A Parallel-Arm, Randomized Trial of Behavioral Activation Therapy for Anhedonia versus Mindfulness-Based Cognitive Therapy for Adults with Anhedonia

Abstract

Background:

Anhedonia, deficits in motivation and pleasure, is a transdiagnostic symptom of psychopathology and negative prognostic marker.

Methods:

In this randomized, parallel-arm clinical trial, a novel intervention, Behavioral Activation Treatment for Anhedonia (BATA), was compared to an individually administered Mindfulness-Based Cognitive Therapy (MBCT) in a transdiagnostic cohort of adults with clinically significant anhedonia (ClinicalTrials.gov Identifiers NCT02874534 and NCT04036136). Participants received 8 – 15 individual psychotherapy sessions, once weekly, with either BATA (n=61) or MBCT (n=55) and completed repeated self-report assessment of anhedonia and other internalizing symptoms.

Results:

Indicators of treatment feasibility were similar across conditions, though MBCT showed a trend towards greater attrition rates than BATA, with an adjusted odd’s ratio of 2.04 [0.88, 4.73]. Treatment effects on the primary clinical endpoint of anhedonia symptoms did not significantly differ, with a 14-week estimated difference on the Snaith Hamilton Pleasure Scale (SHAPS) of −0.20 [−2.25, 1.84] points in BATA compared to MBCT (z = 0.19, p = 0.845, d = 0.05). The expected 14-week change in SHAPS scores across conditions was −7.18 [−8.22, −6.15] points (z = 13.6, p < 0.001, d = 1.69). There were no significant differences in the proportion of participants demonstrating reliable and clinically significant improvements in SHAPS scores, or in the magnitude of internalizing symptom reductions.

Limitations:

Limitations included a modest sample size, lack of longer-term follow up data, and non-preregistered analytic plan.

Discussion:

There was no evidence to support superior clinical efficacy of BATA over MBCT in a transdiagnostic cohort of adults with elevated anhedonia. Both interventions reduced anhedonia symptoms to a comparable magnitude of other existing treatments.

Introduction

Disturbed positive affect is increasingly recognized as a significant feature of psychopathology. Anhedonia, characterized by a diminished capacity for motivation and pleasure from typically enjoyable activities, stands out as a key diagnostic criterion for major depressive disorder (MDD) and is prevalent across anxiety, traumatic-stress, and psychotic disorders (Krzyzanowski et al., 2022; Rizvi et al., 2016; Taylor et al., 2022). Within these conditions, anhedonia contributes to impaired daily functioning and reduced quality of life. Research indicates that anhedonia is a strong predictor of impaired social functioning among depressed individuals (Vinckier et al., 2017). Furthermore, low motivation and pleasure have emerged as risk factors for suicidal thoughts and behaviors across diverse populations and age groups (Auerbach et al., 2022; Bonanni et al., 2019), supported by prospective longitudinal studies (Ducasse et al., 2021; Spijker et al., 2010; Zielinski et al., 2017).

Depression treatments often result in less improvement in positive affect compared to negative affect or somatic symptoms (Nierenberg, 2015). This residual blunting of positive emotions may hinder patients from experiencing full recovery and wellness post-treatment. Challenges addressing anhedonia are evident across behavioral (Alsayednasser et al., 2022; Dunn et al., 2019) and pharmacological (Dunn et al., 2019; McMakin et al., 2012; Uher et al., 2012) intervention modalities. Additionally, certain antidepressant medications, such as selective serotonin reuptake inhibitors, can cause anhedonic features like emotional blunting (Goodwin et al., 2017) and decreased libido (Rosen et al., 1999). These antidepressant side effects underscore the need for interventions that effectively target and improve positive affect.

In recent clinical trials, several pharmacologic and neurostimulation interventions have shown efficacy for anhedonia compared to placebo or sham control conditions. These interventions include a novel selective κ-opioid receptor antagonist (Krystal et al., 2020; Pizzagalli et al., 2020), the KCNQ2/3 potassium channel modulator ezogabine, (Costi et al., 2021), and transcranial direct current stimulation of the left dorsolateral prefrontal cortex (Kong et al., 2024). The Snaith Hamilton Pleasure Scale (SHAPS) served as a common anhedonia outcome measure across these studies (Snaith et al., 1995), facilitating direct comparisons of results. Standardized effect sizes¹ indicate treatment advantages over placebo ranging in size from small (d=0.36) for transcranial direct current stimulation, to medium (d=0.62) for the κ-opioid antagonist, and large (d=0.93) for ezogabine. While none of these studies reported anhedonia remission rates using the SHAPS, post-treatment SHAPS scores suggest that most participants remained in a clinically elevated range of anhedonia severity .

Targeted psychotherapies have also emerged as promising approaches to treat anhedonia. While traditional psychotherapeutic interventions for mood and anxiety disorders have primarily focused on alleviating negative affect and associated maladaptive cognitions, several newer protocols place relatively greater emphasis on the psychological mechanisms mediating reward and behavioral approach. For instance, Positive Affect Treatment (PAT) was developed as a transdiagnostic intervention comprising three modules dedicated to pleasant activity scheduling, attention training towards positive emotion, and cultivating compassion towards self and others through meditation (Craske et al., 2023; Craske et al., 2019). Compared to a Negative Affect Treatment (NAT) control condition, PAT demonstrated superior improvements in positive affect, as well as anxiety, stress, and suicidal ideation. Another novel psychotherapy, Amplification of Positivity Treatment (AMP) was developed as a transdiagnostic intervention consisting of three modules focusing on increased engagement in positive activities, practicing gratitude, and fostering kindness and generosity towards others (Taylor et al., 2017). In a mechanistic trial involving participants with mood or anxiety disorders, AMP enhanced experimental indices of social functioning compared to a waitlist condition (Taylor et al., 2024). Lastly, Augmented Depression Therapy (ADepT) is a recent intervention tailored for anhedonia in MDD, emphasizing values clarification, behavioral activation, and problem-solving for barriers to resilience (Dunn, Widnall, et al., 2023). Results of a pilot trial showed that ADepT outperformed traditional Cognitive Behavioral Therapy (CBT) across multiple measures of depression severity, including the SHAPS (with an effect size of d=0.44 at 6-months).

Behavioral activation (BA) is a common foundational component in these therapies. The objective of BA is to improve recognition of avoidance patterns that contribute to the persistence of low mood and promote engagement in activities that potentially offer positive reinforcement. To that end, treatment predominantly involves activity scheduling based on identified values, and monitoring barriers to pursuing activities. Through repeated engagement with potential reinforcers, patients learn which activities bring the greatest enjoyment and purpose to their lives. This iterative process results in greater likelihood of positive reinforcement which is theorized to improve underlying motivation. The efficacy of BA in the treatment of depression and other internalizing symptoms of psychopathology above waitlist control and care-as-usual conditions is well established (Stein et al., 2021; Uphoff et al., 2020). However, meta-analyses also indicate that in randomized, parallel-arm studies, BA is effective at reducing depression to a comparable degree with CBT, and possibly psychodynamic therapy and antidepressant medications (Uphoff et al., 2020). The acceptability of BA for depression to patients, measured via attrition rates, also appears comparable to active treatments, but may be slightly lower than care-as-usual conditions.

Few studies have reported the efficacy of BA specifically on anhedonia severity. Most notably, Alsayednasser et al. (2022) conducted a secondary analysis of the COBRA trial comparing BA and CBT for depressed adults. The authors compared groups on mean changes in SHAPS total scores as well rates of reliable and clinically significant improvements (RCSI) on the SHAPS. The results of their trial indicated no significant differences between BA and CBT, with 36% and 38% of patients respectively demonstrating RCSI at 18 months post treatment onset, and pre-to-post improvement effect size of d=0.97 across the sample. This finding indicates cognitive restructuring (within CBT) did not improve outcomes beyond the application of core BA techniques.

In 2017, we initiated a randomized, parallel-arm clinical trial evaluating a novel psychotherapy, Behavioral Activation Treatment for Anhedonia (BATA), to treat anhedonia transdiagnostically (ClinicalTrials.gov Identifiers NCT02874534 and NCT04036136). Consistent with the NIMH Research Domain Criteria Project (RDoC) and experimental therapeutic framework for psychosocial trials², the goal of this trial was to evaluate the effects of BATA, relative to an individually administered Mindfulness-Based Cognitive Therapy (MBCT) comparison intervention, on mechanistic endpoints (Insel et al., 2010; Krystal et al., 2018), including fMRI and neurocognitive measures of reward processing, as well as clinical outcomes (i.e., change in self-reported anhedonia severity). A detailed analysis of treatment effects on mechanistic markers of reward processing will be the focus of a separate report, and in this manuscript, we detail clinical outcomes.

