A Systematic Review of Shared Decision–Making Interventions for Service Users With Serious Mental Illnesses: State of the Science and Future Directions

We followed the PRISMA statement and checklist (19) for reporting guidance of the review process. To identify studies to include or consider for this systematic review, we worked with a medical librarian to develop detailed search strategies for each database. The search strategy was piloted in PubMed Legacy (National Library of Medicine) and was applied to Embase (Elsevier), Web of Science (Clarivate Analytics), PsycINFO (EbscoHost), and Applied Social Sciences Index and Abstracts (Proquest) by using a combination of keywords and subject headings. A gray-literature search included the American Psychological Association’s PsycNet. (A list of search terms is available in an online supplement to this article.) The search was limited to the English language (the primary language of the review team) and articles published since 1980, because this was around the time when the concept of SDM began to appear in the academic literature (8). The original search was completed on July 11, 2018, and was updated on April 15, 2020. The reference lists of included articles were also hand searched for other potential studies.

Two authors (E.C.T., Y.Z.-I.) screened the titles and abstracts of all identified articles to determine which full texts should be accessed and evaluated. They then cross-screened 10% of the included full-text articles to ensure consistency in the selection process, dividing up the screening of the remainder of full-text articles after a high level of agreement was achieved (>80%) and discussing all discrepancies to consensus. Articles were included in the review if they met all the following criteria: all participants were service users with serious mental illnesses, as defined by the Substance Abuse and Mental Health Services Administration (1); interventions included “elements of discussion or communication of health information between a provider and patient or caregiver, and aimed to enhance patient participation, involvement, or self-determination in decisions about the guiding or planning of treatment” (11); studies of decision support tools or DAs were included only when they were used as part of an appointment, meeting, or consultation between providers and service users or caregivers; article types included all except review papers, editorials, development papers, protocol papers, and survey studies of views, perceptions, or attitudes toward SDM; studies included quantitative or qualitative measures assessing the process or outcomes of interventions. Articles were excluded if participants, interventions, or article types did not meet these criteria.

Data extraction was performed independently by at least two authors who used a data dictionary. Characteristics of each study’s participants (e.g., country of origin and demographic characteristics), experimental intervention (e.g., intervention name and components, format, duration or frequency, type of interventionist, setting, and intended user or users [service user, other supporter, or mental health provider]), methods (e.g., study design and quality), and outcomes (e.g., constructs and time points assessed) were recorded. Classification of intervention components was informed by Zisman-Ilani and colleagues’ (11) typology, and intervention components included DAs and decision support tools that elicited shared care planning, elicited preferences, facilitated patient motivation, and involved decision coaching, decision guidance, or communication skills training. Interventions were subsequently grouped into decision support tools only, multicomponent interventions involving decision support tools, multicomponent interventions not involving decision support tools, and shared care planning and preference elicitation interventions. We followed Perestelo-Perez et al. (20) in categorizing decision-making outcomes into SDM antecedents, SDM process, and SDM outcomes; all other outcomes were grouped on the basis of patterns that emerged in the data. If primary versus secondary outcomes were specified by study authors, only primary outcomes were extracted; for studies in which this information was not provided, all outcomes were extracted. Discrepancies were discussed to reach consensus. Data were synthesized with count and frequency statistics.

Risk of bias for each study that included quantitative data was independently evaluated by pairs of researchers within the author team by using the Cochrane Collaboration Risk-of-Bias tool (21). Study quality of each study that included qualitative data was independently rated by an author pair by using the Critical Appraisal Skills Program (CASP) qualitative checklist tool (22). Mixed-methods studies were evaluated with the use of both tools. Discrepancies were resolved by discussion until consensus was reached. Given that this research evaluated only preexisting data and did not involve interaction with human subjects, it did not require ethics committee approval by our institutions.

