Does Capitation Affect the Health

of the Chronically Mentally III?

Results From a Randomized Trial
Nicole Lurie, MD, MSPH; Ira S. Moscovice, PhD; Michael Finch, PhD; Jon B. Christianson, PhD; Michael K. Popkin, MD

Objective.\p=m-\Todetermine the effect on health outcomes of enrollment of chron- cause of concerns that they were at risk
ically mentally ill Medicaid recipients in prepaid plans vs traditional fee-for-service for adverse outcomes under capitation.
Medicaid. While most literature about capitation
indicates that utilization rates are lower
Design.\p=m-\Arandomized controlled trial. Clients who were randomly assigned to than in FFS plans,45 there is limited in
prepaid care were then permitted to choose among four capitated health plans. formation on health outcomes, particu
Clients returned to fee-for-service care at the end of the demonstration.
Setting.\p=m-\TheMedicaid Demonstration Project in Hennepin County, Minnesota, larly for low-income populations or those
with mental illness. A recent evaluation
the urban center of which is Minneapolis. of two Medicaid capitation programs
Patients.\p=m-\Sevenhundred thirty-nine Medicaid clients who were classified as found no differences in prenatal care or
having chronic mental illness on the basis of Medicaid claims. Clients were inter- birth outcomes in prepaid and FFS plans.6
viewed at baseline (time 1) and at two follow-up points. Data were available for 96% Ware et al7 found a trend toward worse
of participants at the end of the intervention (time 2). Average duration of follow-up health outcomes among low-income
was 11 months. A subset of 370 clients with schizophrenia was followed up 11 groups enrolled in a single HMO com
months after the return of the prepaid group to fee-for-service care (time 3). pared with FFS enrollees, but Medicaid
Main Outcome Measures.\p=m-\Generalhealth status, physical functioning, social clients were not included in this compar
ison. Another analysis of the experience
functioning, and psychiatric symptoms, assessed using the Schedule of Affective did not find this to be the case.8
Disorders and Schizophrenia\p=m-\Changeversion, the Global Assessment Scale, and The poor, particularly those with chronic
indicators of community function.
Results.\p=m-\Nosignificant differences between prepaid and fee-for-service groups
physical or mental illness, might have a
different experience in prepaid care than
in general health or psychiatric symptoms from baseline to time 2. After regression in the FFS setting. Theoretically, capi
adjustment, 12% fewer clients in the prepaid group reported being victimized tated arrangements give providers incen
(P<.01). At the end of time 3, the regression-adjusted Global Assessment Scale tives to better manage enrollees' utiliza
scores had worsened by 7.6 points more in the prepaid group in comparison with tion, encourage them to seek services early
the fee-for-service group (P<.02). in an illness episode, and substitute low-
cost for high-cost treatment modalities
Conclusion.\p=m-\Therewas no consistent evidence of harmful effects of enrolling when appropriate. Capitation may lead to
chronically mentally ill Medicaid clients in prepaid care, at least in the short run. The better integration of mental health ser
generalizability of these findings may be limited to plans that control utilization by vices, better coordination ofmental health
methods similar to those used in this study setting. Longer-term outcome studies and non-mental health services, and to
should be undertaken to clarify the strength of the findings. improved quality of care.9
(JAMA. 1992;267:3300-3304) Several authors argue that the poor
may suffer adverse outcomes in prepaid
CARE of the chronically mentally ill no Medicaid programs have enrolled dis systems.910 Financial incentives under
(CMI) is consuming increasing state and abled individuals in health maintenance capitation could lead to underservice, a
federal financial resources.1 Because organizations (HMOs). A few programs particular risk for individuals with chronic
many CMI are poor, the cost of their have capitated mental health care for the illness. Concerns have also been raised
care is reflected in increasing costs of CMI. However, these individuals receive regarding whether low-income and CMI
public programs such as Medicaid.2 the remainder oftheir care via traditional individuals can negotiate their way
At least 20 states have considered cap reimbursement approaches. through the bureaucratic structure of
itation to control Medicaid expenses,3 par Beginning in 1986, the Health Care some HMOs to get needed care. Some
ticularly among young women and chil Financing Administration sponsored a suggest that HMOs will shift these pa
dren receiving Aid to Families With De series of demonstrations enrolling Med tients to publicly financed care or ration
pendent Children. Aside from Minnesota, icaid clients in prepaid care. Minnesota mental health services in ways inappro
was a demonstration site, and Hennepin priate for CMI patients.8 Additionally,
From the Division of Health Services Research and County, which includes Minneapolis, was substitution by psychologists and social
Policy, School of Public Health University of Minnesota its urban setting. Hennepin County was workers for psychiatrists, which is more
(Drs Lurie, Moscovice, Finch, and Christianson); the the only county that enrolled a broad common in HMOs, may heighten the need
Departments of Medicine, Hennepin County Medical
Center, and University of Minnesota (Dr Lurie); and the range of recipients, including the eld for urgent evaluation and increase the
Departments of Psychiatry and Medicine, University of
Minnesota (Dr Popkin), Minneapolis. Dr Lurie is a Henry
erly, disabled, and mentally ill, and that risk of admission and crisis intervention.11
J. Kaiser Family Foundation Faculty Scholar in General randomly assigned recipients to prepaid Finally, health plans and providers work
Internal Medicine. vs fee-for-service (FFS) care. In Min ing in prepaid settings may have little
Reprint requests to Division of Health Services
Research and Policy, School of Public Health, Univer-
nesota, enrollment began in 1987. interest or experience in treating chronic
We studied the experience of CMI mental illness and, therefore, may avoid
sity of Minnesota, 420 Delaware St SE, Box 729, Min-
neapolis, MN 55455-0392 (Dr Lurie). individuals in this demonstration be- CMI patients or treat them less aggres-

