Introduction
The use of mental health services is escalating. Utilization and spending rates for mental health care services among commercially insured adults in the USA increased by 38.8% and 53.7%, respectively, from 2019 to 2022 (Cantor et al., Reference Cantor, McBain, Ho, Bravata and Whaley2023). At the same time, various indicators show that mental health is deteriorating at a population level. The Australian National Study of Mental Health and Wellbeing (Australian Bureau of Statistics, 2022) revealed that the prevalence of operationally defined mental disorders in 16–24-year-olds rose from 26% in 2007 to 40% in 2021. Globally, 27% of responders across 64 countries in 2022 were classified as distressed or struggling with mental health, representing a decline of 11% between 2019–2021, with no recovery towards pre-pandemic levels evident (Sapien Labs, 2023).
While potential solutions are varied and complex, one approach seeks to identify ways in which we can optimize our existing evidence-based interventions. For example, personalizing treatment to the individual (e.g., tailoring treatment to a unique symptom profile measured at baseline, or using machine learning to identify patient subgroups) demonstrates better outcomes (small but significant) relative to a “one-size-fits-all” approach (Nye, Delgadillo, & Barkham, Reference Nye, Delgadillo and Barkham2023). Adding further significant benefit to the use of a purely clinician-tailored treatment is the incorporation of some degree of client choice regarding therapy (Andersson et al., Reference Andersson, Kall, Juhlin, Wahlstrom, de Fine Licht, Fardeman and Berg2023). Understanding the contribution of different optimization approaches and how best to apply these has the potential to translate to a sizeable increase in the number of clients meeting remission after treatment, thus improving the cost-effectiveness of mental health treatments.
While robust meta-analyses have emerged pertaining to different optimization approaches (e.g., de Jong et al., Reference de Jong, Conijn, Gallagher, Reshetnikova, Heij and Lutz2021; Nye, Delgadillo, & Barkham, Reference Nye, Delgadillo and Barkham2023), outcomes related to the utilization of client treatment choice are less certain. To date, four systematic reviews with meta-analyses have investigated the impact of patient preference (Table 1). Two included both medical and psychological conditions (Delevry & Le, Reference Delevry and Le2019; Lindhiem, Bennett, Trentacosta, & McLear, Reference Lindhiem, Bennett, Trentacosta and McLear2014), and two focused on mental health (Swift, Callahan, Cooper, & Parkin, Reference Swift, Callahan, Cooper and Parkin2018; Windle et al., Reference Windle, Tee, Sabitova, Jovanovic, Priebe and Carr2020). The latter two studies showed a benefit of patient preference on reduced dropout rates (OR 1.79; RR 0.62 respectively). One demonstrated superior clinical outcomes (n = 53; d = 0.28; Swift, Callahan, Cooper, & Parkin, Reference Swift, Callahan, Cooper and Parkin2018) moderated by diagnosis (depression/anxiety > psychosis/behavioral/substance use disorders), but one found no effect on depression, anxiety, or global outcomes (n = 29; Windle et al., Reference Windle, Tee, Sabitova, Jovanovic, Priebe and Carr2020).
Overview of systematic reviews and meta-analyses investigating patient preference
a Of these nine mental health studies, three were DRPT design, six were FRPT; eight trials compared psychotherapy to pharmacotherapy, one compared psychological approaches.
b This group excluded from results as contributed to substantial heterogeneity.
c One included study recruited from age 12 in a trial comparing medical approaches (M age of the study = 23.7 years).
These mixed results are likely explained by the inclusion of three differing preference trial designs, as described by Delevry and Le (Reference Delevry and Le2019). The first are Partially Randomised Preference Trials (PRPTs) where those who decline randomization choose their preferred treatment and continue in a trial. While this design retains more participants, it does not allow an unbiased measure of the impact of treatment choice. Second, a Fully Randomised Preference Trial (FRPT) assesses patient preference at baseline, but all participants proceed to randomization, allowing comparison of those who are matched versus mismatched to their preferred treatment. However, mismatch may be biased by “resentful demoralization.” Third, and by contrast, a Doubly Randomised Preference Trial (DRPT) randomizes participants to a choice vs no-choice arm, before further randomizing to Treatment A or B within the no-choice arm, and is thus the only design allowing true estimation of the impact of active choice. The FRPT design is more common given it is easily accommodated in a standard Randomised Controlled Trial (RCT) through the addition of a simple preference question at baseline. Delevry and Le (Reference Delevry and Le2019) were the first to exclude PRPT designs in the most recent of the four meta-analyses in a mixed sample across mental health (n = 9) and pain/functional disorders (n = 14), showing client preference matching conferred superior clinical outcomes (d = 0.18). The effect was stronger for mental health treatments (d = 0.23) and when limiting studies to more stringent DRPT designs (d = 0.27).