The systematic database search resulted in 15,358 records (including 98 grey-literature records) (see PRISMA flow diagram in the online supplement). After removal of duplicate records, 11,711 eligible records were exported to Covidence (at https://www.covidence.org), the recommended systematic review platform by Cochrane Reviews. In total, 59 published records of 53 separate studies were included in this review. Characteristics of the methods, participants, and interventions of the included studies are summarized in a table in the online supplement (23–78).

Participants represented many nationalities, because studies were conducted in the United States (25 studies resulting in 29 published records) (14–16, 24, 28–32, 34, 36, 41–43, 45, 49, 51, 52, 55, 56, 62, 64, 65, 68–70, 73, 75, 78), the United Kingdom (10 studies, 12 records) (26, 33, 40, 50, 53, 54, 57, 58, 66, 67, 71, 74), Germany (six studies/records) (25, 35, 37, 59–61), the Netherlands (four studies/records) (38, 39, 46, 47), Australia (two studies/records) (44, 63), Japan (two studies/records) (27, 72), Saudi Arabia (one study/record) (23), Finland (one study/record) (77), Israel (one study/record) (48), and across multiple countries (one study/record) (76). Six studies (eight records) (27, 42, 44, 62, 70, 73, 75, 77) were conducted with young adults between the ages of 18 and 30, and 47 studies (51 records) (14–16, 23–26, 28–41, 43, 45–61, 63–69, 71, 72, 74, 76, 78) evaluated SDM interventions primarily among middle- and older-age adults with serious mental illnesses. Half of the studies included more male than female participants. Of the studies that reported information on the racial and ethnic background of participants (N=26), most included predominantly White participants. The mean percentages of participants in other racial and ethnic categories were relatively small in these studies (Black, 37%; Asian, 3%; Native American, 1%; multiracial, 3%; and Hispanic/Latinx, 16%). Psychiatric diagnoses of participants included schizophrenia spectrum disorders and other primary psychotic disorders (e.g., schizophrenia and delusional disorder) in 38 studies, affective disorders (e.g., bipolar disorder and major depression) in 36 studies, anxiety disorders (e.g., posttraumatic stress disorder) in nine studies, personality disorders (e.g., borderline personality disorder) in nine studies, and unspecified serious mental illness in three studies.

Studies explored a range of SDM interventions. Five studies were of decision support tools only, which focused on psychiatric medication (15, 23, 26), treatment options for depression (24), or questions to ask during an outpatient clinical encounter (25). Twenty-three studies described multicomponent interventions involving decision support tools. Of these, the most frequently evaluated intervention (in eight studies) was CommonGround, a computerized decision support center staffed by peer specialists and intended to be used in preparation for psychiatric medication consultation meetings (14, 28, 30–32, 34, 43, 45). Other interventions in this category also focused on decisions related to psychiatric medications (35, 36, 40, 42), psychiatric rehabilitation services (48), smoking cessation (29), or selection of mental health treatment options in primary or outpatient psychiatric care settings (27, 33, 37–39, 41, 44, 46, 47). Seventeen studies (21 records) were of multicomponent interventions not involving decision support tools (16, 49–68). Most commonly, interventions in this category were designed to elicit service users’ preferences for future mental health treatment, including joint crisis planning and facilitated psychiatric advance directives (53–58, 66, 67). Finally, 10 studies were of interventions focused exclusively on shared care planning and service user preference elicitation (69–78). These interventions did not include decision support tools or other SDM components, such as coaching or guidance. For example, two of these studies were of Open Dialogue, an approach to engage young adults with early psychosis in SDM with treatment providers and other supporters 73, 77).

Most interventions were delivered in a face-to-face format, with many also including either paper or electronic materials; one intervention was delivered by telephone only (61). When the intended duration of the intervention was reported (in 44 studies), it ranged from a single session to up to 3 years. Most commonly, interventions were delivered by mental health providers (e.g., psychiatrists or therapists), sometimes in concert with a peer specialist, and in some cases they were delivered by a trained research assistant or primary care provider. Most interventions were implemented in outpatient settings, six were delivered in inpatient settings, and five within primary care. Interventions were intended to be used by service users and mental health providers in 30 studies (57%); service users only in 12 studies (23%); service users, mental health providers, and other supporters in five studies (9%); service users and other providers (e.g., pharmacists and primary care physicians) in three studies (6%); and service users and other supporters (e.g., family members) in three studies (6%).