Downloaded From: http://jama.jamanetwork.com/ by a George Washington University User on 04/04/2016

 sively than would FFS providers.12 This
article compares health outcomes of CMI
individuals who received their care in pre
paid plans with those remaining in FFS
settings during the demonstration.
METHODS
Study Setting
In the Hennepin County Medicaid Dem
onstration Project,13 35% of all Medicaid
beneficiaries in Hennepin County were
randomly assigned to prepaid care. The
remaining 65% continued to use FFS pro
viders. Once randomized to prepaid care,
clients were given a choice of four health
plans. An independent broker educated
clients about the plans and encouraged
them to choose one. Clients who did not
choose a plan were randomly assigned to
one. The choice process has been previ
ously described.14 Only 17% of clients did
not voluntarily choose a plan; 15% changed
providers. The decision to choose a plan
was related to having a regular source of
mental health care and the participants'
education level. Prior health care use and
health status did not significantly affect
the decision to choose.14 Once clients were
enrolled, they were required to remain in
the plan for at least a year, unless granted
an appeal. Plans received a capitation pay
ment, targeted at 95% of estimated FFS
costs, to care for enrolled individuals. They
were required to provide all services avail
able under Medicaid. Rates were estab
lished using a rate-cell approach, which
considered age, gender, Medicare status,
and Medicaid eligibility category. Plan par
ticipation was voluntary, but plans en
rolling clients of Aid to Families With
Dependent Children were required to en
roll at least one high-risk group. Four plans
chose to enroll disabled clients, including
the CMI. These included a county-spon
sored network HMO, a Blue Cross and
Blue Shield-sponsored plan, and two
smaller independent practice association
plans. Seven months after the demonstra
tion began, Blue Cross and Blue Shield,
the plan with the most Medicaid benefi
ciaries enrolled, announced its intent to
withdraw from the demonstration. Be
cause of concerns about whether the other
plans could accommodate the large num
ber of disabled enrollees who would need
to be transferred, the state withdrew all
disabled clients from the demonstration
after the first year.14
The prepaid plans used a variety of
approaches to manage provision of men
tal health services. The largest plan used
the approach it had developed for private-
sector enrollees. This included primary-
care case management, use of a nurse
case manager, prior approval for non-
emergency admissions, and review, ap
proval, and monitoring of treatment plans.
Downloaded From: http://jama.jamanetwork.com/ by a George Washington University User on 04/04/2016
The plan's data system was used to iden Table 1.Sample Size and Response Rates
tify and address, retrospectively, patterns
of utilization that appeared excessive. Subjects Prepaid Care FFS Care*
Eligible by algorithm 506 510
The county's sponsored health plan con Refused
"
29 29
tracted with all county-funded mental Unable to locate
Other sample
9 5
health providers, but relied heavily on exclusions 99 106
prior authorization and case management. Baseline sample
369 370
It performed concurrent review for hos (time 1)t
pitalized patients and attempted to ar Refused
Unable to locate
11
2
10
2
range services so that patients could be Moved out of state 1 3
discharged, at the soonest appropriate Spoke foreign
language 1 0
time, to the care of community providers. Completed Interviews
Care of high-risk individuals was case- at time 2 354 (96%) 355 (96%)
Eligible sample time 3
managed by a psychiatric nurse. More (clients with schizo
intensive case management was used im phrenia) 196 174
Refused 10 8
mediately after hospitalization. A physi Unable to locate 0 1
cian, nurse, and case-manager team co Moved out of state 2 2
ordinated care with the primary physi Completed Interviews
at time 3 184 (93%) 163 (94%)
