Sixty participants were randomised (Fig. 1), per our protocol target. Twenty-five participants were randomised pre-COVID (recruitment period March 2019–March 2020) and 35 during COVID (June 2020–January 2021). Due to the re-randomisation design, for the last recruitment wave (Wave 7, February to March 2021), participants first allocated to WL (n = 4) based on the predetermined randomisation sequence were immediately re-randomised to either AP or AE by opening the second randomisation envelope. This was done to ensure all participants could receive an intervention. This resulted in 22 participants each in AP (re-randomised n = 29), AE (re-randomised n = 28), and 16 WL.

Fig. 1

CONSORT flow chart of recruitment into the MAPLES study. AP = Activity Planning group, AE = Activity Engagement group, WL = Waitlist Controls

AP participants began their allocated intervention at an average of 21.07 days (SD = 12.55; median = 16 days) following their baseline assessment, compared to 17.64 days (SD = 8.89, median = 15 days) for AE participants. There were 3 participants (n = 2 AP, n = 1 AE) for whom circumstances dictated significantly longer delays been initial assessment and group participation. Their delays were 119, 166, and 140 days, respectively, and are not included in the mean values to avoid potentially giving a misleading impression of the majority pattern.

The main reason for exclusion was not meeting study eligibility criteria (n = 37 of 74; see Additional File 2: Table S1 for detailed reasons for exclusion).

Participant demographics are presented in Table 2.

Table 2 MAPLES participant characteristicsPrimary objective—feasibility outcomes

Due to COVID-19 requiring completely remote research, feasibility data were summarised for in-person and remote sessions separately.

Recruitment outcomes

Recruitment targets were met only when including re-randomised WL participants; otherwise, minimum recruitment targets were only met for Wave 1 and Wave 4 (Table 3).

Table 3 Total number of new participants recruited per wave, with participants re-randomised to an active intervention (Activity Planning or Activity Engagement) included in brackets. Recruitment for solely online adaptations of the active interventions began in Wave 4

Before the first UK COVID-19 lockdown (March 23 2020), the majority of participants recruited were from NHS ABI services (19 of 25). During COVID-19, participants were mostly recruited from ABI charities (22 of 35; see Additional File 2: Table S2 for eligibility per recruitment source). Self-referral via social media was the most efficient referral route (14 of 21 screened randomised; 66.7% eligible).

Study withdrawal

When including re-randomised participants, attrition was less than 20% across the three arms (Table 4). Not including those re-randomised, withdrawal rates were 13.6%, 9.1%, and 18.8% in AP, AE, and WL, respectively.

Table 4 Summary of attrition between the Activity Planning and Activity Engagement groups

The most common reason for withdrawal was difficulties travelling to the study location (n = 3; 1 per trial arm), followed by expressive aphasia affecting group participation (n = 1 AP and n = 1 AE) (see Fig. 1 for all reasons).

Intervention session attendance

Including participants who withdrew, average attendance did not differ between the AP and AE group (t =  − 0.38, p = 0.70; Table 5). However, median attendance rose from 7 to 8 when delivered online. Average attendance did not differ in-person or online for the AP (t =  − 1.25, p = 0.22) or AE group (t =  − 0.55, p = 0.59). Session attendance per wave and reasons for non-attendance are in Additional File 2: Figures S1 and S2.

Table 5 Summary of average attendance between the Activity Planning and Activity Engagement groups, including those who withdrewAcceptability outcomesCredibility and expectations of interventions

Baseline CEQ summary data are in Additional File 2: Table S3. In terms of how logical each group was perceived to be, the AP group had a mean of 7.24 (SD = 1.62) compared to the AE group (M = 6.61, SD = 2.33). Comparable means were found for perceptions of how effective each group would be in increasing activity levels (AP M = 7.00 [SD = 1.73]; AE M = 6.32 [SD = 2.45]), in recommending the group to a friend (AP M = 6.86 [SD = 1.77]; AE M = 6.68 [SD = 2.39]), and in reporting feeling that their activity levels would improve (AP M = 6.45 [SD = 2.13]; AE M = 5.89 [SD = 2.36]).

Post-study questionnaire—quantitative data

PSQ responses were positive for both groups (Table 6). Irrespective of mode of delivery, AP and AE were rated as similarly enjoyable (t = 0.27, p = 0.79) and helpful (t = 1.99, p = 0.05). The AP group was rated as similarly helpful (t = 0.43, p = 0.67) and enjoyable (t = 0.14, p = 0.89), whether in-person or online. The AE group was similarly helpful online and in-person (t =  − 0.56, p = 0.58), though rated as more enjoyable online (t =  − 2.54, p = 0.02).