Given only nine studies of mental health preference were included in the most stringent meta-analysis, we searched the literature for more recent studies. In particular, we ensured our search terms could detect studies of eating disorders, a notable omission in the extant meta-analyses given these are a common, costly public health concern among youth and young adults, affecting one in six females and one in 40 males (Silén et al., Reference Silén, Sipilä, Raevuori, Mustelin, Marttunen, Kaprio and Keski-Rahkonen2020). Only 35.4% of clients completing treatment for bulimia nervosa experience remission (Linardon & Wade, Reference Linardon and Wade2018), so understanding how client choice might enhance outcomes is of vital importance. We also wished to search for preference studies in youth, as none to date have been included in meta-analyses. Swift, Callahan, Cooper, and Parkin (Reference Swift, Callahan, Cooper and Parkin2018) described the difficulty of including children and adolescents in preference trials, where it is more difficult to determine whose preference is driving treatment decisions. However, choice may be particularly powerful in adolescent samples, given the developmental push for autonomy that emerges at this age (e.g., Johnson, Taylor, Dray, & Dunning, Reference Johnson, Taylor, Dray and Dunning2024).
The primary aim of the current study was therefore to conduct an updated meta-analysis of the impact of patient choice on treatment outcomes across three common and frequently comorbid mental health disorders: anxiety, depression, and eating disorders. We included any age group to determine if studies have emerged in youth. At the systematic review stage, we maintained a broad net (i.e., allowing any study design) to identify emerging research on eating disorders and/or in youth, neither of which have been reported in existing preference reviews. For meta-analyses, we followed the recommendations of Delevry and Le (Reference Delevry and Le2019) to focus on robust preference trial designs (i.e., excluding PRPTs). Finally, we also sought to conduct the first meta-analysis of treatment uptake.
Method
Search strategy and selection criteria
This review was conducted according to the PRISMA guidelines (Page, McKenzie, et al., Reference Page, McKenzie, Bossuyt, Boutron, Hoffmann, Mulrow and Moher2021; Page, Moher, et al., Reference Page, Moher, Bossuyt, Boutron, Hoffmann, Mulrow and McKenzie2021). The following search string was used: ((choice or prefer*) and (treatment or interv* or trial or therapy or self-help or program* or e-health) and random*) in the title only and (depress* or anxi* or eating* or bulimi* or anorexi* or psychia*) in the abstract.
Inclusion criteria were as follows: (1) Peer-reviewed (published) journal articles; (2) RCTs where at least one treatment arm offered patients a choice of approach for eating disorders, depression, and/or anxiety; (3) outpatient setting; (4) quantitative mental health outcomes reported for patient preference-matched or choice versus preference mismatched or no choice groups; (5) any age range; (6) any publication year; (7) any language. Studies were excluded if they met the following criteria (1) Reviews, meta-analyses, or protocols; (2) secondary analyses with no additional extractable data related to patient preference. Studies that met the first exclusion criterion were tagged for reference list searching, however.
The study protocol was registered on the Open Science Framework (https://doi.org/10.17605/OSF.IO/SNJXR) on April 8, 2024, and searches of three electronic databases Medline, PsycINFO, and Scopus, were conducted on April 11, 2024. An updated search was performed on December 5, 2024, including the addition of grey literature via the database ProQuest dissertations & theses global, to more fully explore publication bias. Results of the search were loaded into Covidence software where duplicates were removed, and then the title and abstracts were double-screened by two independent reviewers (CJ and MR) before the same authors independently examined the full text of the surviving articles. Inter-rater reliability was very good at both stages (Cohen’s Kappa = .84, .93 respectively). Conflicts were resolved by discussion between reviewers.
