june-cjp

IN REVIEW

Economic Impacts of Assertive Community Treatment: A Review of the Literature

Eric A Latimer, PhD¹

Background: Assertive community treatment (ACT) is an extensively studied and widely imitated community support treatment model for severely mentally ill individuals. Several previous reviews have documented its favourable effects on clients and their families. This is the first review to focus on economic outcomes.

Methods: Nineteen randomized studies and 15 nonrandomized studies describing ACT programs were identified based on 2 criteria: 1) provision of services primarily in the community and 2) shared caseloads. Percentage reduction in hospital days was calculated for the 34 study sites where reported data allowed it. Multiple-regression methods were used to relate reduction in hospital days to program fidelity and other contextual factors. The impacts of ACT on emergency-room use, use of outpatient services, housing, costs, and other economic outcomes were also examined.

Results: Higher-fidelity programs appear to reduce hospital days by about 23 percentage points more than lower-fidelity programs (95% CI = –41.2, –5.2). The estimated regression coefficients imply that a high-fidelity program reduces hospitalizations by about 58% over 1 year if the alternative involves some type of case management and by 78% if it does not. ACT appears to increase the proportion of clients who live in independent housing situations, but the effect on use of supervised housing, and therefore on housing costs, is ambiguous. The effects on use of most other resources are inconsistent across studies. Overall, ACT appears to result in somewhat lower costs, whatever the perspective of analysis adopted.

Conclusions: The most reliable cost offset to ACT treatment costs appears to be reduced hospital use. Using Quebec costs, an ACT program must enroll people with prior hospital use of about 50 days yearly, on average, to break even. As care systems evolve to reduce their reliance on hospitalization as a care modality with or without ACT, this threshold will become increasingly difficult to achieve. The primary justification for implementing ACT services will then become their clinical benefits.

(Can J Psychiatry 1999;44:443–454)

Key Words: assertive community treatment, economic outcomes, community supports, service costs

Background

Assertive community treatment (ACT) is one of the most extensively studied and widely imitated community support treatment models for severely mentally ill individuals (1–5). Briefly, it implies a multidisciplinary team providing intensive care and support to a group comprised mainly of individuals with psychosis in their own living environments (5). ACT differs from intensive case management (ICM) in that staff share the team’s caseload (6,7). Numerous randomized as well as nonrandomized trials indicate that, compared with outpatient clinic-based aftercare or low-intensity case management, ACT reduces hospitalizations while improving symptom, quality-of-life, and satisfaction outcomes (4,8). There now exist more than 400 ACT teams in the United States (US) (9) and a growing number in Canada.

Some recent work has de-emphasized the distinction between ICM and ACT (7). There exist, however, numerous clinical and organizational reasons to favour the shared caseload approach of ACT over the individual approach of ICM: greater continuity of care, improved ability to respond to crises, reduced staff burnout, and improved job satisfaction (10,11). This review will, accordingly, focus on economic issues associated with the ACT model.

Several studies have included some measurement of costs. They have generally reported lower costs for ACT, primarily through a significant reduction in hospital days. There are, however, numerous inconsistencies in both methods and results. This article reviews the existing literature on the costs of ACT and offers some interpretations of the findings.

Although several reviews of ACT studies have been published, few to date have focused on the economic impact of ACT. McGrew, Bond, and others, measuring fidelity at 18 programs based on the Thresholds Bridge program using a 14-item measure, found that programs showing the highest fidelity achieved the greatest reductions in days hospitalized. Grouping the 14 items into 3 subscales, they found that organization and staffing were significantly correlated with reduction in days hospitalized, whereas staffing was not (12). Scott and Dixon, in their review of the ACT literature, concluded that ACT was more effective at reducing hospitalizations than was “assertive outreach” (programs based on the Thresholds Bridge program, which do not include a psychiatrist on staff, have smaller teams, and offer less extended coverage) (8). In a recent, comprehensive review of the literature on ACT and ICM, Mueser, Bond, and others argue that, although most studies have found cost savings with ACT, there is no inherent reason for community-based care to be cheaper than hospital-based care. The relative costs of hospital-based and community-based care, they argue, reflect historical priorities. They also point out that providing effective community care may increase costs if access to care was limited in the past. Finally, they argue that models emphasizing rehabilitation may take longer to achieve cost savings than do models such as ACT, with a particular focus on preventing rehospitalization (7).