BATA is a modification of Behavioral Activation Therapy for Depression (Lejuez et al., 2011), a validated treatment for mood disorders predicated on the framework that depression is exacerbated by reduced contact with potential reinforcers in the environment and that anhedonia may be treated by helping individuals engage in rewarding behaviors to establish the reinforcing qualities of these behaviors (Dimidjian et al., 2011; Hopko et al., 2003). Modifications to develop BATA included removal of content specific to depression and the addition of psychoeducation about anhedonia – including education about anticipatory versus consummatory anhedonia, positive versus negative reinforcement, and how anhedonia can foster avoidance – the streamlining of activity monitoring to reduce monitoring effort for low-motivation patients, a focus on increased frequency of initiation of new behaviors to target motivation, and additional exercises to increase present-moment savoring as a means to target pleasure capacity.

In the current trial, MBCT was chosen as the comparator because its mechanisms of action were hypothesized to impact different psychological and brain systems than BA therapies. Briefly, the goal of MBCT is to foster greater awareness of thoughts and emotions to disrupt unhelpful habits of mind and behavior. This is achieved through dedicated mindfulness practice – an attentional style involving intentional, nonjudgmental, present-moment awareness. Mindfulness practice, as incorporated in modern CBTs, is an adaptation from ancient meditative techniques originating in Theravada and Mahayana Buddhism (Gunaratana, 2011). While mindfulness-based interventions were originally developed and validated as an effective means of reducing depressive relapse (McCartney et al., 2021; Teasdale et al., 2000), recent meta-analyses have demonstrated that MBCT is as effective as standard interventions, including CBT, in treating active depression (Goldberg et al., 2019; Sverre et al., 2023). Mindfulness-based interventions have also demonstrated efficacy for the treatment of diverse psychiatric symptoms and conditions beyond depression, including anxiety and trauma-related disorders (Keng et al., 2011; Sverre et al., 2023), indicating suitability as a transdiagnostic intervention. While traditionally administered in a group format, individually administered MBCT has demonstrated efficacy in reducing depression symptoms (Paterniti et al., 2022) at rates that do not differ from group MBCT (Schroevers et al., 2016) or individual CBT (Paterniti et al., 2022).

As with any psychotherapy, there are likely several mechanisms of action for MBCT. Systematic and meta-analytic reviews of both self-report (Gu et al., 2015; van der Velden et al., 2015) and neuroimaging(Young et al., 2018) outcomes of mindfulness-based interventions emphasize increased present-moment metacognitive awareness, interoceptive awareness, and non-judgmental acceptance as mechanisms of mindfulness. Reductions in perseverative negative thinking, including rumination and worry, are key mediators of reduced depression symptoms and relapse risk (Gu et al., 2015; van der Velden et al., 2015). Mindfulness-based interventions may be effective for enhancing positive emotions (Garland et al., 2015) or reward-related neural activation (Young et al., 2018). However, impacts on positive affect or neural indicators of reward sensitivity are not consistently observed across studies (Collard et al., 2008) and effect sizes of changes in affective processes may not be as strong as those associated with changes in perseverative negative thinking in reducing depression symptoms (Batink et al., 2013). Increases in positive affect associated with MBCT may be enhanced through increased attention to external sensory stimuli (Dunn, Wiedemann, et al., 2023), as opposed to Behavioral Activation, which is theorized to increase positive affect via increased behavioral engagement with potentially reinforcing stimuli (Nagy et al., 2020).

We have reported in interim analyses significant and equivalent reductions in anhedonia severity across BATA and MBCT. We also reported shared patterns of attenuating resting state functional connectivity within the brain’s default mode and frontoparietal control networks after treatment, though neural changes were unrelated to changes in SHAPS scores (Cernasov et al., 2021). Equivalency was reported again in Phillips et al. (2023) and Cernasov et al. (2023) which used different modeling strategies to specify change in symptoms over time. The current report extends our previous work by analyzing clinical endpoints in the full study cohort. We sought to answer the following four research questions:

• (1)How do BATA and MBCT compare on indicators of treatment feasibility, including attrition, attendance, homework completion or satisfaction ratings?
• (2)How do BATA and MBCT compare in the temporal trajectories and magnitude of change in anhedonia symptoms?
• (3)How do the proportion of participants in BATA and MBCT exhibiting RCSI in anhedonia symptoms compare?
• (4)How do BATA and MBCT compare in the temporal trajectories and magnitude of change in other internalizing symptoms?

Methods

Participants

Clinical and demographic sample characteristics are reported in Table 1. Participants were adults, ages 18 – 50 years old, with clinically significant anhedonia, as defined by Snaith Hamilton Pleasure Scale (SHAPS) scores ≥ 20 (Franken et al., 2007) and Clinical Global Impression–Severity (CGI-S) scores ≥ 3, representing a sample that was at least ‘mildly ill’ (Busner & Targum, 2007). Though 74% of participants met criteria for at least one DSM-5 diagnosis, this was not an eligibility requirement. Exclusionary criteria were psychotropic medication use within 30 days of screening, receiving concurrent psychotherapy, past behavioral activation or mindfulness-based interventions, contraindications for MRI, or history of significant neurologic disorder, eating disorder, bipolar disorder, psychotic disorder, or moderate to severe substance use disorder, assessed via the Structured Clinical Interview for DSM-5, Research Version (SCID-5) (First MB, 2015). Figure 1 depicts the CONSORT diagram of participant enrollment.

Table 1.

Baseline Demographic and Clinical Characteristics of the Intent-to-Treat Sample

Characteristics	Total (n=116)	MBCT (n=55)	BATA (n=61)
Age
Mean (SD)	29.18 (8.99)	30.20 (9.45)	28.26 (8.51)
Median (IQR)	26.50 (13.00)	28.00 (13.50)	25.00 (13.00)
Sex
Female	79 (68%)	38 (69%)	41 (67%)
Male	37 (32%)	17 (31%)	20 (33%)
Race
Asian	23 (20%)	9 (16%)	14 (23%)
Black	18 (16%)	8 (15%)	10 (16%)
Indigenous American	2 (1.7%)	1 (1.8%)	1 (1.6%)
Multiracial	4 (3.4%)	0 (0%)	4 (6.6%)
Other	4 (3.4%)	3 (5.5%)	1 (1.6%)
White	65 (56%)	34 (62%)	31 (51%)
Ethnicity
Hispanic/Latino	14 (12%)	9 (16%)	5 (8.2%)
Not Hispanic/Latino	101 (87%)	45 (82%)	56 (92%)
Declined Response	1 (0.9%)	1 (1.8%)	0 (0%)
Education
High School	5 (4.3%)	2 (3.6%)	3 (4.9%)
Some College	29 (25%)	14 (25%)	15 (25%)
Associates	6 (5.2%)	2 (3.6%)	4 (6.6%)
Bachelors	44 (38%)	20 (36%)	24 (39%)
Masters	24 (21%)	14 (25%)	10 (16%)
Doctoral	8 (6.9%)	3 (5.5%)	5 (8.2%)
Relationship Status
Never Married	67 (58%)	29 (53%)	38 (62%)
Cohabiting	10 (8.6%)	4 (7.3%)	6 (9.8%)
Married	32 (28%)	17 (31%)	15 (25%)
Separated	3 (2.6%)	1 (1.8%)	2 (3.3%)
Divorced	4 (3.4%)	4 (7.3%)	0 (0%)
Income
Mean (SD)	85.62 (85.80)	75.41 (58.77)	94.83 (104.03)
Median (IQR)	60.00 (90.00)	50.00 (83.75)	60.00 (100.00)
Snaith Hamilton Pleasure Scale
Mean (SD)	37.34 (4.24)	36.96 (4.75)	37.67 (3.73)
Median (IQR)	37.50 (5.25)	36.00 (5.50)	38.00 (5.00)
Beck Depression Inventory ‡
Mean (SD)	22.62 (9.13)	22.78 (9.66)	22.48 (8.69)
Median (IQR)	22.00 (13.00)	20.50 (12.25)	22.00 (13.50)
PTSD Checklist for DSM-5
Mean (SD)	25.20 (14.48)	25.31 (14.80)	25.10 (14.31)
Median (IQR)	21.50 (18.75)	22.00 (18.50)	21.50 (19.75)
Perceived Stress Scale ‡
Mean (SD)	23.71 (5.83)	23.25 (6.06)	24.11 (5.65)
Median (IQR)	24.00 (9.00)	24.00 (7.50)	25.00 (9.90)
Penn State Worry Questionnaire ‡
Mean (SD)	55.27 (16.00)	53.02 (16.68)	57.33 (15.20)
Median (IQR)	56.00 (26.00)	55.00 (28.00)	56.50 (24.75)
Primary DSM-5 Diagnosis
Attention Deficit/Hyperactivity Disorder	2 (1.7%)	2 (3.6%)	0 (0%)
Adjustment Disorder	1 (0.9%)	0 (0%)	1 (1.6%)
Generalized Anxiety Disorder	14 (12%)	5 (9.1%)	9 (15%)
Major Depressive Disorder	43 (37%)	21 (38%)	22 (36%)
None	30 (26%)	13 (24%)	17 (28%)
Other Specified Anxiety Disorder	2 (1.7%)	2 (3.6%)	0 (0%)
Other Specified Traumatic Stress Disorder	1 (0.9%)	0 (0%)	1 (1.6%)
Persistent Depressive Disorder	10 (8.6%)	4 (7.3%)	6 (9.8%)
Premenstrual Dysphoric Disorder	1 (0.9%)	0 (0%)	1 (1.6%)
Post-traumatic Stress Disorder	7 (6.0%)	6 (11%)	1 (1.6%)
Social Anxiety Disorder	3 (2.6%)	1 (1.8%)	2 (3.3%)
Specific Phobia	1 (0.9%)	1 (1.8%)	0 (0%)
Unspecified Depressive Disorder	1 (0.9%)	0 (0%)	1 (1.6%)

^‡Missing values were registered for scores in which two or more items were unanswered (three for the Beck Depression Inventory, two for the PTSD Checklist for DSM-5, and one for the Penn State Worry Questionnaire).