cian and evaluated patient progress
weekly. One of the plans relied on nurse *FFS Indicates fee-for-service.
case managers to control utilization of tTimes 1 through 3 are defined in the "Methods"
section.
high-frequency/high-cost clients; the other
relied on a capitated contract with a phy the period between client notification of
sician group to control costs. the state's intent (14 days prior to disen-
Study Sample rollment) and 2 weeks following their dis-
enrollment from the prepaid plan (time
We developed an algorithm to identify 2). Since intake dates for enrollment var
CMI individuals from Medicaid claims
ied, follow-up interview periods ranged
tapes. This algorithm is described else from 7 to 12 months; average length was
where, as are results of a validation study approximately 11 months. Follow-up data
of claims for schizophrenia.1415 We ap for the comparison group were collected
plied this algorithm to all Hennepin according to the same schedule, with in
County Medicaid clients who were eligi dividuals randomly assigned interview
ble for Medicaid because they were dis dates from 7 to 12 months after their base
abled. We identified all eligible clients line interviews.
randomized to the experimental (prepaid) In addition to the predemonstration
group and an equal random sample of and postdemonstration data collection,
clients remaining in FFS Medicaid. Rea we tracked a subsample of clients with
sons for loss from the sample are pre
schizophrenia through their transition
sented in Table 1 but were primarily that back to FFS care to look for any delayed
the clients were no longer on Medicaid or
problems or rebound effects that might
living in Hennepin County when we tried have been due to unmet need during the
to interview them. This sample size al demonstration. We collected this final
lowed us to detect a difference in Global set of data from 9 to 11 months after
Assessment Scale (GAS) scores between
study clients were withdrawn from the
groups of one third of an SD, a small to demonstration (time 3).
moderate effect size.16
Data Collection Health Status Measures
Data were collected through face-to- We used separate measures to assess
face interviews with study clients. All overall health and mental health status.
interviewers had prior experience with While all measures used were standard
CMI individuals and received training instruments validated in other studies,
in excess of the guidelines recommended some required minor modification after
by the (then) National Center for Health pilot testing. Health status was assessed
Services Research.17 along the dimensions outlined by the
We interviewed clients at baseline (time World Health Organization.18 Clients
1), the 2-week period between assignment first rated their health as excellent, good,
to prepaid plans and actual receipt of ser fair, or poor. Physical functioning was as
vices in that plan for the experimental sessed with the nine-item battery used in
group. We conducted control-group in the RAND Health Insurance Experi
terviews during a similar period. Our in ment.19 It was scored in terms of number
tent was to reinterview all clients 1 year of limitations, with a higher score indicat
later. However, because of the prema ing worse health. Social functioning was
ture withdrawal of disabled clients from measured using five items modified by
the demonstration, the interview sched Kane et al.20 Role function was assessed
ule was modified. Individuals enrolled in with two items from the RAND health
prepaid plans were reinterviewed during insurance study,19 and general-health
 Table 2.Covariates Used in the Estimation of
Regression-Adjusted Differences*
Prepaid vs fee-for-servicet
Aget
Sexf
Racet
Education
Income
in Medicare
Participation
Possession of private insurance
Number of mental health admissions, prior year
Number of physical health admissions, prior year
Number of chemical dependency admissions, prior year
Mental health outpatient visits, past 3 mot