Twenty-six studies (29 records) (49%) were RCTs, 17 (32%) were quasi-experimental studies, and five (9%) were naturalistic studies. Eighteen studies (20 records) (34%) were qualitative or had a qualitative component. Sample sizes ranged from 12 to 3,379 participants.

Risk-of-bias ratings of quantitative studies and study quality ratings of qualitative studies were calculated (see online supplement). For quantitative studies, the “allocation concealment,” “blinding of participants and personnel,” and “blinding of outcome assessment” items of the Cochrane Collaboration Risk-of-Bias tool (21) received the highest percentage of high-risk ratings (53%, 27 records; 88%, 45 records; and 53%, 27 records, respectively), and the “selective reporting” item received the lowest percentage of high-risk ratings (4%, N=2). “Other bias” was noted in 8% (N=4) of studies, for reasons including selection bias, internal validity concerns, and implementation issues. For qualitative studies, the greatest percentage of studies (50%, N=10) failed to satisfy the following item of CASP qualitative checklist tool: “has the relationship between researcher and participants been adequately considered” (22). However, we noted that most of the qualitative studies (≥80%, at least 16 studies) provided a clear statement of the research, justified the use of qualitative methods, used an appropriate design to achieve the study aims, used an appropriate recruitment strategy, appropriately attended to ethical issues, and provided a clear statement of findings.

Nine studies collected data at a single time point (e.g., after exposure to the intervention), 21 used a pre-post design or otherwise collected data at two time points, 18 included follow-up assessments ranging from 4 weeks to 5 years after exposure to the intervention, and the remaining five studies used a data collection procedure that was ongoing throughout the study period (Table 1).

Outcome characteristics of quantitative studies that compared differences between experimental and control groups (if multigroup) or that examined change over time (if single group) are summarized in Table 1. Although process and outcome measures were variable across studies, we describe patterns in the findings across intervention types.

Decision support tools were associated with positive findings related to SDM outcomes (i.e., decisional conflict [15]) and treatment engagement or adherence (23). Evidence was mixed about their impact on the SDM process (i.e., treatment satisfaction [23, 25] or perceived involvement in decision making [24]) and other outcomes (i.e., beliefs about medication [23], length of clinical encounter [25], and other feasibility outcomes [24]). Studies did not detect differences between experimental and control groups in terms of SDM antecedents (i.e., participation preferences or decision self-efficacy [25, 26]) or mental health or functional outcomes (i.e., depression symptom severity or quality of life [23]).

Evidence was limited regarding the impact of multicomponent interventions involving decision support tools on SDM antecedents (i.e., service users’ decision-making preferences [40, 43]), except for one study that found a favorable effect on decision self-efficacy in the experimental group (48). In terms of SDM process, five studies reported a positive impact on service user involvement in decision making (37, 40, 43, 44, 48), and one study failed to find an effect of the intervention on patient-centered communication (28). Mixed findings also pertained to SDM outcomes (i.e., decisional conflict [38–40, 44], perceived effectiveness of the decision-making process [27], satisfaction with the decision [41], and knowledge about treatment options [48]), treatment engagement or adherence (31, 35, 37, 41, 42, 45, 48), mental health outcomes (i.e., symptoms [35, 37, 41, 43, 48]), and other outcomes (i.e., smoking cessation outcomes [29], psychiatrists’ adherence to clinical practice guidelines [36, 42], length of clinical encounter [37], attitudes toward medication, cost-effectiveness [40], side effects [42], and service user activation [43]). Negative findings pertained to service user–provider relationships (40) and global functioning (35).

Qualitative studies reported favorable attitudes toward CommonGround among both mental health providers and service users (14, 30, 34). Another qualitative study noted favorable attitudes toward and comfort engaging in SDM among service users participating in depression treatment (33).