Data extraction
One of two authors (CJ, MR) extracted the following information for each included study: author, year, country, sample size, participant demographics (mean age, gender, ethnicity, and socioeconomic status), mental health disorder targeted (eating disorder, depression or anxiety), study design (partial, fully or doubly randomized preference trial), intervention arm comparators (e.g., antidepressant medication versus psychotherapy) and whether the strength of preference was assessed. The following data were extracted for meta-analyses: mean and standard deviation of mental health outcomes at baseline and end of treatment in choice/matched preference versus no-choice/non-matched preference groups to calculate effect sizes (Cohen’s d). Frequency/percentage of uptake (i.e., participants that attended the initial session) and dropout rates (i.e., participants who did not complete treatment, or did not complete a sufficient dose if defined a priori by study) were also extracted for both groups to calculate odds ratios. Where raw data was not available for these syntheses, we used an online effect size calculator (https://www.campbellcollaboration.org/calculator/) to convert reported results; where insufficient data were reported we emailed authors for raw data if papers were < 15 years old; where no data were extractable through any approach described, we report narrative descriptors of preference effect from study text. Where more than one outcome was reported for a mental health construct (e.g., depression), the following sequence was applied: the primary outcome was extracted if nominated (if two were nominated both were extracted given multilevel approach adjusts for a unit of analysis error); then a clinician-reported outcome if available; then self-report outcome most relevant to target condition (e.g., social anxiety rather than generalized anxiety for a study targeting social anxiety disorder). If a study reported two different outcomes (e.g., targeted anxiety but also measured depression) both were extracted. Where low study numbers precluded meta-analyses, we extracted descriptors of preference effects from study text for narrative syntheses. Where these were not reported but data were provided, we manually calculated outcome (Cohen’s d) or engagement effects (Odds ratios) for single studies to include a result (i.e., choice or preference did or did not improve outcome) in our descriptive syntheses.
Quality assessment
Quality assessment utilized a subset of items from the 25-item Consolidated Standards of Reporting Trials (CONSORT) checklist (Schulz et al., Reference Schulz, Altman and Moher2010). Ten criteria of most relevance to included studies (i.e., RCTs) were selected by the study team as follows: Item 4a: eligibility criteria for participants/recruitment method; Item 4b: settings and locations where the data were collected/care providers described; Item 5: Description of interventions with sufficient detail to allow replication; Item 6a: use of validated scales with properties reported and primary outcome(s) nominated; Item 7: how sample size was determined; Item 8a: method used to generate randomization reported; Item 12a: statistical methods described in sufficient detail, with treatment effects including 95% confidence intervals plus reporting of ITT; Item 13a: the numbers of participants who were assessed for eligibility, randomly assigned, received intended treatment and were analyzed for the primary outcome; Item 13b: reasons for losses and exclusions; Item 15: a table showing baseline demographic and clinical characteristics. Items were scored ‘Y’ when completely conforming with the criteria, ‘N’ when not fully conforming with the criteria, and ‘P’ when partially conforming to the CONSORT criteria. Overall, studies were rated as high quality if they fully conformed with 8 of 10 criteria.
Meta-analyses
All meta-analyses were conducted with the R statistical software program, using the metafor package. Following Cochrane’s recommendations (Higgins & Green, Reference Higgins and Green2011) we required each outcome to have a minimum of 10 studies to proceed with a rigorous meta-analysis. Meta-analyses were conducted by comparing outcomes for participants offered choice or matched to preference to those who were randomized or non-matched to preference. For each group separately, we first calculated the mean gain (such that a positive score indicated improvement) and standard deviation of the difference. Between-group effect sizes were then precalculated from these scores using the Campbell Collaboration effect size calculator (https://www.campbellcollaboration.org/calculator/). Where a study reported preference data separately by treatment arm (e.g., match vs mismatch for antidepressant therapy, and for psychological therapy) both were extracted and entered into the analysis separately; we used a multi-level approach to account for non-independence of effects (Harrer, Cuijpers, Furukawa, & Ebert, Reference Harrer, Cuijpers, Furukawa and Ebert2021). Binary outcomes (i.e., odds ratios for uptake and dropout) were manually converted to log odds ratios for the multilevel analyses and exponentiated back to odds ratios for ease of interpretation in tables and forest plots (Harrer, Cuijpers, Furukawa, & Ebert, Reference Harrer, Cuijpers, Furukawa and Ebert2021). PRPT designs were not included in meta-analyses. Meta-analyses were planned with both double and fully randomized trials for maximum power, and then just with the former design as a more stringent test of choice effect. Heterogeneity was assessed with a combination of the Q and I2 statistics, and Egger’s regression test was used as one indicator of publication bias (Egger, Smith, Schneider, & Minder, Reference Egger, Smith, Schneider and Minder1997), entered as a moderator in our multilevel models.