Physical health outpatient visits, past 3 mo
Chemical dependency outpatient visits, past month
General health status (excellent-poor)t
No. of comorbid conditions
Living arrangement!
Marital statust
Physical functioning index
General health status index
Global Assessment Scale scoret
Scores on SADS-C subscalesft
No. of impairments in community function
Number of days in plant
Interaction with mental and physical health care
sions
Number of visits
*Measured at baseline.
fThese covariates were used in the analysis of sub
groups.
+.SADS-C indicates Schedule of Affective Disorders
and Schizophrenia-Change version.
perceptions were measured with a four-
item general-health scale also developed
at RAND.19 Other than physical func
tioning, all these measures are scored so
that higher scores indicate better health.
We used the Schedule of Affective
Disorders and Schizophrenia-Change
version21 to measure change in psychi
atric symptoms. This measure is brief,
can be administered by a nonpsychia-
trist, and covers a wide range of psy
chiatric symptoms. It consists of six sub-
scales: depression, anxiety, endogenous
features, mania, delusions/hallucinations,
and miscellaneous symptoms. Higher-
scale scores indicate more symptoms. It
also includes the GAS,22 a well-validated
100-point scale assessing overall com
munity function in which higher scores
indicate better function. We adapted sev
eral other measures from the work of
Lehman23 to assess community function.
These included counts of victimization
and arrests and information about liv
ing and working in sheltered settings.
Finally, we collected information regard
ing sociodemographic characteristics, ac
cess, satisfaction, and utilization of health
services.
Data Analysis
We compared the distributions of vari
ables within experimental and control
populations for the baseline and each
follow-up period using Student's t tests
and 2 techniques. Mann-Whitney U sta
tistic tests were used for ordinal vari
ables. To control for differences that
might have been present in the two
groups at baseline, we used a variety of
Downloaded From: http://jama.jamanetwork.com/ by a George Washington University User on 04/04/2016
Table 3.Baseline Demographic Characteristics of Prepaid and Fee-for-Service (FFS) Care Samples*
Variables
Age, yt
Female, %
Married, %
No. of years In Hennepin
County, Minnesota!
No. of years education!
White, %
Employed! or student, % 14.6
Monthly Income, $!
No. of chronic health conditions, GAS range,
0-18!
with % 55.9
Subjects schizophrenia,
*None of the above is
comparisons statistically significant.
!Numbers are expressed as mean (SD).
!lncludes full- and part-time employment.
regression techniques. For continuous
variables, we used ordinary least squares
models; logistic regression was used for
admis dichotomous dependent variables. In all
cases, the dependent variable was a
given health status measure during the
follow-up period. For measures that
were continuous variables, we also con
structed change measures, computing
the difference between the value of the
variable from time 1 to time 2, and, for
those with schizophrenia, from time 1 to
time 2, and from time 1 to time 3. We
then used that difference as the depen
dent variable. This approach avoids the
problems that can occur when the de
pendent variable has a high proportion
of zero values, as was sometimes the
case. Mean sample values were substi
tuted for missing values if the number
of observations for which data of a spe
cific variable were missing was less than
10%. Otherwise, observations with miss
ing data were dropped from the analy
sis. In the regression models, we con
trolled for sociodemographic character
istics, inpatient and outpatient use, gen
eral health status, mental health status,
social function, insurance, and length of
time in the plan. A fuller listing of these
variables is presented in Table 2. All
models provided consistent results. With
one exception, there were only minor
differences between the unadjusted and