The present review differs from previous ones in that: 1) its focus is exclusively on economic impacts; 2) a more up-to-date set of ACT studies is included in the review; 3) explicit attention is paid to contextual factors in assessing the economic impact of ACT programs; and 4) explicit consideration is given to several outcomes other than hospitalizations and costs, including use of emergency services, outpatient visits, and housing.

Methods

Potentially relevant studies were identified by the author through literature search, identification of references noted in literature reviews, and communication with experts. Relevant studies were then selected from the pool of potentially relevant ones based on indications in published descriptions of the use of a shared caseload approach with an emphasis on providing services in the community, 2 hallmarks of ACT (7,13). Mueser and others’ classification of studies as ACT or ICM (7) was considered, but in some cases the author’s application of these 2 criteria led to the opposite classification. Thus, programs that appear to be significantly office-based (14,15) or that do not employ the shared-caseload model (for example, 16–19) were excluded. The concept of shared caseload, however, was interpreted broadly: involvement of team medical personnel in care for clients, together with a single case manager, with team meetings to discuss treatment plans, was classified as following the shared-caseload model (as in 20,21). Studies that provided too little information for the treatment program to be characterized as ACT or ICM according to this study’s criteria were excluded (for example, 22). Mueser and others (7) cite 4 unpublished studies, which they classified as ACT and which the author was not successful in obtaining, that were not included in this review (23–26). Studies reporting no economic outcomes were excluded (27). Finally, following Mueser and others (7), studies that described programs requiring participation of a family member were also excluded (28,29).

Studies differ in terms of several important factors likely to affect differences in service use and costs. Five in particular are noted: 1) study design (larger effects expected in pre–post than in randomized designs due to regression to the mean in pre–post designs) (Note 1); 2) nature of program offered to the experimental group (more intensive, high-fidelity programs likely to have greater program costs but may achieve greater reductions in use of other services); 3) nature of services offered to control or comparison group (the more the control or comparison condition resembles the experimental one, the smaller the expected economic effects); 4) study population characteristics (study subjects selected for high service use more likely to experience large reductions in service use); and 5) duration of follow-up (hospitalization rates might change over time as clients adjust to the experimental treatment, or as treatment itself evolves—they have been observed to increase [20] as well as decrease [6]). These differences imply that any estimated combined effect size, obtained using usual metaanalytic methods, would be an average of estimates of different true effect sizes. Such an average, reflecting as it would the particular mix of studies that were included in the analysis, would not be especially meaningful. The analytic strategy, therefore, is to seek evidence of systematic patterns in the relation between economic effects and the 5 factors noted.

To implement this strategy, several measures were extracted from reports for each study site: 1) the percentage difference or reduction in hospital days; 2) the direction and statistical significance of differences in use of various resources, such as the emergency room or the outpatient clinic; 3) a 2-level variable representing study design (randomized or pre–post) (Note 2); 4) a 3-level variable indicating degree of fidelity to the ACT model; 5) a 2-level variable indicating whether the control or comparison service included low-intensity case management or not (the alternative being traditional “passive” regular aftercare from an outpatient clinic); 6) a 2-level variable indicating whether the study population was selected on the basis of high hospital use; and 7) duration of follow-up.