Results
Description of included studies
The search yielded 311 studies; after removing duplicates, 163 remained (Figure 1). No unpublished studies were identified. Title and abstract screening excluded 105 studies, with 58 proceeding to full text screening where a further 27 were excluded. A full list of exclusion reasons appears in Supplementary Table S1.
PRISMA diagram of study selection process.
The final sample for the systematic review consisted of 31 papers reporting 30 studies across 7055 participants (M age 42.5 years, SD 11.7; 69.5% female; Table 2). Of these, one secondary paper contributed an additional outcome measure. Most studies targeted depression (n = 23; 76.7%), with anxiety represented in 5 studies (16.7%) including its variants panic disorder, social anxiety, and obsessive-compulsive disorder. Only two studies (6.7%) targeted eating disorders (binge eating disorder and prodromal anorexia nervosa). Most frequently, the choice offered to patients was between antidepressant medication and psychotherapy (n = 13; 43.3%) or between two types of psychotherapy (n = 11; 36.7%). Most studies (n = 15; 50.0%) were conducted in the USA, followed by Europe and the UK (n = 5; 16.7% each). Smaller numbers of studies were from Scandinavia (n = 2; 0.07%), Australia, Canada, and Iran (n = 1; 0.03% each).
Study characteristics
Abbreviations: ADM = antidepressant medication; BBQ = Brunnsviken Brief Quality of Life Scale; BDI=Beck Depression Inventory; DRPT = Doubly Randomised Preference Trial; DRPT = Doubly Randomised Preference Trial; EQ-5D = EuroQoL 5 Dimension scale; FRPT = Fully Randomised Preference Trial; FRPT = Fully Randomised Preference Trial; GAD-7 = Generalised Anxiety Disorder scale; HARS=Hamilton Anxiety Rating Scale; HDRS=Hamilton Depression Rating Scale; HSC-90 = Hopkins Symptoms Checklist; MADRS = Montgomery-Asberg Depression Rating Scale; PDSS=Panic disorder severity scale; PHQ = Patient Health Questionnaire; PRPT = Partially Randomised Preference Trial; PRPT = Partially Randomised Preference Trial; QIDS = Quick Inventory of Depression Symptomatology; QLDS = Quality of Life in Depression Scale; ROMIS=Patient-Reported Outcomes Measurement Information System; SAS=Social Anxiety Scale; SDS=Sheehan Disability Scale; SWLS=Satisfaction with Life Scale; YBOCS=Yale-Brown Obsessive Compulsive Scale.
a Denotes included in meta-analyses.
By preference study design, in decreasing order of rigor, 9 studies (30.0%) were doubly randomized (DRPT), 15 studies (50.0%) were fully randomized (FRPT), and 6 (20.0%) were partially randomized (PRPT). PRPT designs appear in Table 2, but were not included in our narrative or meta-analytic syntheses.
Meta-analyses
Mental health outcomes. Follow-up data were reported in only 50% of studies, so post-intervention data were utilized. There were insufficient numbers of studies to allow meta-analyses for two of three mental health outcomes (anxiety and eating disorders) and insufficient DRPT designs for depression to run a more stringent analysis utilizing this design alone. One meta-analysis was therefore conducted, for depression, using a combined sample of DRPT and FRPT designs (Table 3), and including one study that targeted anxiety but also measured depression. This analysis showed a small significant effect (Cohen’s d = 0.17; 95% CI .08, .27; I2 = 13.55) such that depression was lower for those who chose or were matched to their preferred treatment. The Forest plot appears in Supplementary Figure 1. Egger’s test suggested no evidence of publication bias; we also found no unpublished studies in our search of the grey literature.
