion criteria for potential comparators additionally included: clients with at least 3 months of eligibility available in the baseline period; clients with at least 3 months of eligibility after the programme was started; and clients with at least one schizophrenia diagnosis code in the baseline period and at least two antipsychotic prescriptions in the baseline period, or at least two schizophrenia diagnosis codes in the baseline period. In addition, to establish consistency with DM clients, possible comparators were excluded based on organisation (ORGN) code classification if they: had medical claims for ‘Old Age Pensioners’ (i.e. ORGN code 5412); were admitted to Class I nursing facilities (i.e. ORGN code 5440); or were receiving care under a state Medicaid waiver plan (i.e. ORGN codes 8210, 8211 and 8212). These inclusion criteria provided 11,318 possible matches in the comparator group from the original pool of 33,419 from which the propensity scores algorithm was applied.

Outcome measures and statistical methods

Outcomes reported in this study were the respective overall medical and pharmacy costs (US$). Medical and pharmacy claims incurred during the baseline or study period were summed for each client, and claim costs per member per month (PMPM) were determined by dividing the medical or pharmacy costs in the baseline or intervention period by the length of time in the programme. Differences in expenditures between the DM client group and the matched comparator group were presented by the ORGN code as defined by the Colorado Department of Health Care Policy and Financing. The ORGN code results were presented for all medical claims both in the DM and comparator groups. The time frame for all analyses extended from the 1st July 2000 to the 31st May 2004. Institutional Review Board approval was granted by the Human Subjects Protection Program at the University of Arizona.

All statistical analyses were conducted using Stata 9 software (Stata Corp., College Station, TX)21. To address any skewing of the data and to meet the assumptions of parametric testing procedures, a log plus constant model was employed for total medical and pharmacy costs for each client in the DM and comparator groups22. One-sample t-tests were used to determine whether baseline costs and costs following programme implementation differed from zero. Two-sample t-tests with unequal variance were used to compare baseline medical and pharmacy costs, age and propensity scores between groups. χ² tests were used to assess whether there were significant differences in the proportion of men versus women and the proportion of White versus other races between groups.

Regression techniques were employed to determine whether there were statistically significant differences between groups while holding baseline costs constant for respective medical or pharmacy claim costs as:where a is the intercept of the model, b1 and b2 are the parameter estimates, and e is the stochastic residual. Cost = overall medical or pharmacy costs, respectively; DisMgmt = group dummy variable (0 = propensity score-matched comparator group and 1 = DM group); and BaselineCost = baseline overall medical and pharmacy costs, respectively.

Least-squares analysis was used to obtain the parameter estimates of each regression model. As all costs within the regression model were log-transformed, results yielded constant elasticities between these cost-related dependent and independent variables22. Diagnostics (e.g. heteroskedasticity, normality) were conducted on the residual term of the regression equation22. Finally, a univariate pre-post analysis of the change in overall medical and pharmacy costs for the DM group only was performed. This secondary analysis was not initially planned, but was included to give an indication of where the highest expenditures may have occurred. For all tests, an α level of 0.05 was established for statistical significance. An a priori power analysis to compare two groups with a medium effect size of 0.3, α of 0.05 and power of 0.95 yields a total sample size of 134 subjects. Similarly, a total sample size of 107 subjects is specified within a regression analysis defined by a squared effect size of 0.15 (i.e. medium effect), α of 0.05, power of 0.95 and two predictor variables23.

Results

A total of 73 (58%) of the 126 clients initially enrolled in the DM programme completed more than 6 months of interventions. The average time that these clients spent within the DM programme was 11.9 months (standard deviation (sd) 0.6 months; minimum 6 months, maximum 12 months). Following the propensity score matching based on an initial 1:3 match ratio, there were a total of 159 matched individuals included in the comparator group, representing an approximate 1:2 ratio. Table 1 summarises the descriptive characteristics for each of these groups. Findings indicated that the DM client group was statistically equivalent to the propensity score-matched comparator group with regard to age, gender, race, baseline costs, RxRisk and propensity score values. The typical DM participant was mid-40s, White and male. To better reflect any skewed distributions in cost categories, Table 2 presents the descriptive median values of PMPM costs (rather than the mean PMPM costs) for baseline and post-initiation time frames both for the DM client group and the propensity-matched comparator group. Finally, Table 3 presents the Pearson correlation matrix for the continuous variables of interest. These results indicate a strong and significant correlation of ≥ 0.50 (p≤0.01) between baseline/post-medical costs, baseline/post-pharmacy costs and post-medical/post-pharmacy costs.