Three studies found positive effects associated with multicomponent interventions not involving decision support tools on functional outcomes (i.e., global functioning [52, 63], residential and employment status [49], and quality of life [52]). There was mixed evidence about their impact on SDM antecedents (i.e., decision-making competence [55], decision-making preferences [59, 68], and decision self-efficacy [59]). Mixed findings also pertained to SDM process (i.e., treatment satisfaction [52, 56, 59, 64, 68] and responsibility for decision-making [59]), SDM outcomes (i.e., decisional conflict [64], decision-making skills and knowledge [68], and knowledge about mental health [65]), mental health (i.e., psychiatric symptoms [16, 52, 59, 65, 68], perceived recovery and mental health [16, 63, 68], and hospitalizations [53, 61, 66, 67]), and other outcomes (i.e., treatment costs [52], family involvement in treatment [16], attitudes toward medication, health locus of control [59], self-management [63], stigma beliefs [65], and implementation outcomes [68]). Three studies failed to find an effect of these interventions on treatment engagement or adherence (59, 60, 67). One study did not detect significant differences in service user–provider relationships (59).

Qualitative analysis of the content of advance directives and joint crisis plans revealed how service users may use these tools to disclose crisis symptoms, request respectful and compassionate treatment, and express preferences for medication, hospitalization, and medical care (54, 56, 57). Two qualitative studies identified barriers to implementation of joint crisis planning and collaborative care from the perspective of providers and service users (50, 58).

One study found that shared care planning and preference elicitation interventions were associated with improved SDM outcomes (i.e., knowledge of care plan [78]). Evidence was mixed about the impact of these interventions on SDM process (i.e., perceived autonomy support [70, 74] and treatment satisfaction [72, 78]), mental health (i.e., psychiatric symptoms [73, 77] and hospitalizations [77]), and functional outcomes (i.e., level of functioning [73], quality of life [76], and employment status [77]). One study of this intervention type found no significant differences in SDM antecedents (i.e., decision self-efficacy [73]). Another study found no differences in other outcomes (i.e., duration of untreated psychosis [77]).

A qualitative study of service users’ experiences with early intervention in psychosis services reported that a focus on shared care planning and preference elicitation, especially regarding medication, was considered to facilitate engagement and adherence (75). Another qualitative study of SDM on an inpatient psychiatric unit generally supported feasibility of implementation (71).

This review provides a comprehensive account of the state of the science related to SDM interventions for service users with serious mental illnesses. It expands on findings from previous reviews and meta-analyses by describing participant, intervention, and methodological characteristics across studies and illuminating the range of outcomes assessed and reported.

Study samples were relatively homogeneous. As we assessed on the basis of the available data, most studies were conducted with middle-age, male, and White individuals from Western countries. Disproportionately fewer studies were conducted with young adults. It should be noted that many studies, especially those conducted outside the United States, did not report race and ethnicity data, precluding the ability to draw conclusions about the potential role of these factors on outcomes of SDM interventions among service users with serious mental illnesses. This is important, because problems with provider bias, literacy, and provider mistrust are particularly pronounced among individuals from racial and ethnic minority backgrounds in other service user populations, which may limit the degree to which these individuals can engage in SDM (79–82).

Consistent with the review by Zisman-Ilani and colleagues (11), our review found that a variety of SDM interventions have been tested among service users with serious mental illnesses. With the exception of studies of CommonGround and joint crisis planning, few interventions were the focus of more than one study. Many interventions focused specifically on medication-related decisions, with some exceptions targeting other decisions (e.g., goal setting, treatment planning, smoking cessation, and family involvement in care). Most interventions were delivered in a face-to-face format by mental health providers in outpatient settings. Peer specialists cofacilitated the decision-making process in a subset of studies, most often by assisting individuals with using digital decision support tools and providing educational and motivational support. Intervention duration was highly variable, with decision support tools and joint crisis planning and advance directives having the shortest duration and CommonGround having the longest. Most interventions were designed to support SDM between mental health providers and service users.