Table 6 Summary of responses to Post-Study Questionnaire (PSQ) per group for participants who attended in-person versus onlineBarriers to attendance

All participants (including those who withdrew) were asked whether there were any barriers to attending study sessions, even if they were able to overcome these barriers. Average number of barriers to attending the AP (M = 2.84) and AE (M = 2.04) groups did not significantly differ (t = 1.40, p = 0.17). There appeared to be variations in barriers present (Fig. 2), though only fatigue being less frequently reported as a barrier in AE was statistically significant (t = 2.99, p < 0.01).

Fig. 2

Frequency of each barrier to attendance across all participants in the AP group (left) and AE group (right)

The number of barriers in the AP group was not statistically different online compared to in-person (t =  − 1.79, p = 0.09). There were no statistically significant differences in number of barriers for AE online versus in-person (t =  − 2.10, p = 0.05). Regardless of group, attending online resulted in more reported attention, fatigue, technology and organisational barriers (ts =  − 2.89 to − 3.78, ps < 0.01). For AP specifically, technology, fatigue, and attention barriers were more frequently reported online (ts =  − 2.58 to − 4.24, ps < 0.05). For AE specifically, organisational barriers were more frequently reported online (t =  − 2.28, p = 0.04).

Acceptability outcomes—qualitative feedback on groups

PSQ and exit interview qualitative data were used to explore perspectives of acceptability, specifically aspects that positively or negatively contributed to helpfulness or enjoyableness, of the groups and overall participation in the study. Table 7 synthesises perceived strengths of the groups or areas for improvement, with example quotes in Additional File 2: Table S4. Recommendations presented below should be considered within any local context of recreating the groups described here. For space, in-depth qualitative data analysis of specific group experiences will be reported elsewhere.

Table 7 Recommendations for important areas to maintain or modifications of the groups, based on qualitative dataSupporting participants with aphasia

To evaluate the potential suitability of the groups for people with language difficulties, qualitative data from participants with aphasia are presented in Additional File 2: Document S4 and Table S5. In brief, participants with fluent aphasia seemed to have positive experiences within groups; however, for those with non-fluent aphasia, individual sessions were preferred.

Secondary objective—clinical outcomes

The secondary objective was to provide estimates on the primary efficacy outcome measure and sample size determination for a subsequent trial. We summarise missing data, AP and AE fidelity assessment results, and efficacy estimations on the primary efficacy outcome measure between groups.

Fidelity assessment results

Both the AP and the AE groups were delivered as intended, with percent fidelity estimates of 95.06% (SD = 6.59, range 80–100%) and 99.17% (SD = 2.89, range 90–100%), respectively.

Missing data and acceptability of questionnaires

Generally, participants who did not withdraw had complete data.

Reasons for non-completion included discomfort with the measure content (n = 1 SHAPS; AE group), providing opinions about rehabilitation staff (n = 1 MOT-Q; AE group), due to technical errors with online data collection (n = 1 BMQ-S AP group; n = 1 MOM-POPS AP group) or accidental omission from the questionnaire battery (n = 1 MOT-Q WL group), or for unknown reasons (n = 1 MOT-Q WL group; n = 1 SCS AE group).

Baseline outcome measures

Randomisation produced well-matched groups on study variables (see Table 8).

Table 8 Scores on all outcome measures at baseline by groupEfficacy of intervention on primary efficacy outcome measure

The unadjusted mean difference on the BADS from Time 1 to Time 2 was − 10.41 (95% CI − 19.67 to − 1.27) for the AP group, compared to − 7.35 (95% CI − 14.84 to 0.14) in the AE group and − 1.51 (95% CI − 13.78 to 10.76) for the WL group, indicating positive trend toward improvements in activity levels in the AP and AE groups.

A mixed-effects linear model was conducted using total BADS scores via the lmerTest R package [55]. Participants were initially modelled as a random effect. However, adding BADS baseline scores as a covariate resulted in model singularity. As participants were recruited in “waves” where the AP, AE, and WL groups ran in parallel, we instead considered cohort effects because (a) this allowed accounting for COVID-19 onset and resultant fluctuating restrictions on socialising, mood, and activity levels; (b) COVID-19 dramatically affected recruitment, and (c) in the last wave, those first sent to WL were immediately re-randomised. Whilst randomisation should protect against potential systematic differences, study wave (Waves 1–7) was modelled as a random effect to account for this.