Table 1. Baseline characteristics by group.

Characteristic	Disease management group (n = 73)	Propensity score-matched comparator group (n = 159)	Between-group differences (p-value)	Statistical significance
Age (mean (sd)) (years)	45.5 (9.5)	44.5 (12.0)	0.486	NS
Male gender (%)	63.0	59.1	0.317	NS
Race White (%)	73.9	67.9	0.869	NS
RxRisk (mean (sd))	5.0 (3.7)	5.4 (4.1)	0.496	NS
Baseline overall medical costs (mean PMPM (sd)) ($) *	908 (630)	1,196 (1667)	0.094	NS
Baseline overall pharmacy costs (mean PMPM (sd)) ($) *	623 (451)	521 (556)	0.624	NS
Propensity score (mean (sd)) ^†	0.149 (0.144)	0.114 (0.131)	0.079	NS

sd, standard deviation; NS, not statistically signifi cant at the ≤0.05 level; PMPM, per member per month.

* Between-group mean differences were assessed via two-sample t-test, log plus constant values.

^† Propensity score computed via a logistic regression that included age, gender, total baseline medical claim costs, total baseline pharmacy claim costs and the RxRisk case-mix risk adjustor.

Table 2. Median medical and pharmacy costs by group according to cost category.

Cost category	Disease management group		Propensity score-matched group
	Baseline	Post-intervention	Baseline	Post-intervention
Overall medical costs (median PMPM (25%, 75%)) ($)	1,050 (298, 1,263)	939 (184, 1,227)	554 (115, 1,443)	562 (105, 1,394)
Overall pharmacy costs (median PMPM (25%, 75%)) ($)	574 (277, 924)	618 (235, 900)	362 (148, 668)	338 (114, 755)

PMPM, per member per month.

Table 3. Pearson correlation matrix for continuous variables.

	Age	RxRisk	Log baseline medical costs	Log baseline pharmacy costs	Log post-medical costs	Log post-pharmacy costs
Age	1.000	0.054	-0.068	-0.149 *	-0.084	0.156
RxRisk	0.054	1.000	0.305 ^†	0.416 ^†	0.186 ^†	0.361 ^†
Log baseline medical costs	-0.068	0.305 ^†	1.000	0.443 ^†	0.755 ^†	0.378 ^†
Log baseline pharmacy costs	-0.149 *	0.416 ^†	0.443 ^†	1.000	0.335 ^†	0.815 ^†
Log post-medical costs	-0.084	0.186 ^†	0.335 ^†	0.335 ^†	1.000	0.514 ^†
Log post-pharmacy costs	-0.156 *	0.361 ^†	0.378 ^†	0.815 ^†	0.514 ^†	1.000

* Statistically significant at the 0.05 level;

^† statistically significant at the 0.01 level.

The regression analyses that controlled for differences in baseline costs indicated a significant reduction of 30% for total medical claim costs with the DM group compared with the comparator group (coefficient = -0.358, p=0.031), as presented in Table 4. However, there were no significant changes in total pharmacy costs (p=0.870). Baseline costs were observed to have a statistically significant association with post-period costs. A 1% change in baseline PMPM medical costs was associated with a 0.871% change in post-period overall medical costs (p<0.001), whereas a 1% change in baseline pharmacy costs was associated with a 0.876% change in overall pharmacy costs (p<0.001). Overall, the regression analyses explained between 57.5 and 66.1% of the variance as indicated by the adjusted R² values.