It was not possible to reliably assess program fidelity retrospectively using the Dartmouth ACT fidelity scale (31), which requires considerably more data than are ever reported or than could reliably be reported by authors years after the fact. A simpler method for assessing fidelity had to be devised. Experimental programs were coded as being of high fidelity if, in addition to following a shared-caseload model and providing the majority of services in the community, they explicitly met at least 4 of the following 5 criteria: 1) staff:client ratio of 1:12 or better, 2) a psychiatrist on staff, 3) at least 1 nurse on staff, 4) at least some coverage outside of normal working hours, and 5) at least 2 team meetings every week (missing information on just 1 characteristic did not prevent being classified as high fidelity). These characteristics were selected on the basis of commonly shared views about important ingredients of ACT (12,31–33). Programs that explicitly met 3 or 4 of the criteria were classified as being of medium fidelity; programs that explicitly met only 2 or fewer criteria were classified as of low fidelity. Fidelity for programs for which information could not be assessed from published reports on 3 or more of the criteria was considered to be missing.

Linear multiple-regression analysis was used to assess the relationship between reduction in hospital days and the above-noted independent variables. An insufficient number of studies reported resource use or cost effects on any other variable for multiple regression methods to be used. The synthesis of these results is narrative in character.

Results

Tables 1 and 2 describe the studies included in this review based on the criteria enumerated above. A total of 19 randomized and 15 nonrandomized (14 pre–post and 3 quasi-experimental) studies were identified.

Table 1. Main features of randomized design studies
Study and location	Population studied	%^a	Program	Groups (n)	Duration
Stein and Test (1); Weisbrod and others (34). Madison, WI	Presenting at emergency room (ER)	50	ACT	ACT (65); SAC (65)	1 year
Mulder (35); Mowbray and others (36). Kent County, MI	Previous hospitalizations	79	ACT	ACT (59); SAC (62)	30 months
Hoult and others (37). Sydney, Australia	Presenting at ER	50	ACT	ACT (60); SAC (60)	1 year
Bond and others (38). 3 sites, Indiana	Previous hospitalizations	75	AO	AO (84); CM (83)	6 months
Bond and others (39). Chicago, IL	Previous hospitalizations	67	AO	AO (45); DIC (43)	1 year
Bush and others (40). Atlanta, GA	Previous hospitalizations	86	AO	AO (14); SAC (14)	1 year
Test and others (41); Test and others (42). Madison, WI	18–30 years of age; < 12 months institutionalized	100	ACT	ACT (75); SAC (47)	2 years
Morse and others (43); St Louis, MO	No stable housing	30	AO	ACT (52); SAC (64), DIC (62)	1 year
Marks and others (21); Knapp and others (44). London, England	Presenting at ER	49	ACT	ACT (92); SAC (97)	20 months
Rosenheck and others (45). 10 sites, US Veterans Administration	Previous hospitalizations	50	More or less faithful ACT adaptations	ACT (454); SAC (419)	2 years
Åberg-Wistedt and others (46). Stockholm, Sweden	Recent admissions or outpatients	88	Coordinated inpatient and outpatient ACT-like teams^b	ACT (20); SAC (20)	2 years
Quinlivan and others (47). San Diego, CA	Previous hospitalizations	68	AO	AO (30); CM (30) No CM (30)	2 years
Solomon and Draine (48). Philadelphia, PA	No stable housing; criminal record	84	ACT	ACT (60); CTJ (60); SAC (80)	1 year
Chandler and others (49,50). 2 sites, California	Functional impairment and on public assistance	61	ACT with capitation funding	ACT (217); SAC (222)	1 year
LaFave and others (51). Brockville, ON	High service users	57	ACT	ACT (24); SAC (41)	25 months
Morse and others (52). St Louis, MO	No stable housing	66	AO	AO, AO + P, SAC (total 165)	18 months
Lehman and others (53). Baltimore, MD	Previous hospitalizations; no stable housing	58	ACT with staff for clients and relatives	ACT (77); SAC (75)	1 year
Drake and others (54). 7 sites, New Hampshire	Comorbid substance abuse	76	ACT with integrated substance abuse treatment	ACT (109); CM (114)	3 years
Essock and Kontos (55); Essock and others (56). 3 sites, Connecticut	Previous hospitalizations	67	ACT	ACT (131); CM (131)	18 months