Figure 1.

CONSORT flow diagram.

Trial Design

Procedures took place at Duke University Medical Center. One hundred and sixteen participants, enrolled between July 2017 and September 2022, were randomized to either BATA (n=61) or MBCT (n=55) in a 1:1 ratio using the REDCap electronic data capture system. During the initial R61 phase of the trial, random assignment was performed through a dynamic balancing protocol between groups of lower (20 – 30) and higher (31+) SHAPS scorers. Within the R33 phase, eligibility criteria were modified so that all participants had a SHAPS score of 30 or greater. Data coder and assessors were blind to treatment condition.

Treatments consisted of 45 – 60-minute individual sessions, delivered once weekly in a crossed-therapist design. Treatment protocols comprised a flexible course of 8 – 15 sessions, allowing for rapid responders to terminate early. Due to SARS-CoV-2 pandemic-related restrictions, 44 participants received teletherapy. Attrition included either participants who expressed an intention to cease participation or were withdrawn from the study due to missing four consecutive weeks of treatment, failing to maintain contact with study staff, or initiating contraindicated medications.

Figure 2 depicts the schedule of assessments in the trial. The primary clinical outcome, the SHAPS, was assessed at every appointment, approximately weekly. The number of unique SHAPS observations ranged between 3 – 20 (modal frequency of 15) per participant. Secondary clinical outcomes measuring other internalizing symptoms were assessed up to five times, approximately three weeks apart. Additional biological and neurocognitive measures were assessed at different schedules but are not included in this report.

Figure 2.

Schematic representation of the schedule of clinical assessments. The interval between the screening and baseline visit, and the baseline and treatment onset visit ranged between 0 to 4 weeks. Participants were randomized to BATA or MBCT in a 1:1 ratio. There was variability in the total number of sessions attended; treatment completers could receive as few as 8 and as many as 15 sessions, depending on mutual agreement between the study therapist and participant that sufficient benefit was achieved. The primary clinical endpoint, the SHAPS, was collected at every appointment in the trial. The percentage of therapy homework completion was assessed at each therapy session by both the participant and the clinician. Assessment of internalizing psychopathology using the BAI, BDI, PCL5, PSS, and PSWQ was conducted prior to treatment onset, approximately at weeks 4, 8, and 12 during treatment, and at the treatment completion visit. Overall satisfaction with the treatment was evaluated using the TAQ at the treatment completion visit. BAI – Beck Anxiety Inventory; BATA – Behavioral Activation Therapy for Anhedonia; BDI – Beck Depression Inventory; MBCT – Mindfulness-Based Cognitive Therapy; PCL5 – PTSD Checklist for the DSM-5; PSS – Perceived Stress Scale; PSWQ – Penn State Worry Questionnaire; TAQ – Treatment Acceptability Questionnaire

Participants were queried at every appointment about changes in their health or medications. Adverse events were documented regardless of relatedness to study procedures and causality was determined upon review by the principal investigators. Pre-existing conditions that were not exacerbated in frequency or severity were not recorded.

Interventions

Study therapists were PhD-level clinical psychologists with at least one year of prior training in CBT and mindfulness-based interventions. Detailed descriptions of the psychotherapeutic interventions are reported in the Supplemental Materials along with therapist training experiences and adherence rating results. In brief, BATA was adapted from BATD (Lejuez et al., 2011) to treat anhedonia transdiagnostically. Intervention components included psychoeducation, activity monitoring, values clarification exercises, goal setting, scheduling of valued activities, support for behavioral initiation, encouragement of savoring, and problem solving around barriers to activation. Increased positive affect and decreased negative affect were theorized to result from enhanced awareness of behavioral patterns, reduced avoidance, and increased contact with potential reinforcers. MBCT was administered in an individual format retaining the primary components of traditional group MBCT including didactic instruction, guided in-session meditations, between-session home practice, and inquiry of subjective experience (Segal & Teasdale, 2018; Wahbeh et al., 2014).

Measures

Self-Report Symptom Scales

The primary clinical outcome measure was the SHAPS, a 14-item self-report questionnaire assessing capacity for pleasure, specifically with respect to consummatory reward processing (Snaith et al., 1995). Each item pertains to a typically enjoyable activity, such as watching television, taking a shower, or receiving praise from other people. Total scores range between 14 – 56, with higher values indicative of worse anhedonia (Franken et al., 2007).

Given the transdiagnostic nature of the sample, five self-report questionnaires assessing wide-ranging internalizing symptoms of psychopathology were administered as secondary clinical endpoints – the Beck Anxiety Inventory (BAI) (Steer & Beck, 1997), Beck Depression Inventory, 2^nd edition (BDI) (Beck et al., 1996), PTSD Checklist for DSM-5 (PCL-5) (Blevins et al., 2015), Perceived Stress Scale (PSS) (Cohen et al., 1983), and Penn State Worry Questionnaire (PSWQ) (Meyer et al., 1990). Brief descriptions of each scale are reported in the Supplemental Materials. In all cases, higher total scores indicate greater severity, distress, and impairment.

Therapy Homework Completion Ratings

Homework completion was gauged by the participant and the clinician at each therapy session, using a 0 – 100% scale. Clinicians were instructed to consider documentation from home activity logs and the content of discussions during therapy to make their judgements.

Treatment Acceptability Questionnaire (TAQ)

A modified, seven item self-report TAQ was administered at the conclusion of study participation (Hunsley, 1992). Items pertained to satisfaction with the treatment, therapist, therapy sessions, therapy handouts, and homework assignments, as well as expectations of enduring benefits, and the likelihood of recommending treatment to others. Responses were made on a five-point Likert scale and coded such that higher positive values indicate more favorable opinions of treatment (total scores −14 – +14).

Clinical Global Impressions (CGI) Scale

The CGI scale is a standardized assessment tool used to evaluate the severity of a patient’s condition and their improvement over time, relying largely on clinician judgment. Initial severity scores (CGI-S) were recorded at the baseline assessment, and therapists provided ratings of improvement (CGI-I) relative to this baseline functioning at each subsequent treatment session. There are seven levels of CGI-I scores spanning “very much improved,” “much improved,” “minimally improved,” “no change,” “minimally worse,” “much worse,” and “very much worse.” The anchors for these descriptive labels can be reviewed in Busner and Targum (2007).

Analytic Plan

Details concerning R software packages and general statistical modeling strategies including diagnostics and validation, are reported in the Supplemental Materials. The analyses pertaining to the SHAPS were not pre-registered and deviate from the statistical and analytic plan outlined on clinicaltrials.gov³. We believe the approach adopted, using mixed effects growth curve models (detailed below), is supported by a stronger theoretical rationale than the originally proposed post-treatment ANCOVA model as it allows for the full utilization of mid-treatment data and offers a wealth of information beyond endpoint mean differences between the treatments.

Question 1: Treatment Feasibility

Differences in attrition rates between BATA and MBCT were examined using a generalized linear model with a logit link function (McCullagh, 1984). Age, sex, and baseline visit SHAPS scores were included as covariates to derive adjusted odds of attrition for each intervention arm.

Non-parametric Wilcoxon rank sum tests (McKnight & Najab, 2010) compared the distributions of four additional feasibility variables between BATA and MBCT. These variables were (1) the total number of therapy sessions attended, (2) the proportion of homework assignments completed as reported by participants, (3) the proportion of homework assignments completed as reported by clinicians, and (4) TAQ total scores.

Question 2: Anhedonia Trajectories

SHAPS scores were analyzed longitudinally using linear mixed effects models, also known as multilevel growth curve models (Curran et al., 2010). Time was coded as a continuous variable measured in weeks, centered on the initial therapy session date, and rounded to the nearest integer. Initially, SHAPS score trajectories were estimated across the whole sample in an unconditional growth model containing time as the sole predictor. The optimal pattern of change over time and the appropriate variance component structure were identified in a stepwise procedure comparing nested models using ${\chi }^2$ likelihood ratio tests, Akaike information criterion (AIC), and Bayesian information criterion (BIC) values (Portet, 2020). Detailed model specifications can be found in the Supplemental Materials.