Table 4. Regression analysis of costs according to cost category.

Regression model and variables *	Coefficient	t-value	p-value	Adjusted R^2
Model 1: dependent variable, overall PMPM medical costs				0.575
Constant/intercept	0.782 ^$	2.50	0.013
Disease management group ^†	-0.358 ^$	-2.17	0.031
Baseline overall medical costs ^‡	0.871 ^¶	17.72	<0.001
Model 2: dependent variable, overall PMPM pharmacy costs				0.661
Constant/intercept	0.555 ^$	2.27	0.024
Disease management group ^†	0.031	0.16	0.870
Baseline overall pharmacy costs ^‡	0.876 ^¶	21.25	<0.001

PMPM, per member per month.

* All cost variables employed log plus constant transformations.

^† Coefficient interpreted as a percent reduction relative to the comparator group, where d is defined as the dichotomous treatment dummy variable:

^† Coefficient interpreted as a constant elasticity :%ΔY = β%ΔX.

^$ Statistically significant at the 0.05 level;

^¶ Statistically significant at the 0.01 level.

Within the DM group alone, and not controlling for relevant covariates via a regression approach, the post hoc univariate pre-post cost assessment (i.e. mean DM intervention time period costs minus mean baseline costs) did not indicate any significant change in mean PMPM costs. Cost changes in this regard, noted in Table 5, seemed to be small and ranged from -$14 (sd $633) for overall medical costs (p=0.733) to +$15 (sd $281) for overall pharmacy costs (p=0.425).

Table 5. Pre–post changes in costs among disease management clients according to cost category.

Cost category	Mean PMPM dollar change from baseline (pre) to disease management intervention (post)	Difference between zero (p-value)	Statistical significance
Overall medical costs (mean change (sd)) ($)	-14 (633)	0.733	NS
Overall pharmacy costs (mean change (sd)) ($)	15 (281)	0.425	NS

PMPM, per member per month; sd, standard deviation; NS, not statistically significant at the 0.05 level.

Beyond the cost analyses addressed earlier, the overall cost to provide the DM programme was approximately $31,250 per month throughout the time frame of this investigation, regardless of the number of clients enrolled in the programme (although ultimately equating to $937,500 over 2 years). Thus, although the 73 clients who completed 6–12 months of the programme had a reduction in total medical costs compared with the propensity score-matched comparators, the costs to provide the programme itself potentially poses a substantial barrier in achieving overall programme cost effectiveness for a small number of participants. Table 6 presents the comprehensive claims and costs data by group according to ORGN code.

Table 6. Overall costs and claims by group according to organisation (ORGN) code.

ORGN code	Organisation description	Number of claims (overall)	Medical costs (overall)	Disease management group			Propensity score-matched comparator group
				Clients	Baseline	Post	Clients	Baseline	Post
5433	Home and community-based service—mentally ill	3,358	$757,429	51	$285,032	$403,408	9	$5,164	$63,825
5437	Home and community-based service—client services	3,883	$746,673	5	$24,521	$18,350	73	$46,180	$657,623
5439	Home and community-based service—persons with AIDS	41	$2,644	0	$0	$0	1	$113	$2,531
5452	Physician services and EPSDT	9	$109	0	$0	$0	2	$0	$109
5455	Physician services and EPSDT	107	$4,906	12	$357	$943	22	$616	$2,989
5457	Lab and X-ray	30	$536	3	$108	$474	5	$66	$287
5459	Dental services	6	$126	0	$0	$0	1	$0	$126
5460	Durable medical equipment	1,909	$165,927	10	$1,357	$8,305	81	$36,415	$119,849
5461	Emergency transportation	525	$6,701	11	$2,178	$419	15	$531	$3,573
5462	County transportation	2,902	$55,194	31	$8,838	$2,895	47	$4,168	$39,293
5475	Co-insurance (Title XVIII-Medicare)	10,875	$160,359	67	$11,462	$14,338	144	$24,468	$110,092
5481	School-based CL	18	$1,044	1	$108	$58	1	$0	$879
8159	HCBS-MI, CM	899	$65,612	52	$26,994	$25,972	10	$703	$11,943
8180	INPT PSYCH DHS	11	$47,275	0	$0	$0	1	$0	$47,275
	Total of claims with organisation coding	24,573	$2,014,535		$360,956	$474,763		$118,424	$1,060,393
	TCN-ICN to GL code missing	45,030	$4,055,515	73	$434,807	$256,184	159	$2,123,097	$1,241,427
	Grand total medical cost	69,603	$6,070,050		$795,763	$730,947		$2,241,521	$2,301,820