any regression-adjusted results. Thus,
we focus on the unadjusted findings. Re
gression adjustment did alter estimates
of victimizations and suicide attempts
and the magnitude of GAS score changes
in clients with schizophrenia. These find
ings are discussed below.
To determine whether other sub
groups were differentially affected by
capitation, we examined the experience
of the following: individuals with any
comorbid medical condition; clients with
a specific comorbid condition (eg, dia
betes, lung disease, heart disease, ar
thritis, neurologic disease); and benefi
ciaries beginning the demonstration in
poorer mental health.
FFS
Prepaid
41.5(11.7) 41.6(12.0)
52.8 58.9
4.5 5.1
21.9(14.7) 22.9(16.0)
12.1 (2.5) 11.7(2.5)
84.3 83.9
12.5
384.2 (246.7) 404.1 (266.1)
3.6
3.6 (2.7) (2.5)
49.0
GAS indicates Global Assessment Scale.
Human Subjects
The study was approved by the hu
man subjects protection committee and
informed consent was obtained from all
participants.
RESULTS
At baseline, we interviewed 369 pre
paid and 370 FFS clients. At the end of
the demonstration period (time 2), we
reinterviewed 354 prepaid clients (96%)
and 355 FFS clients (95%). There were
196 prepaid group clients and 174 FFS
group clients with schizophrenia after
the demonstration ended, and we col
lected data at time 3 from 174 prepaid
clients (94%) and 184 FFS clients (93%).
Proxy respondents were only used to
obtain utilization information on the two
clients in each group who died. Reasons
for nonparticipation at baseline and fol
low-up appear in Table 1.
There were no significant differences
in characteristics of the prepaid and FFS
populations at baseline (Table 3). At base
line, those assigned to the prepaid group
had slightly fewer outpatient physical
health visits in the prior 3 months (2.7 vs
3.4, P=.07).
The effectiveness of cost-containment
strategies is evidenced by the utiliza
tion data. Briefly, based on unadjusted
self-report measures, fewer prepaid than
comparison group enrollees received out
patient physical health care (61% vs 71%,
P=.01). The prepaid group also aver
aged 4.6 fewer annual visits (P<.01) and
fewer inpatient admissions for physical
health problems (0.20 vs 0.39, P<.01).
Fewer prepaid enrollees received out
patient chemical dependency treatment
(4.5% vs 10.3%, P<.01) or inpatient
chemical dependency treatment (1.9%
vs 5.2%, P=.05). There was little differ
ence in outpatient mental health visits
unrelated to chemical dependency, or in
the proportion with a mental health ad
mission. However, a separate analysis
of claims data indicated a shorter length
of stay for prepaid-group clients (1.56 vs
 4.30 days, P<.01).24 Of the prepaid-group Table 4.Health Status of Prepaid and Fee-for-Service (FFS) Care Groups at Baseline and Follow-up*
clients, 17% reported being refused care Time 1 Time 2
in the prior year, in contrast with 12% I I I I
in the comparison group (P=.06). These Variables Prepaid FFS Prepaid FFS
were primarily emergency department General health, physical function, and social health
visits. Prepaid-group clients did not re Self-rated health, % fair/poor 53.0 55.7 50.0 49.0
General health status Index (4-16) 10.2 10.1 10.0 9.7
port poorer access to care on a variety
of other measures. Physical functioning index (0-9) 4.2 4.6 4.2 4.6
Table 4 presents unadjusted health sta Social contact index
(0-120)_3JL6_29J_309_29.6
tus measures at baseline and follow-up for status
Psychiatric
Mean Global Assessment Scale (0-100) 55.1 55.0 54.1 53.7
both groups. In general, enrolling Medi
caid CMI beneficiaries in prepaid plans SADS-C subscale
Depression (0-78) 24.2 23.5 25.0 25.6
did not have a detrimental effect on mea
Anxiety (0-18) 6.1 5.7 6.3 6.6
sures of their health status. This was the
Endogenous features (0-42) 11.5 11.4 12.2 12.6
case for the general and social health mea
sures as well as most measures of psychi
Mania (0-30) 6~4 6^4 6~6 TO
atric status and community function. One Delusions and hallucinations (0-48) 11.2 11.0 11.0 10.8