^aPercentage with schizophrenia or schizoaffective disorders.
^bProgram involving a hospital team providing treatment evenings and weekends and an outpatient clinic team with a psychiatrist that provides treatment during regular office hours. Once every 2 weeks the client meets with the entire team; a designated team member spends 4 hours weekly with the client.
ACT = assertive community treatment; AO = assertive outreach (“Bridge”-type adaptation, no psychiatrist on staff); AO + P = AO with team comprising paraprofessionnel community workers; CM = case management (less intensive community treatment with 25 clients or more per staff worker); CTJ = community treatment by workers prepared to deal with clients who have been in jail; DIC = drop-in centre, a community resource for leisure and social activities; SAC = standard aftercare by outpatient clinic with no community treatment.

Table 2. Main features of nonrandomized design studies
Study and location	Population studied	%^a	Program	Design	Groups (n)	Duration
Witheridge and others (57); Bond (58). Chicago, IL	Mostly previous hospitalizations	ns	AO	PP	ACT (50)	1 year
Bond and others (59). Chicago, IL	Previous hospitalizations	ns	AO	PP	ACT (30)	ns
Borland and others (20). Spokane, WA	Previous hospitalizations	99	ACT	PP	ACT (72)	5 years
Wright and others (60). Seattle, WA	Previous hospitalizations	75	ACT adaptation	PP	ACT (196)	4 years
Bond and others (61). 3 sites, Indiana	Young substance abusers with previous hospitalizations	70	AO	QE	ACT (31); EG (23); SAC (43)	18 months
Bond and others (6). Philadelphia, PA	Previous hospitalizations	64	AO	QE	ACT (30); CM (10)	2 years
Teesson and Hambridge (62); Sydney, Australia	Previous hospitalizations	85	AO	PP	AO (27)	6 months
Santos and others (63). Rural South Carolina	Previous hospitalizations or persistent symptoms	74	Rural adaptation of ACT with psychiatrist + 2 nurses	PP	ACT (23)	4–26 months
Santos and others (64). Charleston, SC	Previous hospitalizations	100	ACT	PP	ACT (52)	1 year
Dincin and others (65). Chicago, IL	Previous hospitalizations	ns	AO	PP	AO (66)	1 year; 3 years
Dharwadkar (66). Dandenong, Australia	Previous hospitalizations	ns	ACT adaptation	PP	ACT (50)	1 year
Hambridge and Rosen (67). Sydney, Australia	Previous hospitalizations	83	AO	PP	AO (50)	1 year
Sands and Cnaan (30). Philadelphia, PA	Previous hospitalizations	90	ACT	QE	ACT (30); ICM (30)	1–12 months
McGrew and others (68). 6 sites, Indiana	Previous hospitalizations	65	Rural adaptations of ACT	PP	ACT (212)	18 months
Meisler and others (69). Wilmington, DE	No stable housing; substance abuse	ns	ACT with integrated substance abuse treatment	PP	ACT (114)	12–48 months

^aPercentage with schizophrenia or schizoaffective disorders.
ACT = assertive community treatment; AO = assertive outreach (“Bridge”-type adaptation, no psychiatrist on staff); CM = case management (less intensive community treatment, with 25 clients or more per staff worker); EG = educational groups; ICM = intensive case management; ns = not specified; PP = pre–post (before–after comparison); QE = quasi-experimental (comparison group without random assignment); SAC = standard aftercare by outpatient clinic with no community treatment.

Effects of ACT on Time Spent in Hospital

The most consistent effect of ACT is the reduction of time spent in hospital (7,8,70,71). Where data in the identified studies allowed, the percentage reduction in hospital days was calculated (Note 3). Three sites from multisite studies were excluded, which, according to study authors, had not implemented the model: site B (Fort Wayne) of the Bond and others 1988 study (38) and sites GMS-2 and GMS-5 of the Rosenheck and others 1995 study (45) of US Veterans Administration (VA) sites. This yielded 34 sites (Table 3).