Subsequently, a conditional growth model was fitted, incorporating age, sex, total therapy sessions attended, and treatment condition as time-invariant covariates. Continuous variables were mean centered. Female sex and MBCT were used as reference levels for factor variables. The formulation for the final conditional model is expressed below. The effect of time was specified as a piecewise function with two segments, joined by a knot at the first therapy session. The pretreatment segment contained three observations and modeled linear change. The active treatment segment contained up to 16 observations and modeled change as a quadratic function. Treatment condition was specified as a cross-level interaction with the time effects within the active treatment segment to evaluate differences in SHAPS score trajectories between BATA and MBCT⁴.

Level-1

$$\begin{gathered} SHAP{S}_{\text{ij}}={\text{β}}_{0\text{j}}+{\text{β}}_{1\text{j}}TimeSegment{1}_{\text{ij}}+{\text{β}}_{2\text{j}}TimeSegment{2}_{\text{ij}}+{\text{β}}_{3\text{j}}TimeSegment{2}_{\text{ij}}^2+{\text{ε}}_{\text{ij}} \\ \begin{array}{c}{\text{ε}}_{\text{ij}}\sim \text{N}(0,\text{Σ})\\ \text{Σ}=\left(\begin{array}{cccc}{\text{σ}}^2& -& \cdots & -\\ \text{σϕ}& {\text{σ}}^2& \cdots & -\\ ⋮& ⋮& \ddots & ⋮\\ \text{σ}{\text{ϕ}}^{\text{k}}& \text{σ}{\text{ϕ}}^{\text{k}-1}& \cdots & {\text{σ}}^2\end{array}\right)\end{array} \end{gathered}$$

Level-2

$$\begin{gathered} {\text{β}}_{0\text{j}}={\text{γ}}_{00}+{\text{γ}}_{01}Ag{e}_{\text{j}}+{\text{γ}}_{02}Se{x}_{\text{j}}+{\text{γ}}_{03}TotalSession{s}_{\text{j}}+{\text{γ}}_{04}Treatmen{t}_{\text{j}}+{\text{u}}_{0\text{j}} \\ \begin{array}{c}{\text{β}}_{1\text{j}}={\text{γ}}_{10}+{\text{u}}_{1\text{j}}\\ {\text{β}}_{2\text{j}}={\text{γ}}_{20}+{\text{γ}}_{21}Treatmen{t}_{\text{j}}+{\text{u}}_{2\text{j}}\\ {\text{β}}_{3\text{j}}={\text{γ}}_{30}+{\text{γ}}_{31}Treatmen{t}_{\text{j}}+{\text{u}}_{3\text{j}}\end{array} \\ \left(\begin{array}{l}{\text{u}}_{0\text{j}}\hfill \\ {\text{u}}_{1\text{j}}\hfill \\ {\text{u}}_{2\text{j}}\hfill \\ {\text{u}}_{3\text{j}}\hfill \end{array}\right)\sim \text{N}\left[\left(\begin{array}{l}0\hfill \\ 0\hfill \\ 0\hfill \\ 0\hfill \end{array}\right),\left(\begin{array}{cccc}{\text{τ}}_{00}& -& -& -\\ {\text{τ}}_{01}& {\text{τ}}_{11}& -& -\\ {\text{τ}}_{02}& {\text{τ}}_{12}& {\text{τ}}_{22}& -\\ {\text{τ}}_{03}& {\text{τ}}_{13}& {\text{τ}}_{23}& {\text{τ}}_{33}\end{array}\right)\right] \end{gathered}$$

Finally, a multiple-group conditional growth model was fitted, specifying separate level-2 random effects variance parameters and level-1 residual variance for BATA and MBCT. The significance of these additional parameters in the multiple-group model were evaluated using χ² likelihood ratio tests and changes in AIC and BIC values relative to the single-group model. Essentially, the multiple-group model tested the hypothesis that the variability of participant response to treatment differed between BATA and MBCT, independently of group mean improvements.

While these growth models generate numerous results of substantive interest, we focused on estimating the difference in SHAPS scores between BATA and MBCT at the median trial duration of 14 weeks as our primary measure of treatment superiority. In addition to the null hypothesis significance test of any effect greater than 0, we also compared treatment effects against an estimated value for the minimum clinically important difference (MCID) on the SHAPS. We used final CGI-I scores to identify a threshold for meaningful change by computing mean ΔSHAPS scores (i.e., baseline values minus the last observation carried forward) across each level of CGI-I. The lower end of the 95% confidence interval (CI) for the mean ΔSHAPS score among participants who “minimally improved” provided a liberal estimate for the MCID value.

Question 3: Anhedonia RCSI Rates

Drawing from Alsayednasser et al. (2022), we defined a binary indicator of RCSI by employing two requisite distribution-based criteria. First, we set the threshold of < 25.88 for the last observed SHAPS score, representing a transition to greater probability of belonging to the non-clinical population distribution (Trøstheim et al., 2020) than to our sample baseline visit distribution. Under this criterion scores classified as RCSI are suggestive of anhedonia remission. Second, we determined the difference between the last observed and baseline SHAPS scores should be ≤ −6.45 points, which corresponds to the minimal detectable change (MDC) on the SHAPS, per 1.96 times the estimated standard error of the difference scores (Jacobson & Truax, 1991). Detailed formulae for these calculations are provided in the Supplemental Materials.

Differences in the rates of participants demonstrating RCSI on the SHAPS between BATA and MBCT were examined using a generalized linear model with a logit link function. Age, sex, total sessions attended, and baseline visit SHAPS scores were included as covariates to calculate adjusted odds of achieving RCSI within each intervention.

Question 4: Internalizing Symptom Trajectories

BAI, BDI, PCL-5, PSS, and PSWQ scores were analyzed longitudinally using mixed effects models. Growth modeling procedures for these scales resembled the process for SHAPS, however, the smaller number of observations limited the complexity of models fitted (see Supplemental Materials for details). Once the optimal unconditional model was identified, age, sex, total sessions attended, and treatment condition were included as covariates in conditional growth curve models, including a treatment-by-time interaction term to evaluate differences in symptom trajectories between BATA and MBCT.

Power Calculations

Power calculations were initially conducted for the originally proposed post-treatment SHAPS ANCOVA model using G*Power v3.1.9.7. Our sample size of 116 participants afforded 80% power to detect a standardized effect size of $d=0.525$ at two tailed $\alpha =0.05$. Analytic power calculations for mixed effects models with complex covariance parameter formulations, such as our SHAPS growth model, are challenging in comparison. Monte Carlo simulations offer a solution to estimate power for such models but nonetheless require a priori specification of many parameter values. Absent pilot data or comparable models reported in the literature to inform these assumptions, we conducted “partially post-hoc” calculations of what power would have been to detect various standardized effect sizes based on our final sample size and observed standard error from the growth model point estimate (Dziak et al., 2020).

Results

Question 1: Treatment Feasibility

There were no adverse events determined to be related to the interventions.

Attrition

The observed attrition rates [95% CIs] were 24.6% [13.8, 35.4] in BATA and 36.4% [23.7, 49.1] in MBCT. The unadjusted odds of attrition were 1.75 [0.79, 3.90] times higher for participants in MBCT compared to BATA.

Multiple logistic regression model results are reported in Table 2 and depicted in Figure 3. The adjusted attrition rates (marginalized across levels of sex and held at the mean of age and baseline SHAPS score) were 21.3% [10.6, 31.9] in BATA and 35.5% [21.7, 49.3] in MBCT. The adjusted odds of attrition were 2.04 [0.88, 4.73] times higher in MBCT compared to BATA, which did not reach statistical significance $(z=-1.67,p=0.095)$. There was a trending relation between baseline SHAPS scores and attrition such that participants with greater anhedonia severity were more likely to dropout $(z=1.89,p=0.059)$. The odds ratio of attrition for the interquartile range (IQR) effect of baseline SHAPS (i.e., increasing by 5.25 points) was 1.68 [0.98, 2.87].

Table 2.

Multiple Logistic Regression Model of Attrition

	Odds Ratio	Standard Error	p	95% Confidence Interval	Bias
Predictor
β0 Intercept	0.634	0.206	0.160	0.304 – 1.271	0.021
β1 Treatment [BATA]	0.489	0.209	0.095	0.187 – 1.185	0.012
β2 Age	0.958	0.025	0.096	0.887 – 1.000	-0.003
β3 Sex [Male]	0.749	0.357	0.545	0.257 – 1.878	0.030
β4 Baseline SHAPS	1.104	0.058	0.059	0.990 – 1.246	-0.001
Performance Metrics
R2	0.118
AIC	142.0
BIC	155.7
Brier Score	0.192
C-Index	0.680

The model includes observations from 116 participants. The 95% confidence intervals for the estimates were derived using the percentile method from parametric bootstrapping of 1,000 iterations. Bias refers to the difference between the mean of the bootstrapped estimates and the observed estimate. The odds ratios for continuous variables represent the effect of a one unit increase in the predictor.

Figure 3.