Meta-analysis results for depression and dropout rates, by preference group
a Effect size = Cohen’s d.
b Effect size = Odds Ratio.
c Mixed group comprises depression = 13 arms/12 studies, anxiety = 7 arms /4 studies; eating disorders = 2 arms /1 study.
Dropout and uptake rates. There were insufficient numbers of studies to conduct a quantitative synthesis for uptake rates, or for dropout with DRPT designs alone; we ran the following two meta-analyses using a combined sample of DRPT and FRPT designs. First, dropout rates across a mixed group of target disorders for maximum power, and second, dropout rates for depression alone (Table 3).
Treatment choice had a significant effect on dropout rates, in both samples. In the larger combined sample (targeting depression, n = 12; anxiety, n = 4 and eating disorders, n = 1 study), participants who were not offered a choice, or were not matched to their preferred treatment, were 1.46 times (95% CI 1.17,1.83; I2 = 9.69%) more likely to drop out. In the subsample of 12 studies targeting depression alone, those randomized to treatment were 1.65 times (95% CI 1.05,2.59; I2 = 47.68%) more likely to drop out compared to those allowed to choose their treatment approach. However, we note the wider confidence interval and greater heterogeneity in the smaller sample. Combined with the absence of unpublished studies, Eggers’s test added no evidence of publication bias for either sample. Forest plots for dropout rates appear in Supplementary Figures S2–S3.
Studies not included in meta-analyses
Mental health outcomes. Depression. Data were not able to be extracted for four studies (three FRPT, one DRPT) targeting depression to include in the above meta-analyses. Three found no benefit to preference matching (Gum et al., Reference Gum, Areán, Hunkeler, Tang, Katon, Hitchcock and Unützer2006, N = 1602; Kwan, Dimidjian, & Rizvi, Reference Kwan, Dimidjian and Rizvi2010; N = 106) or choice (Hegerl et al., Reference Hegerl, Hautzinger, Mergl, Kohnen, Schutze, Scheunemann and Henkel2010, DRPT, N = 368). Moradveisi, Huibers, Renner, and Arntz (Reference Moradveisi, Huibers, Renner and Arntz2014), N = 100) found preference scores predicted dropout from pharmacological but not psychological (behavioral activation) treatment; preference scores also positively impacted the initial response within the behavioral activation arm. Considering DRPT studies (N = 7 studies) as a separate group, three found a benefit for choice (Andersson et al., Reference Andersson, Kall, Juhlin, Wahlstrom, de Fine Licht, Fardeman and Berg2023, N = 197; Callaghan, Khalil, Morres, & Carter, Reference Callaghan, Khalil, Morres and Carter2011, N = 38; Mergl et al., Reference Mergl, Henkel, Allgaier, Kramer, Hautzinger, Kohnen and Hegerl2011, N = 117) while four studies found no effect (Danhauer et al., Reference Danhauer, Miller, Divers, Anderson, Hargis and Brenes2022, N = 500; Raue et al., Reference Raue, Schulberg, Heo, Klimstra and Bruce2009, N = 60; Rokke, Tomhave, & Jocic, Reference Rokke, Tomhave and Jocic1999, N = 40; Svensson et al., Reference Svensson, Nilsson, Perrin, Johansson, Viborg, Falkenstrom and Sandell2021, N = 221). No pattern of the type of choice (e.g., the choice between two types of psychotherapy, or between medication and psychotherapy) driving benefit or lack thereof was evident between the groups.
Anxiety. In the three studies that utilized a DRPT design (Andersson et al., Reference Andersson, Kall, Juhlin, Wahlstrom, de Fine Licht, Fardeman and Berg2023, N = 197; Brenes et al., Reference Brenes, Divers, Miller, Anderson, Hargis and Danhauer2020, N = 500; Svensson et al., Reference Svensson, Nilsson, Perrin, Johansson, Viborg, Falkenstrom and Sandell2021, N = 221), no benefit of offering choice was found. Of three FRPT studies, two found no impact of preference matching (Dunlop et al., Reference Dunlop, Kelley, Mletzko, Velasquez, Craighead and Mayberg2012, N = 80; Koszycki, Ilton, Dowell, & Bradwejn, Reference Koszycki, Ilton, Dowell and Bradwejn2022; N = 97) while one (Wheaton et al., Reference Wheaton, Carpenter, Kalanthroff, Foa and Simpson2016; N = 80) found the effect differed by condition: matching did not affect outcomes for participants who received exposure therapy augmentation of antidepressant medication but did confer a benefit for those who preferred and received antipsychotic augmentation.