Table 3. Percentage change in hospital days, contextual and programmatic factors at identified ACT sites
		Contextual factors		Programmatic factors
Study	Percentage change in hospital days (%)	Received case management services	High prior hospital use	< 1:12^a	Psychiatrist on team	Nurse on team	2 or more team meetings per week	Coverage outside normal hours	Fidelity
Stein and Test (1)	–83.3	N	N	Y	Y	Y	Y	Y	High
Mulder (82); Mowbray and others (36)^b	–84.3	N	N	Y	N	Y	ns	Y	Medium
Hoult and others (37)	–82.9^c	N	N	Y	Y	Y	ns	Y	High
Bond and others—site A (38)	–70.5	Y	Y	Y	N	Y	Y	Y	High
Bond and others—site C (38)	–81.8	Y	Y	Y	N	Y	Y	N	Medium
Bond and others (39)	–49.9	N	Y	Y	N	Y	Y	N	Medium
Bush and others (40)	–34.1	N	Y	ns	Y	ns	ns	ns	Not rated
Test and others (41); Test and others (42)	–84.7	Y	N	Y	Y	Y	Y	Y	High
Marks and others (21)	–82.7	N	N	Y	Y	Y	ns	Y	High
Rosenheck and others—NP-1 (45)	–10.4	N	Y	Y	Y	Y	ns	ns	Medium
Rosenheck and others—NP-2 (45)	–24.2	N	Y	Y	Y	Y	ns	ns	Medium
Rosenheck and others—NP-3 (45)	–34.9	N	Y	Y	Y	Y	ns	ns	Medium
Rosenheck and others—NP-4 (45)	–59.3	N	Y	N	Y	Y	ns	ns	Low
Rosenheck and others—GMS-1 (45)	–13.1	N	Y	Y	Y	Y	ns	ns	Medium
Rosenheck and others—GMS-3 (45)	–27.7	N	Y	Y	Y	Y	ns	ns	Medium
Rosenheck and others—GMS-4 (45)	–15.1	N	Y	Y	Y	Y	ns	ns	Medium
Rosenheck and others—GMS-6 (45)	–22.0	N	Y	Y	Y	Y	ns	ns	Medium
Quinlivan and others (47)	–80.3	N	Y	Y	N	N	Y	ns	Low
LaFave and others (51)	–84.8	Y	Y	ns	Y	Y	ns	Y	Medium
Lehman and others (53)	–44.1	Y	N	Y	Y	Y	ns	Y	High
Essock and others (56)	–16.8	Y	Y	Y	Y	Y	ns	ns	Medium
Witheridge and others (57); Bond and others (56)	–58.0	N	Y	Y	N	Y	Y	N	Medium
Bond and others (59)	–61.3	N	Y	Y	N	Y	Y	N	Medium
Borland and others (20)	–82.9	N	Y	Y	Y	Y	ns	ns	Medium
Wright and others (60)	–77.4	ns	Y	Y	Y	ns	ns	ns	Not rated
Bond and others (61)	–45.9	Y	Y	Y	N	ns	ns	ns	Not rated
Teesson and Hambridge (62)	+29.9	Y	Y	Y	N	Y	Y	N	Medium
Santos and others (63)	–78.8	N	Y	Y	Y	Y	Y	Y	High
Santos and others (64)	–94.2	N	Y	Y	Y	Y	ns	N	Medium
Dincin and others (65)	–65.4	N	Y	Y	N	N	Y	N	Low
Dharwadkar (66)	–80.2	N	Y	Y	N	Y	ns	N	Medium
Hambridge and Rosen (67)	–62.4	Y	Y	Y	N	Y	Y	N	Medium
Sands and Cnaan (30)	–54.7	Y	Y	Y	ns	ns	Y	Y	Medium
McGrew and others (68)	–63.9	Y	Y	Y	ns	ns	ns	ns	Medium^d

GMS = general medical site; NP = neuropsychiatric; ns = not specified.^aRatio of staff to clients is 1:12 or better.^bThe original study is by Mulder; ratings are based on the summary in Mowbray and others (36).
^cApproximate percentage due to imprecision in data reporting.
^dThe medium fidelity rating for McGrew and others (68) is based on a secondary source (McGrew and others [12]).