Multiple logistic regression model of attrition. (A) Predicted attrition rates between BATA and MBCT, across varying levels of initial anhedonia severity, conditioned on female sex and mean age of sample (30 years old). Ribbons denote 95% confidence intervals. (B) Model-implied odds ratios of attrition. The dark, medium, and light purple bars correspond to the 68%, 95%, and 99% confidence intervals respectively. Ratios for continuous variables represent the interquartile range effect (i.e., an individual age 35 vs 22 years old, or an individual with a baseline SHAPS score of 40 vs 34.75 points). The adjusted odds of attrition were 2.04 [0.88, 4.73] times higher in the MBCT group compared to the BATA group. The baseline SHAPS score was the only statistically significant predictor. (C) The area under the receiver operating characteristic curve was 0.68 indicating the model had relatively poor predictive ability to classify participants who dropped out.

Total Sessions Attended, Homework Completed, & TAQ Scores

Non-parametric test results are presented in Figure 4. There were no statistically significant differences between BATA and MBCT with respect to any treatment feasibility indicators. The median (IQR) number of sessions attended were 11 (3) in BATA and 10 (4) in MBCT. A total of 93 out of 116 participants completed the TAQ. The median scores were 9 (5.75) in BATA and 9 (7) in MBCT, indicating a reasonably high degree of satisfaction among those not lost to follow up. The median proportion of homework assignments completed per clinician-report were 82% (27.5) in BATA and 75.56% (37.33) in MBCT. The median proportion of homework assignments completed per participant-report were 85% (25.97) in BATA and 78.89% (32.35) in MBCT.

Figure 4.

Nonparametric analyses comparing the distributions of feasibility outcome variables between BATA and MBCT. Overall, there were no statistically significant differences. The median of all possible differences between observations across groups, known as the Hodges-Lehmann estimate of location shift, is reported alongside the median value for each treatment and the smaller of the two sums of ranks (i.e., the W statistic) within treatment condition. (A) The median (IQR) number of sessions attended was 11 (3) in the BATA condition and 10 (4) in the MBCT condition. (B) 93 out of 116 randomized participants completed the TAQ. The median (IQR) score was 9 (5.75) in the BATA condition and 9 (7) in the MBCT condition, suggesting a relatively high degree of satisfaction among those not lost to follow up. (C) Missing data from approximately 5% of clinician-reported homework completion ratings were imputed via mixed effect model predicted values incorporating random and fixed effects. Clinician-reported homework was specified as an outcome variable predicted by time-varying participant-reported homework completion, with random effects for intercept and slope. The correlation between predicted and observed values was r = 0.88. The median (IQR) proportion of clinician-reported homework completion was 82% (27.5) in the BATA condition and 75.56% (37.33) in the MBCT condition. (D) Missing data from approximately 3.96% of participant-reported homework completion ratings were imputed via mixed effect model predicted values in an analogous fashion to the clinician-reported homework completion ratings (i.e., using the other informant variable as a predictor). The resulting correlation between predicted and observed values was r = 0.90). The median (IQR) proportion of participant-reported homework completion was 85% (25.97) in the BATA condition and 78.89% (32.25) in the MBCT condition.

Question 2: Anhedonia Trajectories

SHAPS Reductions Across the Sample

Raw trajectories of SHAPS scores are depicted in Figure 5. Change in SHAPS scores over time was specified as a piecewise function with a pretreatment segment and an active treatment segment. Linear reductions in SHAPS scores were observed between the baseline visit and initial therapy session, at a rate of −0.26 [−0.41, −0.11] points per week $(t=-3.46,p=0.001)$. The expected value for SHAPS scores at the initial therapy session was 36.2 [35.4, 37.0] $(t=89.04,p<0.001)$. During the active treatment segment, the trajectory of SHAPS scores followed a quadratic curve that was shallow, concaveup, and decreasing. The rate of SHAPS score reduction significantly increased at the start of psychotherapy, relative to pretreatment reductions $(t=-3.08,p=0.002)$. The instantaneous slope at treatment onset was −0.66 [−0.82, −0.50] points per week $(z=-7.99,p<0.001)$, and subsequently decelerated by 0.01 [0.00, 0.02] points weekly $(t=2.41,p=0.016)$. The point estimate for the expected change in SHAPS scores after 14 weeks (the median duration of participation) was −7.18 [−8.22, −6.15] points $(z=13.6,p<0.001)$. Dividing this model-implied change score by the baseline visit standard deviation yielded a standardized effect size (i.e., comparable to effect sizes reported in traditional two-timepoint repeated measures designs) of $d=1.69$ (Feingold, 2009).

Figure 5.

Spaghetti plot depicting trajectories of observed SHAPS scores over time across 116 participants. Each participant is represented by a separate line. All observations following the screening visit are included, with a median and modal number of 15 assessments per participant (range: 3 – 20). Scores in the darker gray rectangular region of the graph, SHAPS < 25.88, have greater probability of belonging to the non-clinical sample distribution (as defined by meta-analysis) than the clinical sample distribution (as defined by the baseline visit data) – see the Supplemental Materials for an explanation of this calculation. The plot showcases possibly nonlinear improvements over time, with considerable variability across participants, and no pronounced differences between treatment conditions.

SHAPS Variability Across the Sample

The unconditional growth model revealed significant variability among participant’s trajectories, including pretreatment slopes (SD = 0.34), symptom severities at the initial therapy session (SD = 4.02), initial slopes upon treatment (SD = 0.77), and subsequent deceleration rates (SD = 0.03). Participants with higher SHAPS scores at their initial therapy session tended to show slower improvement rates before receiving treatment $(r=0.58)$ and faster improvement rates upon initiating treatment $(r=-0.35)$. Conversely, faster improvement rates before treatment were associated with slower improvement rates upon starting treatment (r = −0.54). Meanwhile, the residual variance was constant over time and there was a modest correlation between adjacent observations within participants $(\text{φ}=0.18)$.

SHAPS MCID

Among the last observed CGI-I ratings, the proportions of participants were roughly equal across categories for “very much improved” at 28.9% [20.6, 37.3], “much improved” at 30.7%, [22.2, 39.2], and “minimally improved” at 32.5%, [23.9, 41.1]. Approximately 7% [2.3, 11.7] were classified as “no change” and a single participant as “minimally worse”. The mean ΔSHAPS score associated with each category is depicted in Figure 6. The low end of the 95% CI for the mean ΔSHAPS score for participants whom “minimally improved” was 3.33, serving as an MCID for subsequent null-hypothesis significance testing.

Figure 6.

(Top) The proportion of CGI-I scores among the last observation for 114 participants in the trial. (Bottom) The mean and 95% CI for the ΔSHAPS score associated with each level of CGI-I. ΔSHAPS was calculated using the baseline visit and last observation carried forward. The low end estimate from the “minimally improved” category was used as an MCID on the SHAPS in our sample.

SHAPS Reductions in BATA versus MBCT

Conditional growth model results are detailed in Table 3 and depicted in Figure 7. Age, sex, and total therapy sessions attended were unrelated to SHAPS scores. The 14-week point estimate for the predicted difference in SHAPS scores between participants in the parallel arms, given equal values at treatment onset, was −0.20 [−2.25, 1.84] points lower in BATA than in MBCT $(d=0.05,z=0.19,p=0.845)$⁵. Accordingly, neither treatment-by-time interaction term was significant (i.e., the parameter estimates for the first or second order polynomial terms for the active treatment segment), indicating the trajectories of SHAPS scores did not differ between BATA and MBCT. The model-implied 14-week reduction in SHAPS scores for BATA was −7.20 [8.61, −5.79] points $(d=1.70,z=10.0,p<0.001)$, which was significantly greater than the proposed MCID of 3.33 $(z=5.39,p<0.001)$. . Meanwhile, the model-implied 14-week reduction in SHAPS scores for MBCT was −7.00 [−8.52, −5.48] points $(d=1.65,z=9.04,p<0.001)$, which was also significantly greater than the proposed MCID $(z=4.74,p=0.478)$.

Table 3.