Our review points to an established and maturing literature on SDM interventions for service users with serious mental illnesses. Approximately half of the quantitative studies were RCTs with sufficient sample sizes, and many qualitative studies fulfilled a large proportion of quality appraisal criteria. However, methodological limitations were noted. More than half of the studies collected data at a single time point or used a pre-post design, limiting the ability to determine longer-term impacts of SDM interventions on outcomes. Issues with blinding, selection bias, internal validity, and implementation were also noted. Furthermore, many studies were lacking sufficient detail about methodology, making quality appraisal more challenging. This was especially true regarding outcome reporting of quantitative studies and data analysis procedures of qualitative studies. These findings call for the development of guidelines for reporting SDM intervention studies for this population.

As noted in the review by Perestelo-Perez and colleagues (20) of measurement of SDM interventions in mental health, outcome constructs and measures were highly variable across the studies reviewed here. Commonly assessed were involvement in decision making (most often measured subjectively according to service users’ perspectives), decisional conflict, service users’ satisfaction with care planning processes or treatment, psychiatric symptoms, and medication or treatment adherence. Other outcomes included quality of life, functioning, therapeutic relationships, psychiatric hospitalizations, and implementation outcomes.

It is no surprise that given this diversity of outcomes and the range of interventions evaluated, findings across studies were mixed. However, an examination of patterns in findings across the studies points to possible benefits associated with specific types of interventions. For example, consistent with findings by Zisman-Ilani et al. (11), this review found that decision support tools yielded positive effects only on treatment engagement or adherence in a single RCT and on SDM outcomes (i.e., decisional conflict) in another. Studies of multicomponent interventions involving decision support tools consistently showed positive impacts on service user involvement in decision making; most were quasi-experimental in nature. Also similar to the findings of Zisman-Ilani et al. (11), this review found that multicomponent interventions that did not involve decision support tools yielded positive effects on functional outcomes across various study designs, with many studies also reporting favorable effects for SDM antecedent, process, and outcome variables (e.g., decision-making competence and preferences, treatment satisfaction, and decisional conflict). Finally, preference elicitation and shared care planning interventions gave positive effects only on SDM outcomes (i.e., knowledge) in a single RCT; findings were mixed in other outcome domains. In accordance with Stovell et al. (18), no intervention types clearly yielded benefits regarding service user–provider relationships. One possible explanation for these mixed findings is that only some of these interventions improved service user–provider communication, and thus they had limited impact on later health outcomes. Of course, these findings may also be attributable to methodological factors (e.g., variability in measurement tools, study design, and sample characteristics), rather than to intervention effectiveness, and should be interpreted with caution. Future comparative effectiveness research and meta-analytic studies might further examine which SDM interventions work best in relation to these outcomes.

Several limitations to this review merit discussion. First, we did not contact study authors to determine whether additional articles should be included. Furthermore, studies of person-centered interventions that were not characterized with terms such as “SDM,” “decision aids,” or “decision support” may not have been identified by our search. Therefore, it is possible that relevant articles were missed. However, the comprehensiveness of the search strategy bolstered our confidence that key studies were identified. Second, because many interventions were multicomponent treatments, it was not possible to isolate the effect of specific components on outcomes. Future dismantling studies may be especially useful for this purpose. Finally, although the comprehensiveness of this review allowed for inclusion of multiple study designs and may be considered a strength, drawing conclusions across controlled and noncontrolled trials requires careful consideration of variability in methodological rigor. Furthermore, because the heterogeneity of measures, settings, and sample characteristics precluded use of meta-analysis on the full data set (83), the purpose of this review was to provide a descriptive account of the SDM literature and not to synthesize data for analysis. Consequently, judgments about effectiveness were based solely on the detection of statistically significant differences in outcomes and did not account for effect size. We urge caution in the interpretation of the reported positive and negative findings, and we encourage researchers to subject subsets of similar studies from this review to meta-analysis.