Eating disorders. Of the two studies targeting disordered eating, only one (Linardon et al., Reference Linardon, Shatte, Messer, McClure and Fuller-Tyszkiewicz2023; N = 155) utilized a DRPT design, reporting no benefit to offering a choice of approach for binge eating. The second study utilized a PRPT design that evolved part-way through the trial. Given this was our only identified study involving children and adolescents (9–18 years; Loeb et al., Reference Loeb, Weissman, Marcus, Pattanayak, Hail, Kung and Walsh2020; N = 59; targeting prodromal Anorexia Nervosa) we report more detail. A strong preference for family-based treatment over individual supportive psychotherapy emerged among carers of these youth, resulting in rates of randomization refusal rising to 75%. A treatment arm was then added for those refusing randomization to receive their treatment of choice; within the design limitations of this study, no impact of offering choice on outcomes was reported.
Quality of life. Three studies utilizing a DRPT design measured this outcome, with two (both targeting depression; Andersson et al., Reference Andersson, Kall, Juhlin, Wahlstrom, de Fine Licht, Fardeman and Berg2023, N = 197; Callaghan, Khalil, Morres, & Carter, Reference Callaghan, Khalil, Morres and Carter2011) finding a benefit to offering patient choice, while one targeting anxiety (Svensson et al., Reference Svensson, Nilsson, Perrin, Johansson, Viborg, Falkenstrom and Sandell2021, N = 221) did not. Two additional studies with a FRPT design measured quality of life. Koszycki and colleagues (2021 N = 97) found no effect of preference matching for a target group with anxiety. We report the effects from Cooper et al. (Reference Cooper, Messow, McConnachie, Freire, Elliott, Heard and Morrison2018; N = 36; targeting depression) under preference strength in the following section.
Strength of preference
Seven studies (23.3%) measured the strength of preference; six targeted depression (Cooper et al., Reference Cooper, Messow, McConnachie, Freire, Elliott, Heard and Morrison2018; Dunlop et al., Reference Dunlop, Kelley, Mletzko, Velasquez, Craighead and Mayberg2012; Dunlop et al., Reference Dunlop, Kelley, Aponte-Rivera, Mletzko-Crowe, Kinkead, Ritchie and Team2017; Moradveisi, Huibers, Renner, & Arntz, Reference Moradveisi, Huibers, Renner and Arntz2014; Raue et al., Reference Raue, Schulberg, Heo, Klimstra and Bruce2009; Uebelacker et al., Reference Uebelacker, Weinstock, Battle, Abrantes and Miller2018) and one targeted anxiety (obsessive-compulsive disorder; Wheaton et al., Reference Wheaton, Carpenter, Kalanthroff, Foa and Simpson2016), with only one study of a DRPT rather than an FRPT design (Raue et al., Reference Raue, Schulberg, Heo, Klimstra and Bruce2009). Measures varied from categorical (e.g., mild, moderate) to continuous (5–100 point scales). At baseline, two studies reported higher preference scores for psychological compared to pharmacological treatment (Raue et al., Reference Raue, Schulberg, Heo, Klimstra and Bruce2009; Wheaton et al., Reference Wheaton, Carpenter, Kalanthroff, Foa and Simpson2016), and one found no difference (Dunlop et al., Reference Dunlop, Kelley, Aponte-Rivera, Mletzko-Crowe, Kinkead, Ritchie and Team2017). Two studies described the relative effect of preference strength on outcomes. Raue et al. (Reference Raue, Schulberg, Heo, Klimstra and Bruce2009) found preference strength (5-point response scale) was a more sensitive measure of outcome than dichotomous match/mismatch to desired treatment, positively associated with treatment uptake and adherence, but not predictive of depression outcomes. By contrast, Dunlop et al. (Reference Dunlop, Kelley, Aponte-Rivera, Mletzko-Crowe, Kinkead, Ritchie and Team2017) found strength of preference (mild, moderate, very strong) did not impact remission rates. When preference was measured separately for treatment approaches, Moradveisi, Huibers, Renner, and Arntz (Reference Moradveisi, Huibers, Renner and Arntz2014) found preference for one modality was not necessarily negatively correlated with preference for the other approach offered. Cooper et al. (Reference Cooper, Messow, McConnachie, Freire, Elliott, Heard and Morrison2018) found no impact of preference strength at post-intervention, but at 6-month follow-up, those with higher preference scores who were matched to person-centered counseling (but not to low-intensity CBT) reported a higher quality of life.