Missing data was estimated using restricted maximum likelihood estimation. F-statistics with effective degrees of freedom were estimated using Satterthwaite’s method in lmerTest. For each model, adjusted intraclass correlation coefficients were used to estimate the amount of variance attributable to the random effects [56]. Tukey-adjusted post-hoc tests were conducted using estimated marginal means. Full mixed-effects model results (including b-values and standard errors) are shown in Additional File 2: Table S6.

On the BADS (Fig. 3), there was a main effect of Time (Satterthwaite’s F2,167 = 3.82, p < 0.05) and Group (Satterthwaite’s F2,170 = 3.33, p < 0.05) and baseline scores (Satterthwaite’s F2,172 = 479.43, p < 0.001), but no Time by Group interaction (Satterthwaite’s F3,167 = 0.98, p = 0.39). Random effects estimates were low (14.44, SD = 3.80, χ2 = 5.54, p = 0.02, ICCadj = 0.07), indicating that variation in participant intercept due to study wave was likely minor. In post-hoc tests, only AP participants demonstrated improvements on BADS scores from Time 1 to Time 2 (t =  − 2.76, p = 0.01) and to Time 3 (t =  − 2.68, p = 0.02). Time 2 and Time 3 scores did not differ within the AP group (t = 0.10, p = 0.99). BADS scores in the AE group did not differ between Time 1 to Time 2 (t =  − 1.76, p = 0.18) or to Time 3 (t =  − 1.12, p = 0.50). WL participants showed no BADS improvements from Time 1 to Time 2 (t =  − 0.18, p = 0.98). In summary, significant gains in BADS scores were only observed in participants randomised to the AP group and these improvements were well maintained at Time 3.

Fig. 3

Visualisation of individual change between the three groups on the Behavioural Activation for Depression Scale (BADS) across each time point. Visualisations include those who were re-randomised into either the AP or AE group. Higher BADS scores represent greater activity engagement (i.e., improvement)

Exploratory analyses—secondary efficacy outcome measures

To descriptively examine potential effects between groups across study measures, a summary of mean differences with 95% CIs across all study outcome measures is shown in Table 9. A visualisation of point estimates is in Additional File 2: Figure S3.

Table 9 Unadjusted mean differences on study outcome measures from baseline to post-intervention

To examine potential statistical effects on secondary efficacy outcome measures, exploratory analyses on the HADS, BMQ-S, the SCS, and IU-SF were conducted. These were selected based on MCID results (presented below). For ease of exposition, they are reported here because the analysis method was identical to that of the BADS (above).

There was a significant time by group interaction on the HADS-Depression subscale (F3,167 = 5.96, p < 0.001; VC = 0.31, SD = 0.56, ICCadj = 0.06). In post hoc contrasts, both AP (t =  − 4.30, p < 0.001) and AE (t =  − 4.60, p < 0.001) participants demonstrated reduced HADS-Depression scores versus WL at Time 2, whilst AP and AE participants did not differ at Time 2 (t = 0.41, p = 0.91) nor at Time 3 (t =  − 0.84, p = 0.68). In summary, both intervention groups were equally effective in reducing HADS-Depression scores (relative to Waitlist) and these improvements were maintained from Time 2 to Time 3 for both AP (t = 0.01, p = 1.00) and AE participants (t =  − 1.25, p = 0.43).

For HADS-Anxiety scores, there was a significant Time by Group interaction (F3,167 = 4.04, p < 0.01; VC = 0.15, SD = 0.38, ICCadj = 0.03). At Time 2, AP participants showed significantly greater reductions in anxiety than both AE (t =  − 3.98, p < 0.001) and WL participants (t =  − 4.13, p < 0.001). AE did not differ from WL (t =  − 0.78, p = 0.72). HADS-Anxiety reductions within AP were maintained from Time 2 to Time 3 (t =  − 0.83, p = 0.69) and were still lower than AE (t =  − 2.55, p = 0.03). In summary, only Activity Planning group participants showed significant reductions in HADS-Anxiety scores that were well maintained at Time 3.