EPSDT, early prevention, screening and diagnostic testing; CL, community living; HCBS-MI, CM, home and community-based services for the elderly, blind and disabled case management; INPT PSYCH DHS, inpatient psychiatric care, Department of Human Services; TCN-ICN to GL, transaction control number-image control number to general ledger.

Discussion

Schizophrenia is a complex disease process that is often managed by a range of interventions and monitoring. Patient needs are often complex and may require basic skills training to allow individuals with the disease to live and work in the community in order to avoid institutionalisation or homelessness5. The current investigation focused on assessing changes in direct medical and pharmacy costs for Medicaid clients with schizophrenia enrolled in a DM programme relative to matched comparators. A propensity score-matching methodology was employed to identify a comparator group that was statistically equivalent based on age, gender, total baseline overall medical and pharmacy claim costs, and the RxRisk case-mix risk adjustor.

Several findings are important to note within the current study. Foremost, less than one half of clients were able to be contacted via telephone for initial enrolment (155/350; 44%). Of the 126 clients who met all inclusion criteria and ultimately enrolled, 73 (58%) completed at least 6 successive months of interventions. Regarding the 53 who did not partake in the DM programme for that minimum time frame, 42 (79%) had changes in Medicaid eligibility that precluded them from completing and the remaining 11 (21%) dropped out for various other reasons (e.g. lack of telephone contact, death). Future research should seek to identify and positively impact factors that may be associated with programme participation and completion.

Another key finding of this study was that the DM clients had significantly lower overall medical costs of 30% relative to a matched comparator group. However, no significant differences were observed regarding overall pharmacy costs. Furthermore, pre-post costs did not change for clients only enrolled in the DM programme (i.e. costs following the DM programme minus baseline costs). Noting that the current investigation did not directly capture clinical outcomes and did not include programme costs, these findings may be viewed as DM participants were associated with lower overall medical costs compared with matched comparators.

Regarding programme costs, conducting the interventions may potentially be viewed as relatively expensive if limited to a small group of participants (i.e. $31,250 per month ultimately equating to $937,500 over 2 years). These costs may ultimately limit the overall cost effectiveness, particularly if benefits in outcomes are deferred for long periods of time and if a small number of individuals enrol. Irrespective of outcomes, it may be approximated that the costs of delivering the programme to each of the 73 individuals in the current study was almost five-fold higher (e.g. $428–535 PMPM) compared with if the programme was delivered to all of the 350 contacts that were initially eligible (e.g. $89–112 PMPM). Pragmatically, increasing the number of clients remaining enrolled within a programme would seek to capture benefits based on economies of scale. Beyond this, however, future research should also seek to assess the incremental long-term changes in clinical, economic and humanistic outcomes (e.g. costs, clinical endpoints, productivity, quality of life) associated with DM programmes across a number of perspectives (e.g. patient, provider, payer, society). It is also crucial to acknowledge that cost savings alone may not be indicative of overall cost effectiveness (i.e. ‘when the outcome is worth the cost relative to competing alternatives’)24.

As previously addressed, there are relatively few studies that have assessed the impact of coordinated care in the treatment of schizophrenia and related medical co-morbidities. Lehman reported that the PACT model (i.e. a comprehensive, community-based schizophrenia treatment and rehabilitation programme) was cost effective among individuals whose condition could not be adequately treated via typical community-based approaches6. Other studies and adaptations of the PACT model have been associated with decreases in hospitalisations of up to 40% and statistically significant decreases in inpatient lengths of stay25,26. Findings from TMAP have also yielded favourable clinical results and have subsequently provided an evidence-based framework for DM programme interventions27. In addition, TMAP has been reported to improve outcomes without increasing healthcare service costs28.