Uptake rates. Five studies utilized DRPT designs. Of these, three found that offering choice resulted in greater uptake (Brenes et al., Reference Brenes, Divers, Miller, Anderson, Hargis and Danhauer2020, N = 500, targeting anxiety; Hegerl et al., Reference Hegerl, Hautzinger, Mergl, Kohnen, Schutze, Scheunemann and Henkel2010, N = 368 and Raue et al., Reference Raue, Schulberg, Heo, Klimstra and Bruce2009, N = 60, both targeting depression). Linardon et al. (Reference Linardon, Shatte, Messer, McClure and Fuller-Tyszkiewicz2023), N = 155, binge eating) and Svensson et al. (Reference Svensson, Nilsson, Perrin, Johansson, Viborg, Falkenstrom and Sandell2021), N = 221, anxiety) found no benefit. Two studies utilizing FRPT designs found no benefit of preference matching on uptake (Cooper et al., Reference Cooper, Messow, McConnachie, Freire, Elliott, Heard and Morrison2018, N = 36, depression; Koszycki, Ilton, Dowell, & Bradwejn, Reference Koszycki, Ilton, Dowell and Bradwejn2022, N = 97, anxiety).
Quality assessment
Supplementary Figure S4 illustrates the frequency of studies meeting CONSORT quality criteria with data for each individual study in Supplementary Table S2. Overall, only 6 studies (20%) were rated as high quality (meeting ≥8 of 10 criteria). Across studies, the top three criteria with the highest compliance were eligibility criteria (86.6%; n = 26 studies), baseline demographic and clinical characteristics (73.3%; n = 22 studies), and statistical methods used to compare groups for the primary outcome (56.6%; n = 17 studies). The criterion with the lowest compliance rate was an explanation of how the sample size was derived (i.e., a priori power analyses) with only 8 of 30 studies meeting this criterion (26.6%). The second lowest compliance criterion was the provision of the description of participant losses and exclusions (30%; n = 6 studies).
Discussion
We conducted a systematic review and meta-analysis of the impact of patient choice on treatment outcomes, uptake, and dropout in RCTs across three common and comorbid mental health disorders. Of the 30 studies included at the systematic review stage, where any study design was included, the majority targeted depression (n = 23), five focused on anxiety disorders, and two investigated disordered eating. Only one study involved youth (aged 9–18 years), where a strong preference for family-based therapy was driven by carers. All other studies were conducted with middle-aged adults.
Restricting our analyses to the more stringent FRPT and DRPT designs, we identified 22 studies, sufficient to undertake meta-analyses for depression outcomes, and for dropout rates across mixed disorders and within depression studies. This significantly increases our ability to develop robust estimations compared to Delevry and Le (Reference Delevry and Le2019). There were insufficient studies to undertake meta-analyses for anxiety or eating disorders, for uptake rates, or using the most stringent choice design alone (DRPT studies) for any outcome.