Conditional Growth (Mixed Effects) Model of SHAPS

	Estimate	Standard Error	p	95% Confidence Interval	Bias
Fixed Effects
γ00 Intercept	35.894	0.614	< 0.001	34.608 – 37.032	0.012
γ10 Time Pretreatment Segment	−0.256	0.071	< 0.001	−0.399 – −0.127	0.007
γ20 Time Treatment Segment	−0.330	0.149	0.027	−0.629 – −0.032	−0.005
γ30 Time2 Treatment Segment	0.006	0.006	0.305	−0.005 – 0.018	−0.000
γ01 Age	−0.012	0.041	0.767	−0.098 – 0.067	−0.000
γ02 Sex [Male]	−0.045	0.773	0.953	−1.61 – 1.54	0.036
γ03 Total Sessions	−0.158	0.105	0.137	−0.350 – 0.029	−0.003
γ04 Treatment [BATA]	0.581	0.744	0.437	−0.882 – 2.18	−0.025
γ21 Treatment [BATA] x Time Treatment Segment	−0.137	0.158	0.387	−0.428 – 0.178	−0.003
γ31 Treatment [BATA] x Time2 Treatment Segment	0.009	0.008	0.287	−0.008 – 0.023	0.000
Variance Components
σ	2.457
SD τ00 (Intercept)	4.046
SD τ11 (Pretreatment Slope)	0.181
SD τ22 (Treatment Initial Velocity)	0.682
SD τ33 (Treatment Acceleration)	0.024
Cor τ01/10	0.992
Cor τ02/20	−0.376
Cor τ03/30	−0.063
Cor τ12/21	−0.308
Cor τ13/31	−0.125
Cor τ23/32	−0.875
ϕ Autocorrelation	0.185
Performance Metrics
R2Marginal	0.227
R2Conditional	0.827
AIC	7731.3
BIC	7848.8
RMSE	2.220

The model encompasses 1,543 observations across 116 participants. The 95% confidence intervals for the estimates were calculated using the percentile method from parametric bootstrapping with 1,000 iterations. Bias refers to the difference between the mean of the bootstrapped estimates and the observed estimate. In this model, γ₀₀ signifies the predicted SHAPS score at treatment onset for a female participant in the MBCT condition at the mean value of age (29.2 years) and total sessions completed (9.72 sessions). γ₁₀ is the slope of SHAPS over time in the pretreatment segment of the piecewise function. γ₂₀ is the change in the slope from the previous segment upon treatment commencement, specifically for the MBCT condition (i.e., γ₁₀ + γ₂₀ is equal to the slope at time = 0 for participants in MBCT). γ₃₀ is the subsequent change in the slope with the passage of each week (i.e., the deceleration rate), specifically for the MBCT condition. γ₂₁ is the difference in initial slope differential for BATA condition (i.e., γ₁₀ + γ₂₀ + γ₂₁ is equal to the slope at time = 0 for participants in BATA). γ₃₁ is the difference in the deceleration rate for the BATA condition (i.e., γ₃₀ + γ₃₁ is equal to the deceleration rate for participants in BATA). It is noteworthy that while the quadratic effect of time was statistically significant when estimated across the entire sample in the unconditional growth model (see Supplementary Materials Table S1), this was no longer the case for either BATA or MBCT when the effect was estimated separately between treatments. This phenomenon is most likely attributable to the reduced precision of the estimator with the smaller sample sizes when testing for the interaction.

Figure 7.

Conditional growth model-implied trajectories of SHAPS scores over time. Thin gray lines represent predicted values for individual participants using a combination of fixed and random effects. Colored lines represent the expected treatment effects in the population (i.e., calculated using fixed effects, with age and total sessions held at their mean values and sex specified as female). Ribbons denote the 95% confidence intervals for these estimates. The trajectory of SHAPS scores is characterized by a piecewise function containing two segments with a knot at the initial therapy session. Participants experienced significant linear reductions in SHAPS scores pretreatment at a rate of −0.26 [−0.4, −0.13] points per week. The slope of change over time significantly increased at the first psychotherapy session, with an instantaneous rate of −0.72 [−1.0, −0.41] points per week in BATA and −0.59 [−0.89, −0.29] points per week in MBCT. This rate slowly decelerated at a weekly pace of 0.01 [−0.01, 0.02] points in MBCT and 0.02 [−0.00, 0.03] points in BATA. There were no statistically significant differences in the trajectories between treatment conditions.

As sensitivity analyses, we ran a mixed effects model with the subset of participants who completed treatment per-protocol as well as a simplified ordinary least squares regression using the full sample, as originally proposed on clinicaltrials.gov. Both models corroborated the finding of equivalence between interventions⁶.

Partial post-hoc power calculations – utilizing the standard deviation of the baseline visit SHAPS score (4.24) and the standard error of the estimate from the growth model-implied 14-week treatment difference (1.04) – indicated 80% power to detect a standardized effect size of d = 0.69. Thus, we were poised to detect a raw SHAPS score difference of 2.91 points or larger between groups.

SHAPS Variability in BATA versus MBCT

Evidence of differences in SHAPS score variability between BATA and MBCT was equivocal. The likelihood ratio test between the single group and multiple group conditional models was marginally significant $\left({\chi }_{(11)}^2=19.77,p=0.049\right)$. However, both AIC (Δ = +2.23) and BIC (Δ = +60.92) values increased with the multiple group model, suggesting the complexity introduced by the additional parameters may be unjustified. Additional comparisons against the single group model showed a lack of improved fit when either the level-1 residual variance (${\chi }_{(1)}^2=2.71,p=0.100$; ΔAIC = −0.71, ΔBIC = +4.63) or the level-2 random effects variance components (${{\chi }^2}_{(10)}=17.74,p=0.059$; ΔAIC = +2.26, ΔBIC = +55.61) were specified in separate blocks per treatment, motivating the decision to retain the single group model as the final result.

A summary of the multiple-group conditional growth model is reported in Table 4. The most salient difference in random effects parameter was for the covariance between intercept and treatment onset slope. For participants assigned to BATA, there was almost no correlation between anhedonia severity at treatment onset and the subsequent rate of improvement $(r=-0.09)$. In contrast, for MBCT, participants with greater initial anhedonia showed more rapid improvement rates upon starting psychotherapy $(r=-0.54)$. Additionally, there was greater variability in the initial slopes within MBCT (SD = 0.88) compared to BATA (SD = 0.74).

Table 4.

Multiple-Group Conditional Growth (Mixed Effects) Model of SHAPS

Fixed Effects				Variance Components
Predictor	Estimate [Std. Error]		Intercept	Pretreatment Slope	Treatment Initial Velocity	Treatment Acceleration	Residual
γ00 Intercept	35.78 [0.697]	BATA
γ10 Time Seg1	−0.258 [0.077]	Intercept	3.173
γ20 Time Seg2	−0.308 [0.158]	Pretreatment Slope	0.489	0.379
γ30 Time2 Seg2	0.005 [0.005]	Initial Velocity	−0.089	−0.556	0.740
γ01 Age	−0.008 [0.039]	Acceleration	−0.331	0.026	−0.819	0.029	2.497
γ02 Sex [Male]	0.229 [0.739]	MBCT
γ03 Total Sessions	−0.130 [0.102]	Intercept	4.846
γ04 Treatment [BATA]	0.604 [0.757]	Pretreatment Slope	0.675	0.346
γ21 Treatment x Time Seg2	−0.177 [0.157]	Initial Velocity	−0.543	−0.642	0.877
γ31 Treatment [BATA] x Time2 Seg2	0.012 [0.008]	Acceleration	0.213	0.366	−0.927	0.020	2.353

The model encompasses 1,543 observations across 116 participants. Fixed effects coefficients are reported along with their standard errors in parentheses. The under script _Seg1 refers to the linear pretreatment segment of the piecewise model, while _Seg2 refers to the nonlinear active treatment segment. Variance components are reported as standard deviations and correlations. The residual auto-correlation parameter was 0.18.

Question 3: Anhedonia RCSI Rates

The observed rates of SHAPS RCSI were 14.8% [5.9, 23.7] in BATA and 25.5% [13.9, 37.0] in MBCT. The unadjusted odds of RCSI were 1.97 [0.78, 5.01] times higher in MBCT than in BATA. Multiple logistic regression model results are reported in Table 5 and depicted in Figure 7. The adjusted SHAPS RCSI rates (marginalized across levels of sex and held at the mean of age, baseline SHAPS score, and total sessions attended) were 11.3% [3.18, 19.5] in BATA and 22.6% [10.2, 35.1] in MBCT. The adjusted odds of RCSI were 2.29 [0.85, 6.18] times higher in MBCT than in BATA, which was not statistically significant $(z=-1.64,p=0.101)$. There was a significant positive association between total sessions attended and likelihood of SHAPS RCSI $(\text{z}=2.43,p=0.015)$. The odds ratio of SHAPS RCSI for the IQR effect of total sessions (i.e., increasing by 4 sessions) was 2.54 [1.20, 5.38].

Table 5.

Multiple Logistic Regression Model of SHAPS RCSI

	Odds Ratio	Standard Error	p	95% Confidence Interval	Bias
Predictor
β0 Intercept	0.328	0.122	0.003	0.135 – 0.621	0.079
β1Treatment [BATA]	0.436	0.221	0.101	0.127 – 1.307	0.084
β2 Age	0.969	0.029	0.291	0.906 – 1.024	0.002
β3 Sex [Male]	0.787	0.427	0.660	0.161 – 2.643	−0.205
β4 Baseline SHAPS	0.949	0.058	0.395	0.821 – 1.075	0.004
β5 Total Sessions	1.262	0.121	0.015	1.128 – 1.573	−0.025
Performance Metrics
R2	0.147
AIC	116.3
BIC	132.8
Brier Score	0.146
C-Index	0.715

The model includes observations from 116 participants. The 95% confidence intervals for the estimates were derived using the percentile method from parametric bootstrapping of 1,000 iterations. Bias refers to the difference between the mean of the bootstrapped estimates and the observed estimate. The odds ratios for continuous variables represent the effect of a one unit increase in the predictor.