To test the hypothesis that greater fidelity is associated with greater reductions in hospital days, controlling for contextual factors, the percentage reduction in hospitalizations was initially regressed on 5 factors: 1) study design, represented by a dummy variable indicating that the design was randomized as opposed to pre–post; 2) the comparison condition, represented by 1 binary variable indicating whether comparison condition included low-intensity case management; 3) duration of follow-up; 4) whether high hospital use was a criterion for admission into the study; and, finally, 5) program fidelity. Linear multiple regression was used, weighting for the size of the experimental group.

Table 3 describes data used in the regression in addition to length of follow-up, which is in Tables 1 and 2. The results of this analysis are shown in Table 4 (Model 1). Despite the relatively few observations (30, once missing values are taken into account), the parameter estimates are jointly significant (P < 0.02) and explain 48% of the variance in percentage reduction in hospital days. Aside from the intercept, however, only 1 predictor is statistically significant: randomized studies appear to be associated with more modest reductions in hospital days, by about 32 percentage points.

Table 4. Estimated models to predict percentage reduction in hospital days (standard error in parentheses)
	Model 1	Model 2	Model 3	Model 4
N	30	30	30	30
Model F (P value)	3.48^b	5.35^c	7.55^d	10.1^d
R²	0.48	0.63	0.74	0.72
Adjusted R²	0.34	0.51	0.64	0.65
Intercept	–107.8 (30.9)^c	–57.1 (31.4)^a	–82.1 (28.1)^c	–61.4 (12.5)^d
Randomized design	32.4 (11.3)^c	18.2 (10.8)	21.3 (9.3)^b	19.9 (8.6)^b
Veterans administration site	—	47.9 (15.8)^c	37.7 (13.9)^b	41.9 (11.2)^d
Teesson study	—	—	81.0 (26.8)^c	78.7 (25.9)^c
Length of follow-up	–0.2 (0.1)	–1.8 (1.0)^a	–0.9 (0.9)	–1.1 (0.8)
Comparison with case management	7.5 (9.1)	21.0 (9.0)^b	16.1 (7.9)^a	19.4 (7.1)^b
High hospital use	27.0 (17.3)	–3.4 (17.9)	7.0 (15.7)	—
High fidelity	–2.8 (22.8)	–15.5 (20.0)	–7.1 (17.3)	–23.2 (8.7)^b
Medium fidelity	17.4 (15.5)	12.9 (13.4)	13.4 (11.4)	—

^aP < 0.10; ^bP < 0.05; ^cP < 0.01; ^dP < 0.001.

Examination of the data, however, reveals that observed reductions in hospital days are generally lower at the sites in the VA hospital study (Table 3). Clearly these sites, all issued from the same study, all VA hospitals, form a group apart. This could have been anticipated: because of the way they are financed, VA hospitals have fewer financial incentives than do most other US hospitals to reduce use of inpatient beds and may even have disincentives to do so (personal communication, Robin Clark, March 1999). If we take this into account by introducing a separate dummy variable for this group of sites, the results change dramatically (Table 4, Model 2). The parameter estimates are now jointly much more significant (P < 0.002) and explain 63% of the variance. Randomized studies still appear to be associated with more modest reductions in hospital days but only by about 18 percentage points; average reductions at VA hospital sites, compared with those in pre–post studies, are lower by 48 points on average, holding other factors fixed. As expected, studies in which the comparison group receives some form of case management obtain a lower percentage reduction in hospital days, by 21 percentage points. Duration of follow-up is significant at the 10% level, with each additional month of follow-up associated with a further reduction in average hospital days of 1.8 percentage points.