Across combined FRPT and DRPT study designs, those who were preference-matched or offered choice had greater improvements in depression (d = 0.17, n = 18). Participants who were not offered a choice or matched to their preferred treatment were 1.46 times more likely to drop out (n = 17 studies; mixed target conditions). When dropout was examined purely for depression as a target disorder (n = 12 studies), effect sizes were similar but with wider confidence intervals, suggesting a lack of power. Although small, our effects are equivalent in size to the treatment effects found in meta-analyses comparing two established psychotherapies (Lindhiem, Bennett, Trentacosta, & McLear, Reference Lindhiem, Bennett, Trentacosta and McLear2014). These effect sizes are also comparable to those found for other optimization strategies, including personalization of therapy (0.14–0.22; Nye, Delgadillo, & Barkham, Reference Nye, Delgadillo and Barkham2023) and use of progress feedback (0.15–0.17; de Jong et al., Reference de Jong, Conijn, Gallagher, Reshetnikova, Heij and Lutz2021).
A major limitation of this meta-analysis is the small number of robust studies investigating the impact of choice on anxiety and eating disorder outcomes, and in youth samples. It thus remains unclear whether offering choice has the same impact across mental health disorders. For example, anxiety increases the likelihood that ambiguous options will be interpreted negatively, and potential negative outcomes avoided, even at the cost of missing potential gains (Hartley & Phelps, Reference Hartley and Phelps2012). In contrast, depression is associated with decreased memory of previous rewards to guide decision-making (Rupprechter et al., Reference Rupprechter, Stankevicius, Huys, Steele and Seriès2018). Further work on the impact of choice for eating disorders is critical, given the barriers to engaging with treatment. Some first-line treatments for eating disorders already incorporate some collaborative choice of treatment elements (Fairburn and Beglin, Reference Fairburn, Beglin and Fairburn2008; Schmidt, Wade, & Treasure, Reference Schmidt, Wade and Treasure2014), and robust testing of choice can inform the improvement of existing therapies.
Only one study investigated the impact of preference on children and adolescents (Loeb et al., Reference Loeb, Weissman, Marcus, Pattanayak, Hail, Kung and Walsh2020), evaluating the choice of carers and not the children. There is an absence of any studies that examine the impact of treatment choice in emerging adults who can be expected to be in treatment independently of family. This somewhat startling omission may indicate some unease about allowing younger people to make such choices. It is clearly a priority for future research, given the escalation of mental health problems in this group, and the subsequent personal and societal costs (McGorry et al., Reference McGorry, Mei, Dalal, Alvarez-Jimenez, Blakemore, Browne and Killackey2024). The explosion of studies examining digital mental health interventions with emerging adults (Taylor et al., Reference Taylor, Liu, Abelson, Eisenberg, Lipson and Schueller2024) and adolescents (Wüllner et al., Reference Wüllner, Hermenau, Krutkova, Petras, Hecker and Siniatchkin2024) poses a very suitable platform to examine choice and its impact on outcomes.
Further limitations include: (1) only 20% of studies were judged to be high quality; (2) the potential impact of using mixed designs (DRPT and FRPT) to develop estimations. Only seven studies of depression using DRPT were located, 57% of which did not support the advantage of choice; (3) our results do not address impact beyond post-intervention.
In summary, this meta-analysis found that treatment choice in depression can improve outcomes and decrease premature cessation of therapy. Future studies would benefit from incorporating high-quality preference (i.e., DRPT) designs that can best answer questions about treatment choice across depression and other mental health disorders. Greater numbers of DRPT designs would also allow moderator analyses to explore factors that may impact the choice effect: these might include demographic factors (e.g., gender, level of education, socioeconomic status, cultural background), amount and format of treatment information given prior to offering choice (e.g., discussion with a practitioner versus reading a guideline), and whether options for choice matter (e.g., selecting between antidepressant medication and psychotherapy, versus two types of psychotherapy). While there is an indication that a more sensitive measure of choice may be more informative, further work on whether this predicts great variance in outcomes is also required. Also of interest is an investigation of situations where the patient’s choice is at variance with the health professional’s belief in what is needed. Following the lead of depression researchers, anxiety disorders, eating disorders, and youth are key target groups for future research on the benefits of patient choice.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291725000066.
Data availability statement
This review has been pre-registered with the Open Science Framework: https://doi.org/10.17605/OSF.IO/SNJXR.
Funding statement
JL is a recipient of Grant APP1196948 National Health and Medical Research Council. TDW is a recipient of Investigator Grant 2025665 National Health and Medical Research Council. Funding support for CJ and MR was received from the same grant.
Competing interest
All authors declare no conflict of interest.
Open access