exception was that, at the end of the dem Miscellaneous (0-36)_9^6_05_07_10.0

onstration period, enrollees in the prepaid Community and role function, %
Arrested during previous year 4.5 4.8 4.2 4.1
plan had a 12% decrease in the regression- Jailed 2.3 4.5
adjusted likelihood of having been as during previous year 2.3 3.4

saulted, raped, or robbed (P<.01) and a Assaulted, robbed, or raped during

previous year 13.8 17.5 15.1t 12.4
7% increase in the self-reported likelihood
Suicide attempt during previous year 6.2 9.6 5.0+. 8.6
of attempting suicide (P<.05). All but one
of the 36 clients reporting a suicide at Living in sheltered setting 29.3 29.4 27,2 24.2
Withnights homeless during previous year 3.4 3.1 3.1 3.8
tempt had used some mental health ser
vices over the study period. Whether this Working in sheltered setting 13.3 11.0 13.1 10.9
occurred before or after the reported sui *None of the above comparisons is statistically significant. Ranges appear in parentheses. SADS-C indicates
cide attempt could not be determined. Schedule of Affective Disorders and Schizophrenia-Change version. Times 1 and 2 are defined in the "Methods"
section.
Health outcomes for the clinical sub
tin multivariate analysis using logistic regression, enrollees in the prepaid group had a 12% lower likelihood of
groups were similar to those for the ag victimization at time 2. P<.01.
gregate study sample. multivariate analysis using logistic regression, enrollees in the prepaid group had a 7% greater likelihood of
suicide attempts than those in the fee-for-service group at time 2. P<.05.
Data for the subgroup with schizo
phrenia were available for clients once
they returned to FFS care (Table 5).
Other than changes in the GAS scores,
there were no significant differences in Table 5.Health Status of Clients With Schizophrenia in Prepaid and Fee-for-Service (FFS) Care Groups'
health status between the groups. At at Baseline and Postdemonstration*
baseline, mean GAS scores were com Time 1 Time 3
parable for prepaid and FFS groups. At -1
time 2, regression-adjusted GAS scores Variables Prepaid FFS Prepaid FFS
had worsened 2.5 points more in pre General health, physical function, and social health
Self-rated health, % fair/poor 44.9 48.6 48.0 43.9
paid than in FFS clients (P, not signif General health status index (4-16) 10.1 9.9 9.8
icant). By time 3, the difference in this 9.6
decline had reached 7.6 points (P<.02). Physical functioning index (0-9) 3.8 4.0 4.0 4.2

Throughout this period, there were no Social contact index (0-120) 109.7 110.6 27.0 25.5
differences in client self-reported med Psychiatric status
Mean Global Assessment Scale (0-100) 52.8 52.3 44.3t 47.6
ication compliance.
SADS-C subscale
COMMENT Depression (0-78) 24.0 23.2 24.5 24.3
Anxiety (0-18) 6.0 5.6 6.1 6.0
This study examined the experience
of CMI Medicaid beneficiaries randomly Endogenous features (0-42) 11.3 11.5 12.4 12.6

assigned to prepaid plans and a com Mania (0-30) 63 6~4 65 5~9

Delusions and hallucinations 11.7 11.6 13.1
parison group remaining in FFS Med (0-48) 12.2
icaid. Using multiple measures of health Miscellaneous (0-36)_9_4_93_06_93
status, we were unable to find consis Community and role function, %
Arrested during previous year 3.5 2.0 4.5 5.0
tent evidence of worsened health status
for prepaid-group clients. However, sev Jailed during previous year 1.8 3.4 5.1 5.7
eral of the outcomes we report deserve Assaulted, robbed, or raped during
previous year 13.5 20.1 8.9 13.6
comment.
Suicide attempt during previous year 4.1 7.4 4.5 9.2
First, regression-adjusted GAS scores Living in sheltered setting 38.2 40.3 41.0 34.8
for clients with schizophrenia declined
With nights homeless during previous year 2.9 2.7 2.5 2.8
over the study period, with prepaid-plan
clients experiencing a greater decline, Working in sheltered setting 20.6 12.1 14.0 19.1

amounting to approximately one half of
the SD. This differential decline began
during the demonstration, and continued
throughout the postdemonstration pe-
*None of the above comparisons is statistically significant. Ranges appear in parentheses. SADS-C indicates
Schedule of Affective Disorders and Schizophrenia-Change version. Times 1 and 3 are defined in the "Methods"
section.
tin multivariate analysis using logistic regression, Global Assessment Scale scores for enrollees in the prepaid
group declined 7.6 points more than did those in the fee-for-service group (P<.02).

Downloaded From: http://jama.jamanetwork.com/ by a George Washington University User on 04/04/2016