Question 4: Internalizing Symptom Trajectories

Symptom Reductions Across the Sample

Per the final unconditional growth models, there were statistically significant linear reductions across the sample for all five internalizing symptom scales. Detailed descriptions of the trajectory characteristics for these secondary clinical outcomes are included in Supplemental Materials. The point estimates for the expected 14-week change scores were as follows: −4.66 [−5.87, −3.44] for the BAI $(z=7.50,p<0.001)$, −10.17 [−11.86, −8.47] for the BDI $(z=11.78,p<0.001)$, −10.85 [−12.90, −8.80] for the PCL-5 $(z=10.37,p<0.001)$, −5.42 [−6.49, −4.35] for the PSS $(z=9.90,p<0.001)$, and −6.56 [−8.22, −4.91] for the PSWQ $(z=7.76,p<0.001)$.

Symptom Reductions in BATA versus MBCT

Per the final conditional growth models, there were no statistically significant differences in the trajectories of internalizing symptom scores between BATA and MBCT (see Figure 8– model summaries reported in the Supplemental Materials). The point estimates for the expected difference in scores between a participant in the parallel interventions at the 14-week mark were as follows (whereby positive values denote higher scores in BATA): −2.03 [−4.68, 0.61] for the BAI $(z=1.51,p=0.132)$, −0.02 [−4.29, 4.24] for the BDI $(z=0.01,p=0.991)$, −2.42 [−7.19, 2.34] for the PCL-5 $(z=1.00,p=0.321)$, 0.35 [−2.26, 2.97] for the PSS $(z=0.26,p=0.792)$, and 2.09 [−3.53, 7.71] for the PSWQ $(z=0.73,p=0.468)$.

Figure 8.

Multiple logistic regression model of SHAPS RCSI. (A) Predicted RCSI rates between BATA and MBCT, across varying total therapy sessions attended, conditioned on female sex, mean age, and mean baseline SHAPS score. Ribbons denote 95% confidence intervals. (B) Model-implied odds ratios of RCSI. The dark, medium, and light purple bars correspond to the 68%, 95%, and 99% confidence intervals respectively. Ratios for continuous variables represent the interquartile range effect (13 years of age, 5.25 points on the SHAPS, and 4 therapy sessions). The adjusted odds of RCSI were 2.29 [0.85, 6.18] times higher in MBCT compared to BATA. The total sessions attended was the only statistically significant predictor. (C) The area under the receiver operating characteristic curve was 0.72 indicating the model had relatively poor predictive ability to classify participants who showed RCSI on the SHAPS.

Discussion

The present study evaluated clinical endpoints from a parallel-arm, randomized trial of psychotherapeutic interventions for a transdiagnostic cohort of adults with clinically significant anhedonia. The trial was designed to test the efficacy of a novel adaptation of BA, called BATA, in comparison with MBCT, a dissimilar psychotherapeutic approach with broad empirical support for internalizing psychopathology. The current analyses addressed four research questions regarding whether BATA and MBCT differed on (1) indicators of treatment feasibility, (2) trajectories of anhedonia symptoms, (3) the proportions of RCSI in anhedonia symptoms, and (4) trajectories of internalizing symptoms.

BATA was designed with theoretical models of reward processing dysfunction as the root cause of anhedonia in mind. Therapeutic instructions included activities intended to sensitize participants to potential rewards in their environment (e.g., repeated pleasant activity scheduling), enhance their subjective experience of pleasure during positive events (e.g., present moment savoring exercises), and develop more accurate mental reward prediction models (e.g., behavioral experimentation with novel tasks). Though BA for depression is well-established (Ekers et al., 2014; Stein et al., 2021), the novel elements of BATA included modules to promote behavioral initiation and reward savoring, streamlining of activity monitoring for people with low motivation, and a sharper emphasis on increasing the frequency of new behaviors. Given this framework, it was expected that BATA would outperform MBCT with respect to improvement of anhedonia symptoms (the primary outcome). Contrary to original expectations, BATA and MBCT performed similarly across all areas of analysis, including treatment feasibility indicators, primary, and secondary clinical outcomes.

Question 1: Treatment Feasibility

The observed attrition rates in our study were 24.6% in BATA and 36.4% in MBCT. Although high, these figures align with meta-analytic estimates of attrition within BA studies for depression (Uphoff et al., 2020) and mindfulness-based interventions overall (Nam & Toneatto, 2016). It is important to interpret our high attrition rates in the context of a demanding schedule of assessments, including up to four MRI scans over four months, and a study cohort selected for their low motivation. To that point, among the variables examined, the strongest predictor of attrition appeared to be baseline anhedonia severity, such that more severely anhedonic participants were marginally more likely to drop out (p = 0.059).

Though the adjusted odds ratio of dropout was 0.49 [0.21, 1.13] times lower in BATA, this did not achieve statistical significance. However, the power to detect reliable differences in these two proportions at a sample size of 116 is only 50%. If we assumed an a priori attrition rate of 29% within the MBCT condition per Nam & Toneatto (2016), the attrition rate within BATA would need to be approximately 14% to achieve 80% statistical power at this sample size. The recent report by Craske et al. (2023) showed similar (non-statistically significant) attrition differences in a transdiagnostic cohort between their novel interventions, PAT (21.4%) and NAT (39.5%), using similar sample sizes.

Homework, Attendance, and TAQ

Therapy homework assignments are a crucial component of CBTs in general and the “main ingredients” of BA therapies. Interim analyses of the current trial demonstrated greater within-person improvements in anhedonia symptoms between therapy sessions in which there was relatively greater adherence to homework assignments, underscoring the importance of homework (Cernasov et al., 2023). The present analyses revealed no significant differences between BATA and MBCT in the proportion of homework assignments completed per either clinician or participant report. However, Hodges-Lehmann estimates⁷ showed a trend towards relatively lesser homework completion within MBCT by approximately 5.5 percentage points for the clinician-reported variable and 4.5 percentage points for the participant-reported variable.

Participants indicated a reasonably high level of satisfaction with the interventions they received. The median TAQ score for both treatment conditions was 9 on a scale in which −14 indicated total dissatisfaction and 14 indicated complete satisfaction. Notably, these estimates did not include the 57.2% of dropouts that did not complete the questionnaire. One interesting observation is that the dispersion of TAQ scores was relatively greater in MBCT (IQR = 7) than in BATA (IQR = 5), suggesting mindfulness garnered more polarizing reactions among participants. Overall, evidence from attrition rates, homework completion, and TAQ scores hints that MBCT was slightly less acceptable for individuals with anhedonia compared to BATA, though larger samples are necessary to confirm this.

Question 2: Anhedonia Trajectories

Anhedonia symptoms, measured via the SHAPS, represented the primary clinical outcome of this trial. A methodological strength of the design was the high volume of repeated assessments that allowed us to test several hypotheses regarding the temporal dynamics of anhedonia symptoms. Mixed effects models were specified in a growth curve fashion wherein the estimation of SHAPS scores was continuously linked over time, as opposed to an ANOVA-type model with timepoints as separate blocks. Growth models are useful because they inform total change as well as how participants arrived at their endpoint levels (e.g., whether improvements are more rapid earlier in the treatment process, when do benefits start to level off, etc.). An advantage of mixed effects models is the accommodation of missing data for intent-to-treat analyses. With restricted maximum likelihood estimation, the assumption is that data are missing at random, meaning missingness may be related to observed variables, but not unobserved variables.

Anhedonia Improvements Across the Sample

A piecewise growth model was fitted estimating a pretreatment slope value, and then separately estimating a value representing the deflection from that trajectory upon initiating treatment. This specification was possible because three measurements of the SHAPS were collected prior to the participants receiving any psychotherapy. In the absence of a waitlist control condition, this added-rate parameterization method for defining the effect of time arguably provides the strongest evidence that observed reductions in SHAPS scores within our study reflected a treatment response directly attributable to psychotherapy processes, rather than simply change over time.

Participants showed significant reductions in SHAPS scores before treatment, at a typical rate of −0.26 [−0.41, −0.11] points per week. This finding is probably reflective of positive expectations associated with enrollment in the study and may be considered part of the placebo effect. The average rate of improvement significantly increased with initiating psychotherapy, indicating treatment-related benefit⁸. The instantaneous slope of change was −0.66 [−0.82, −0.50] points per week at the first session, and then decelerated by 0.01 [0.00, 0.02] points weekly thereafter. Interestingly, more anhedonic participants at the start of therapy tended to show less of a pretreatment expectation effect, in some cases worsening symptoms, as evidenced by a moderate positive correlation (r = 0.58) between these random intercepts and slopes. Conversely, more anhedonic participants tended to show relatively greater symptom reductions upon starting treatment (r = −0.35). This finding either indicates that more symptomatic individuals derive greater benefit from psychotherapy, or that they undergo a generic process of regression to the mean.