There was a significant Time by Group interaction on the BMQ-S (F3,164 = 3.66, p < 0.05; VC = 2.98, SD = 1.73, ICCadj = 0.06). Only AP participants demonstrated reductions in motivation difficulties versus AE (t =  − 2.91, p < 0.05) and WL (t =  − 3.99, p < 0.001) at Time 2 and compared to AE at Time 3 (t =  − 2.59, p < 0.05). AE and WL participants did not differ at Time 2 (t =  − 1.54, p = 0.27). BMQ-S reductions in AP were maintained from Time 2 to Time 3 (t = 0.49, p = 0.87). In summary, again only AP group participants showed significant improvements in motivation which were well maintained at time 3. Changes in HADS-Depression and Anxiety, and in BMQ-S scores, are illustrated in Fig. 4.

Fig. 4

Changes in depression, anxiety, and motivation between the three groups at baseline, post-group, and 1-month follow-up. AP Activity Planning, AE Activity Engagement, WL Waitlist Controls, HADS Hospital Anxiety and Depression Scales, BMQ-S Brain Injury Rehabilitation Trust Motivation Questionnaire

There were no statistically significant differences between groups on the Perceived Constraints (p = 0.77) and Perceived Mastery (p = 0.17) SCS subscales.

On the IU-SF Inhibitory subscale, there was a significant Time by Group interaction (F3,167 = 3.27, p = 0.02, VC = 0.27, SD = 0.53, ICCadj = 0.03). At Time 2, AP participants had greater Inhibitory Anxiety reductions versus AE (t =  − 2.64, p = 0.02) and WL (t =  − 3.86, p < 0.001). AE and WL participants did not differ from each other (t =  − 1.63, p = 0.24). AP participants maintained reductions from Time 2 to Time 3 (t =  − 0.30, p = 0.95). There was no interaction on the IU-SF Prospective subscale. To summarise, again only AP participants showed significant reductions in intolerance of uncertainty Inhibitory Anxiety that were well maintained at Time 3. There were no group effects on the IU-Prospective Anxiety items.

Taken together, results suggest that participants in the AP group had a wider range of improvements in activity levels, depression and anxiety symptoms, difficulties with motivation, and the inhibiting effects of uncertainty and that these improvements persisted for at least a month post-intervention. There were also significant reductions in depression scores in the AE group, relative to Waitlist, that were well-maintained. These outcomes did not seem affected by study wave, though further modelling for random slopes in a larger trial would be beneficial.

Caution is needed in exploratory analyses due to multiple comparisons. For this reason, we analysed secondary efficacy measures based on MCID estimates (see below). In addition, the pattern of results is important. Where a set of conceptually related measures (anxiety, low mood, low motivation) show consistent patterns across the groups, the likelihood of this reflecting a true underlying pattern is increased.

Estimation of the Minimal Clinically Important Difference (MCID)

The MCID for all outcome measures was calculated using the standard error of measurement (SEM):

Sigma (\(\sigma\)) was Time 1 standard deviation and the reliability (r) was Time 1 internal consistency. Time 1 to Time 2 data were used to calculate percent estimates across groups, including those re-randomised. For the BADS, the MCID was 8.28. Missing data at Time 2 was imputed using predictive mean matching (PMM) across 5 imputed datasets via the mice R package [57].

BADS MCID change from Time 1 to Time 2 between groups using complete cases (including those re-randomised) are in Fig. 5, where 58.33% of AP participants showed BADS improvements at or above this minimally clinically important level, versus 52.17% of AE participants and 30.77% of WL participants. Across 5 imputed datasets, AP participants had a MCID improvement range of 51.72–65.52% versus AE (range 42.86–57.14%) and WL (range 25–37.5%).

Fig. 5

Visualisations of participant-level change scores from Time 1 to Time 2 on the Behavioural Activation for Depression Scale (BADS), not including those who withdrew. Dashed lines indicate the minimal clinically important difference (8.28) in either direction. Higher change scores indicate greater improvements in activity levels. Percentages indicate numbers of those within in each group who made MCID improvements on the BADS from Time 1 to Time 2

As shown in Fig. 5, there were also MCID on this primary efficacy outcome measure in the opposite direction in all groups (AP 25%; imputed range 20.69–34.48%; AE 17.39%; imputed range 17.86–32.14%; WL 38.46% imputed range 31.25–50.0%).