 riod. However, this finding was limited to
the subgroup of clients with schizophrenia
and was not accompanied by measurable
changes in psychiatric symptoms, per
haps because the global nature ofthe GAS
reflects a broader construct than the psy
chiatric symptoms measured by the
Schedule of Affective Disorders and
Schizophrenia-Change version. While
the GAS decline may represent the nat
ural history ofdisease for some clients, the
differential between prepaid and FFS
groups suggests that some feature of
prepaid-plan enrollment may have accen
tuated this decline. For example, transi
tions to and from prepaid care may have
resulted in disruptions in care for some pa
tients. However, because most clients did
not change mental health providers, one
could postulate that a change in provider
behavior could have led to altered care or
that access to ancillary services may have
changed. We did not detect such changes
with our measures. Alternatively, pre
paid plans have had little experience deal
ing with CMI clients and their administra
tive structures may not be as responsive
or proactive as necessary to reach these
patients effectively. Finally, attempts to
limit utilization may have resulted in some
treatment limitations or in a shift from
psychiatrists to other providers who
would be less likely to use psychotropic
medications as an element of treatment.
In either case, effects may not have been
fully manifest until months later.
The results concerning victimizations
and suicide attempts are difficult to in
terpret. In view of the multiple compar
isons we made, they may be spurious find
ings. Alternatively, differences in suicide
attempts may represent behavior result
ing from difficulty experienced by prepaid-
plan enrollees accessing the system; it is
noteworthy that more prepaid enrollees
reported being refused care, primarily in
the emergency department. However,
there was no differential reporting ofother
access difficulties, and there was not an
increase in successful suicides.
Four limitations should be considered.
First, clients were followed up for a rel
atively brief period in prepaid plans be
cause the state terminated the demon
stration for the CMI. Such occurrences
are always a hazard when studying nat
ural experiments. Furthermore, this is
not atypical of other Medicaid experi
ences, in that some health plans may fail
or terminate Medicaid contracts. Given
more time in prepaid plans, differences,
either positive or negative, may have
emerged.
Second, we made multiple compari
sons between prepaid and FFS clients.
Thus, the relatively few differences be
tween the groups may have occurred on
the basis of chance alone. Because of the
Downloaded From: http://jama.jamanetwork.com/ by a George Washington University User on 04/04/2016
large numbers of comparisons and the
consistent findings, use of additional
measures probably would not have al
tered ourconclusions.
Third, clients were reinterviewed af
ter the state announced its intent to
withdraw from the demonstration. We
doubt this announcement affected client
responses since many clients were un
aware that they were to be disenrolled
from prepaid plans. Furthermore, be
cause satisfaction with prepaid and FFS
care was comparable, it seems unlikely
that prepaid clients would have under-
or overreported their health status.
Last, only 15% of clients changed pro
viders; rather, the providers practiced
under a different set of financial and
administrative restraints. This makes it
less likely that GAS score differences
resulted from disruption in client-pro
vider relationship. While the only re
ported access problem was to emergency
department services, it is possible that
access to ancillary services was more
difficult. This scenario, too, is likely to
be typical in other states that enroll Med
icaid clients in capitated plans, especially
when these plans are organized along an
independent practice association model.
It is possible that a more dramatic
change in systems of care for clients
would have produced different results.
In summary, this first randomized trial
of prepaid care for CMI Medicaid ben
eficiaries did not find consistent evidence
of short-term adverse health effects in
prepaid plan enrollees. However, the
generalizability of this finding may be
limited to settings in which prepaid plans
use similar cost-containment methods.
Given the increasing popularity of cap
itation for providers of services to the
CMI, we recommend that long-term out
come studies be undertaken to extend
our findings.
Financial support for this research was provided
by Hennepin County (Minnesota), the National In
stitute of Mental Health, the Bush Foundation, the
State of Minnesota, and Hennepin Faculty Associ