Further examination of the data, however, indicates that the result on follow-up is largely attributable to a single observation, the Teesson and Hambridge study (62). This study obtained a 30% increase in hospital days (which the authors attribute to 4 individuals having been placed in long-term care) over a follow-up duration of only 6 months—almost the only study in the data set to have such a short follow-up period (Tables 1 and 2). If this study is, like the VA studies, treated as a case apart, follow-up duration is no longer a significant predictor, although it remains negative, with a coefficient of –0.9 (Table 4, Model 3).

In these 3 specifications, the coefficient on the high-fidelity variable is negative, as hypothesized; it is not statistically significant, however. Grouping together the low- and medium-fidelity categories (only 3 sites were classified as low fidelity), the high-fidelity variable almost reaches statistical significance, with a coefficient of –20 percentage points (results not shown).

The positive and nonsignificant coefficient on admitting individuals with high previous hospital use can be explained by the high negative correlation in the data between a program admitting individuals with a record of previous hospitalizations and a program achieving a high fidelity score (r = –0.712). (This correlation is in part explained by the fact that several early faithful replications of the Madison model also replicated its process of admitting into the program from the emergency room rather than identifying ahead of time high hospital users.) Thus, if this variable is removed, high fidelity becomes significant (P < 0.02), with a coefficient of –23.2 (95% CI = –41.2, –5.2). Conversely, if fidelity is removed, having selected individuals with a record of high hospital use becomes significant and remains positive (P < 0.04, with a coefficient of 23.4, results not shown). Given the likelihood that individuals presenting at the emergency room of a psychiatric hospital would in many cases also be high previous hospital users and the theoretical implausibility of a positive coefficient on the high hospital use variable, the prior hospital use variable appears redundant, and the model excluding high prior hospital use appears to be better specified (Table 4, Model 4).

The estimated coefficients from this final specification imply that a high-fidelity program reduces hospitalizations by 58% over 1 year if the alternative involves some type of case management and by 78% if it does not (Note 4). Both of those percentages are reduced by 23 points, to 35% and 55%, if the program is not of high fidelity. The cost impact of such a reduction will be taken up below. Also noteworthy is the near significance, in Model 4, of follow-up duration (P < 0.15). The data thus point to a trend toward reductions in hospital days increasing with time, at least up to about 2 years (too few studies have follow-ups longer than that to allow conclusions concerning longer follow-up periods).

Effects on Consumption of Resources Other Than Hospitalizations and on Costs

Table 5 summarizes the effects of ACT programs on client housing and housing costs. Eight of the 9 studies evaluating these effects, excluding 2 that target a homeless population, report an increase in independent living, usually to a statistically significant extent. However, the effects of ACT on use of supervised housing appear ambiguous. It appears that ACT can also help some clients with no previous stable living arrangement to live in supervised housing. Thus the net impact of ACT in terms of housing costs is unclear. It may depend on the proportion of the target population that is capable, over the period of observation, to move into independent housing.