A prototypical participant (agnostic to treatment condition) was expected to show a −7.18 [−8.22, −6.15] point reduction in SHAPS scores at the median duration of participation in the trial. This change-score calculation did not factor in modest improvements that occurred pretreatment. While the magnitude of this raw SHAPS reduction is roughly on par with several other trials targeting anhedonia– including other BA therapies (Alsayednasser et al., 2022; Webb et al., 2022), a novel κ-opioid antagonist (Krystal et al., 2020), and ezogabine (Costi et al., 2021) – this stands out as an exceptionally large standardized effect size of pre-to-post change (d = 1.69). To our knowledge the second largest standardized pre-to-post change using the SHAPS was reported in the ADepT trial (d = 1.60) for the novel psychotherapy that heavily utilizes BA (Dunn, Widnall, et al., 2023).

Anhedonia Improvements in BATA versus MBCT

The study did not find evidence to refute the null hypothesis that BATA and MBCT were equally effective at improving anhedonia severity. The growth model-implied 14-week SHAPS score estimate revealed BATA did not outperform MBCT (d = 0.05). Moreover, it could be ruled out with 95% confidence that BATA results in a relative reduction over MBCT of more than 2.25 points (d = 0.53) on this instrument. In other words, any potential advantages of BATA over MBCT are likely smaller than this value. It is important to note that a medium effect size, per Cohen’s guidelines, is unlikely to be clinically relevant in this case. Highlighting this point, we estimated the standard error of the measure for the SHAPS to be 2.33 points – fluctuations below this value are not convincing of a true score change – and the MCID to be at least 3.33 points per the CGI-I scores.

Question 3: Anhedonia RCSI Rates

When considering potential treatments for a client, a clinician’s primary concern is the likelihood that a given intervention approach will result in meaningful benefit, balanced against its potential harms. Defining meaningful benefit in the context of a dimensional symptom such as anhedonia requires statistically principled methods of estimating thresholds on measurement scales that are indicative of clinically relevant improvements (Johnston et al., 2015). Anchor-based methods use external criteria – like clinician judgment, hospitalization status, or other measurement scales – to identify cutoff points (e.g., “scores below [specified value] are unlikely to meet DSM-5 criteria for MDD”). Research reporting such methods for the SHAPS is lacking. In contrast, distribution-based methods identify cutoff points for improbable change due to chance (i.e., minimal detectable change values), or greater probability associated with a non-clinical population than a clinical population.

In the present analyses, we evaluated the proportion of participants demonstrating SHAPS change scores that could be classified as RCSI in anhedonia symptoms. The observed rates of SHAPS RCSI were 14.8% [5.9, 23.7] in BATA and 25.5% [13.9, 37.0] in MBCT, with a non-significant adjusted odds ratio 2.29 [0.85, 6.18] times higher in MBCT than in BATA. The nominally higher RCSI rate in MBCT is surprising but aligns with evidence from feasibility indicators that MBCT may be associated with a more polarizing response than BATA. It may be the case that a portion of participants assigned to MBCT respond exceptionally well to the treatment, while another portion find mindfulness practice tedious or bothersome. This speculation is further supported by the multiple-group conditional growth models for the SHAPS, which showed larger variance in the random effects for the rate of change during treatment within MBCT than in BATA.

The present estimates of RCSI are considerably lower than those reported by Alsayednasser et al. (2022) despite using similar criteria (32% at post-treatment, and 36% at 18-month follow-up). One salient difference between these trials is that Alsayednasser et al. (2022) delivered a greater dose of therapy (up to 20 sessions) over a longer period (5 months), suggesting potential benefits to lengthier treatment. However, in the current trial, length of treatment was flexible, and relatively shorter durations appeared to be preferred (a median of 11 sessions were attended among those who completed treatment). Another notable difference is that our sample had greater baseline anhedonia severity (i.e., treatment onset SHAPS of 36.2 vs 32.1) which made it more difficult to cross the threshold into the non-clinical range. Furthermore,it may be that the transdiagnostic cohort in our sample, rather than an MDD cohort, was harder to treat given a more diverse set of mechanisms maintaining the patient’s anhedonia.

Question 4: Internalizing Symptom Trajectories

Given the transdiagnostic nature of the sample, we analyzed change over time among several scales of internalizing symptoms that commonly co-occur with anhedonia, including global depressive symptomatology, generalized anxiety, post-traumatic stress symptomatology, perceived stress, and uncontrollable worrying. Though linear reductions in symptoms were evident across all five scales over time, we found no significant differences between BATA and MBCT on any of these secondary clinical endpoints.

Limitations

Several limitations of this trial should be noted. First, the analyses reported were not preregistered. To address the threats posed by analytic flexibility and researcher degrees of freedom, we transparently reported several analyses with respect to the SHAPS, including those specified at trial registration. The finding that results were unequivocally non-significant between BATA and MBCT across analytic methods is reassuring.

Second, the absence of a waitlist control condition leaves ambiguity regarding the relative contributions of active treatment components versus expectation effects to the observed symptom benefits. The piecewise model specification for the SHAPS growth curve partially, but incompletely, addresses this concern. That said, the trial was primarily designed to adjudicate superiority of BATA over MBCT, not necessarily to parse placebo effects. In hindsight, given the trending results that MBCT was associated with a more variable treatment response compared to BATA, it would have been beneficial to include a baseline measure of treatment beliefs to further elucidate the role of expectancy in treatment outcomes (Shedden-Mora et al., 2023).

Third, the trial did not evaluate the endurance of symptom reductions or examine differential post-treatment effects in the longer term, such as several months after the conclusion of treatment. Understanding the sustainability of treatment effects over an extended period is crucial for informing clinician recommendations for long-term care.

Fourth, although the SHAPS is often considered a gold-standard instrument for the self-reported assessment of anhedonia, many researchers have recognized that it falls short of painting a comprehensive picture of the construct. The SHAPS focuses on capturing subjective deficits in reward consumption while neglecting crucial components of reward anticipation and reward learning. Furthermore, the inclusion criterion of SHAPS total score ≥ 20, though consistent with the FAST-MAS study on anhedonia (Pizzagalli et al, 2020), reflects the presence of only mild severity, even below the mean value for non-clinical samples. However, this was offset by the inclusion criterion requiring clinician judgment to identify impairment, resulting in the lowest baseline SHAPS score being a 27.

Future Directions

Though our present analyses established equivalent benefits on clinical endpoints between BATA and MBCT, they did not examine what therapeutic components accounted for the observed symptom reductions, or whether the mechanisms of action differed substantially between these treatments. There are two potential scenarios to consider. One possibility is that BATA and MBCT worked equally well because of shared psychotherapeutic factors, such as the alliance between clinician and the client. Indeed, a large body of evidence supports the notion that common factors tend to play a larger role in positive outcomes than specific techniques do in psychotherapy (Ahn & Wampold, 2001; Wampold, 2015). The other possibility is that BATA and MBCT achieved similar benefits via dissimilar processes, for example, activity levels versus increased mindfulness respectively. Future trials should include repeated measures of common factors, as well as of the psychological and/or behavioral processes that are presumed to unfold over time, and empirically test the contributions of specific ingredients in psychotherapy.

In summary, this trial evaluated the efficacy of BATA relative to MBCT for treating anhedonia in a transdiagnostic cohort. Results indicated equivalent improvements on the primary clinical endpoint of anhedonia symptoms as well as on secondary clinical endpoints of several internalizing symptoms. Future reports will examine the possibility of differential target engagement between treatment conditions across fMRI and other neurocognitive measures of reward processing collected during this trial, and the extent to which these variables relate to anhedonia symptom reductions.

Figure 9.

(Top Row) Trajectories of Beck Anxiety Inventory (BAI), Beck Depression Inventory (BDI), PTSD Checklist for DSM-5 (PCL-5), Perceived Stress Scale (PSS), and Penn State Worry Questionnaire (PSWQ) raw scores over time. Each line represents an individual participant. (Bottom Row) Conditional growth model-implied values. Thin gray lines represent predicted values for individual participants using a combination of fixed and random effects. Colored lines represent the expected treatment effects in the population – calculated using fixed effects, with age and total sessions held at their mean values, and sex specified as female. Ribbons denote the 95% confidence intervals for these population estimates.

Highlights.

• Behavioral Activation Treatment for Anhedonia (BATA), a novel psychotherapy, was no more effective than Mindfulness-Based Cognitive Therapy (MBCT) at improving symptom severity for a transdiagnostic cohort of adults with anhedonia.

• Across BATA and MBCT, the estimated treatment-related improvement at the median duration of participation was −7.18 [−8.22, −6.15] points on the Snaith-Hamilton Pleasure Scale, corresponding with a standardized effect size of d=1.69.

• BATA and MBCT did not differ with respect to any secondary outcome measures of internalizing symptoms of psychopathology.

• Trends in attrition rates and other indices of treatment feasibility suggested BATA may be slightly more acceptable to adults with anhedonia than MBCT.