ates Young Investigator Program.
We thank the following individuals: Jean Endicott,
PhD, Anthony Lehman, MD, and Willard Manning,
PhD, for help with study measures and data interpre
tation; Colleen King for supervising data collection;
Mohammad Ahktar and Charles Ng for help with data
analysis; Ellen Benavides, MHA, for support from the
Hennepin County Office of the Medicaid Demonstra
tion Project; Terry Sarazin, MSW, and Kathleen
Schuller, MPH, from the Minnesota Department of
Human Services for facilitating interactions with
Medicaid; and Jane Raasch for preparing the manu
script. Finally, we thank an anonymous reviewer for
critical review of the manuscript.
References
1. Scarlet LJ. Paying for public mental health care:
crucial questions. Health Aff. 1990;9:7-124.
2. Taube CA, Waldman HH, Salkever D. Medicaid
coverage for mental illness: balancing access and
costs. Health Aff. 1990;9:5-18.
3. Freund D, Newschler E. Overview of Medicaid
capitation and case-management initiatives. Health
Care Finan Admin Rev. 1986(suppl):21-30.
4. Manning WG, Liebowitz A, Goldberg GA, Rog-
ers WH, Newhouse JP. A controlled trial of the
effect of prepaid group practice on use of services.
N Engl J Med. 1984;310:1505-1510.
5. Wells KB, Manning WG, Benjamin B. Use of
outpatient mental health services in HMO and fee\x=req-\
for-service: results of a randomized controlled trial.
Health Serv Res. 1986;21:453-474.
6. Carey TS, Weis K, Homer C. Prepaid vs tradi-
tional Medicaid plans: lack of effect on pregnancy
outcomes and prenatal care. Health Serv Res. 1991;
26:165-182.
7. Ware J, Ryan WH, Davies AR, et al. Compar-
ison of health outcomes at a health maintenance
organization with those offee-for-service care. Lan-
cet. 1986;1:1017-1022.
8. Sloss EM, Keeler EB, Brook RH, Operskalski
BH, Goldberg GA, Newhouse JP. Effect of a health
maintenance organization and physiologic health:
results from a randomized trial. Ann Intern Med.
1987;106:130-138.
9. Christianson J. A comparative study of public
sector financing arrangements in mental health care.
Presented at the National Institute of Mental Health
symposium on Public Sector Capitated Funding
Mechanisms in Mental Health Care; September 21\x=req-\
22, 1987; Arlington, Va.
10. Schlesinger M. On the limits of expanding health
care reform: chronic care in prepaid settings. Mil-
bank Q. 1987;64:189-215.
11. Fink DJ, Dubin WR. No free lunch: limitations
on psychiatric care in HMOs. Hosp Community
Psychiatry. 1991;42:363-365.
12. Heinen LA, Fox PD, Anderson MD. Findings
from the Medicaid competition demonstrations: a
guide for states. Health Care Finan Admin Rev.
1990;11:55-68.
13. Christianson JC, Shadle M. The organization of
mental health care delivery in HMOs. Admin Ment
Health 1988;15:201-225.
14. Moscovice I, Finch M, Lurie N. Plan choice by
the mentally ill in Medicaid prepaid health plans.
Adv Health Econ Health Serv Res. 1989;10:265-278.
15. Lurie N, Moscovice I, Finch M, Popkin M, Dysken
M. Accuracy of Medicaid claims for psychiatric diag-
noses: experience with the diagnosis of schizophrenia.
Hosp Community Psychiatry. 1992;43:69-71.
16. Cohen J. Statistical Power Analyses for the Be-
havioral Sciences. New York, NY: Academic Press
Inc; 1977:8.
17. National Center for Health Services Research.
Reducing Interviewer Effects on Health Survey
Data. Rockville, Md: US Dept of Health and Hu-
man Services; 1986.
18. World Health Organization. Constitution of the
World Health Organization. Geneva, Switzerland:
World Health Organization Basic Documents; 1948.
19. Stewart A, Ware JE, Brook RH. Construction
and Scoring of Aggregate Functional Status Mea-
sures. Santa Monica, Calif: The RAND Corp; 1982.
20. Kane RA, Kane RC, Arnold S, et al. Measuring
Social Functioning in Mental Health Studies: Con-
cepts and Instruments. Rockville, Md: National In-
stitute of Mental Health, US Dept of Health and
Human Services; 1985.
21. Endicott S, Spitzer R. A diagnostic interview:
the Schedule for Affective Disorders and Schizo-
phrenia. Arch Gen Psychiatry. 1979;36:706-712.
22. Endicott S, Spitzer RL, Fleiss JF, Cohen J. The
global assessment procedure for measure of the
overall severity of psychiatric disturbance. Arch
Gen Psychiatry. 1976;33:766-771.
23. Lehman A. The effects of psychiatric symp-
toms on quality oflife assessments among the chron-
ically mentally ill. Eval Program Plann. 1983;6:
143-151.
24. Christianson JB, Lurie N, Finch M, Moscovice
IS. Utilization and costs of physical and mental
health services in a mentally ill population: a com-
parison of Medicaid recipients in prepaid plans ver-
sus fee-for-service. Presented at the annual meet-
ing of the Association of Health Services Research;
June 18, 1990; Arlington, Va.