Table 5. Effects of ACT on clients’ housing situation
Study	Effects on independent versus supervised housing	Cost impact (per client per year)
Stein and Test (1); Weisbrod and others (34); Test and Stein (72)	Higher proportion of ACT clients in independent housing (P < 0.05) and lower proportion in supervised housing facilities (not significant).	Experimental: $0 Control: $43, P < 0.05 (“Other costs [including supervised residences]”)
Mulder (35); Mowbray and others (36)	At 30 months, 6% of ACT clients in supervised housing, compared to 34% of controls (P < 0.001); 36% living alone versus 26% of the controls (not tested).	Not reported
Bond and others (39)	Slightly higher proportion in independent housing and in supervised housing.	Not reported
Test and others (41,42)	ACT clients spend more time in independent housing (P < 0.05), less in supervised housing (P < 0.05), less with (older generational) family members (P < 0.05), less being homeless. Over months 7–24 of trial 53.7% of control patients spent the majority of their time in high-supervision settings, while 73.6% of ACT patients were in low-supervision settings, primarily independent apartments.	Not reported
Morse and others (43)	In primarily homeless population, ACT yielded greater reduction in days homeless at 12 months.	Not reported
Marks and others (21); Knapp and others (44)	Not reported.	Experimental: £5698 Control: £5398, not tested (total accommodation costs over first 20 months, annualized)
Chandler and others (49,50)	Authors report “strong and consistent” finding of comparatively greater independent living at both ACT sites.	Long Beach Site Experimental: $54 Control: $133, not tested Stanislaus Site Experimental: $873 Control: $641, not tested (“Residential care”)
Lehman and others (53)	In primarily homeless population, greater use of community housing (210 days for ACT client versus 160 for control subjects, P < 0.01), fewer days on streets (10 versus 24), slightly fewer in shelters (83 versus 89).	Not reported
LaFave and others (51)	After 12 months, 50% of ACT clients in independent housing and 50% in supervised housing; 45% of control group in hospital, 20% in independent housing, 35% in supervised housing (P < 0.001).	Not reported
Borland and others (20)	Increase in use of supervised housing (from 37.4 days per person per year at baseline to 106.9 over the next 5 years, P < 0.001). 24-hour service allowed many to return to residences from which they had been evicted.	Before: $835 After: $2078, P < 0.02 (“Structured residential care”)
Santos and others (64)	The number of patients living independently increased 3.4 fold, from 11 to 37 (out of 51), and only 1 patient moved to a more dependent arrangement (no statistical tests reported).	Not reported
McGrew and others (68)	Nonsignificant (46% to 50%) increase in proportion of clients in the ACT group living in their own apartment.	Not reported

Table 6 shows ACT’s effects on the consumption of resources other than hospitalizations and housing. Even though in theory one would expect that ACT services would reduce emergency-room use, only 2 studies actually report a statistically significant reduction, though the overall trend appears to be in that direction. Also, an ACT team should in principle reduce the use of outpatient services, since those should be provided directly by the team. This is indeed observed with the original ACT team in Madison but not in several other studies. It should be noted, however, that the studies that report increases in use of outpatient services are mostly adaptations of the Madison model rather than high-fidelity replications. Results for other types of resources are generally inconsistent. The less frequent recourse to family physicians observed by Knapp and others (44) is explained by the authors as due to the increased use of ACT nursing services. The extent to which clients use community-based resources, such as food banks or community kitchens, has not been extensively measured.

Table 6. Effects of ACT on use of emergency room (ER), outpatient clinics (OC), day programs (DP), crisis-intervention services (CS), general practitioners (GP), substance abuse treatment (ST), proportion of clients receiving social assistance (SA) , employment income (EI), justice services (J), and use of other community-based resources (CR)
Study	ER	OC	DP	CS	GP	ST	SA	EI	J	CR
Stein and Test (1), Weisbrod and others (34), Test and Stein (72)	– =	–					–	+	– =	– =
Mulder (35); Mowbray and others (36)						=			=
Hoult and others (37)							=		=
Bond and others (38)^a	+=	– =	– =			+=	+=	– =	+=
Bond and others (39)		– =							–
Bush and others (40)	=
Test and others (41,42)									– =
Morse and others (43)		+				+
Marks and others (21); Knapp and others (44)		–	–		– =
Rosenheck and others (45)		+=
Åberg-Wistedt and others (46)	– =
Quinlivan and others (47)	– =	+	– =	+=
Solomon and Draine (48)									+=
Chandler and others (50)		±	– =			– =	±		– =
Lehman and others (53)	–	+				+			– =