Exploratory analysis–secondary outcome measures MCID

Changes in MCID on secondary outcome measures were explored to determine which variables may be useful for a definitive trial. HADS-Depression and HADS-Anxiety MCID changes are presented in Fig. 6. This shows that 54.17% of AP participants demonstrated clinically meaningful reductions on the HADS-Depression scale (versus 56.52% in AE and 7.69% in WL). In terms of HADS-Anxiety, 54.17% of AP participants demonstrated clinically meaningful reductions compared to 20.83% AE and 15.38% WL.

Fig. 6

Visualisations of participant-level change scores from Time 1 to Time 2 on the Hospital Anxiety and Depression Scale (HADS), not including those who withdrew. HADS-Depression change scores are on the top row, and HADS-Anxiety scores are in the bottom row. Dashed lines indicate the minimal clinically important difference (HADS-Depression = 1.96, HADS-Anxiety = 1.83) in either direction. Lower change scores indicate greater reductions in depression and anxiety. Percentages indicate numbers of those within in each group who made MCID improvements on either measure from Time 1 to Time 2

Across five imputed datasets, AP participants had a MCID improvement range of 51.72–58.62%, versus AE (range 57.14–64.28%), and WL (18.75% across all five imputations) on HADS-Depression. For HADS-Anxiety, the imputed range was 51.72–55.17% for AP, versus AE (range 21.42–28.57%), and WL (range 18.75–31.25%).

Complete case MCID improvements on study measures from Time 1 to Time 2 are in Additional File 2: Table S7. The most responsive measures were the HADS, the BMQ-S, the SCS, and the IU-SF.

Sample size analysis for future trials

BADS Cohen’s d estimates were obtained using estimated marginal means via the emmeans R package [58], adjusted for baseline BADS scores. Given its relevance to BA, HADS-Depression effect sizes were also estimated.

At Time 2, BADS effect size for AP vs AE was 0.48 (95% CI =  − 0.12 to 1.09) and 0.79 (95% CI 0.08 to 1.50) against WL. The effect size for AE versus WL was 0.31 (95% CI =  − 0.40 to 1.02).

At Time 2, HADS-Depression effect sizes for AP versus AE were 0.18 (95% CI =  − 0.42 to 0.79) and − 1.63 (95% CI − 2.36 to − 0.89) against WL. The effect size for AE versus WL was − 1.82 (95% CI =  − 2.56 to − 1.07).

Power analyses were conducted using raw BADS and HADS-D estimates, and a range of estimates (range 0.30 to 1.00, based on the Oates et al. [15] systematic review) to account for likely effect size fluctuations in future trials. Power analyses (alpha = 0.05, beta = 0.80) were two-tailed and were conducted for all possible primary comparisons of interest (e.g., detecting effects between only AP and AE). Depending on the primary comparison of interest and outcome measure used, estimates varied widely (Table 10). For example, detecting a difference in BADS scores between AP and WL groups would require 26 participants per group based on raw estimates. Similarly, detecting a difference in HADS-D scores between either AP or AE to WL would require as little as 6 participants per group based on raw estimates; however, detecting a difference in HADS-D scores between AP and AE groups would require 485 participants per group.

Table 10 Sample size calculations based on effect size estimates using trial data on the BADS and HADS-D, as well as a range of possible effect size estimates. Estimates do not include expected attrition. Power analyses were conducted based on the primary two-group comparison of interest. An alpha of 0.05 and a beta of 0.80 were used. BADS = Behavioural Activation for Depression Scale; HADS-D = Hospital Anxiety and Depression Scale-DepressionHarms

Potential harms (worsening HADS scores or changes in reporting of suicidality) of either BA group were explored. Full data are presented in Additional File 2: Tables S8 and S9. In brief, even when accounting for missing data [59], proportion of those improved in terms of HADS-D scores were greater in both the AP and AE groups. There appeared to be no evidence that participation influenced reporting of suicidal ideation.

Challenges in trial implementation

As part of the trial, a formal steering committee consisting of clinicians, researchers, and an ABI survivor was formed to oversee recruitment and study progress. An overview of challenges in implementing the trial from the steering committee’s perspective is presented in Additional File 2: Document S5. In brief, challenges included efficient recruitment through NHS services relative to ABI charities, online intervention delivery resulting in less fatigue due to travel but affecting group social interactions, ABI participants experiencing difficulties with recalling specific group experiences during the qualitative interview at Time 3, and COVID-19 affecting the range of activities participants could partake in throughout the trial.