Focusing more specifically on the use of pharmaceuticals, Eaddy et al reported that clients enrolled within a Medicaid programme, who were adherent with their antipsychotic therapy, were less likely to have an inpatient hospitalisation and had lower inpatient charges than clients who were only partially adherent29. Busch et al found that implementing managed behavioural health carve-outs in Medicaid did not result in changes in medication quality indicators compared with a reference Medicaid group, although changes in psychosocial treatments were observed30. One study conducted by Wallace et al also found no difference in the use of antipsychotic medications31. These authors subsequently noted a decrease in the continuity or intensity of psychosocial treatment with managed care.

Several limitations must be considered when assessing the results of the current study. Although 73 clients completed the DM interventions, this is a relatively low proportion of the total number of clients that could have completed. These enrolment levels, partially due to challenges in making initial contact and in Medicaid eligibilities, also limit the external validity of the analysis. Thus, the conclusions regarding the costs for the 73 clients may not be generalised to all eligible Medicaid clients nor to all individuals with schizophrenia. As no method existed to investigate those who did not enrol or did not complete the DM programme, non-response error may be present wherein those that remained enrolled may have systematically differed from those who did not32. Despite being developed for individuals with schizophrenia, some interventions were broad and may require more than 12 months to impact outcomes (e.g. encouragement to eat a healthy diet, participation in an exercise programme, receipt of recommended vaccinations). It was also beyond the scope of this study to assess any clinical endpoints of schizophrenia monitoring that may have been impacted by the DM programme and, importantly, the utilisation of carve-outs was also not available for analysis. There may have been significant changes in health status that simply cannot be extrapolated by analysing cost data. In addition, client satisfaction with the DM programme was not assessed in this analysis. Also of importance in interpreting findings, a propensity score-matching method was used to reduce the bias among factors used to generate the propensity score; however, the propensity score cannot control for any bias that is not correlated with the factors used to generate it. Thus, there may be unobserved differences between the study group and comparators that were not captured in the data. Furthermore, as this investigation was retrospective and non-randomised, the quasi-experimental design could not fully control for numerous threats to internal validity33.

The interventions in this DM programme included approaches to improve co-morbidities and the overall health of adult Medicaid clients diagnosed with schizophrenia. Relative to the comparator group, clients participating in the DM programme for 6–12 months had an overall reduction in total medical costs. It is possible that the interventions, which included an exercise programme, improved communication techniques and a vaccination programme, may have created more contacts with the health system for schizophrenia-related issues. Conversely, it is also possible that these general health measures reduced the need for contacts with clinicians, leading to a reduction in total medical costs. The basic interventions to promote a healthy lifestyle may be especially important when interacting with clients with a complex and chronic disease such as schizophrenia.

Conclusion

The purpose of this study was to evaluate potential changes in medical and pharmacy costs associated with a DM programme designed for adult Medicaid clients with schizophrenia in the US State of Colorado. Care for these individuals involves several unique and important issues that must be considered both in the development and delivery of an effective DM programme. Furthermore, changes in eligibility and plan enrolment may selectively disqualify potential programme participants and pose particular challenges when implementing a DM programme among a population of Medicaid clients. Based on the current investigation among the 73 clients with schizophrenia who completed 6–12 consecutive months of interventions, significant reductions in total medical costs were observed compared with a matched reference group of individuals not enrolled in the programme. Costs to deliver the interventions, however, exceeded any corresponding cost savings and may partially be attributed to a low enrolment rate. Researchers, clinicians and healthcare decision-makers should continue to identify and impact factors that are associated with the long-term participation and cost effectiveness of these programmes.

Acknowledgements

This study was funded by a research grant provided by Eli Lilly and Company.