Transdiagnostic Ecological Momentary Intervention for Improving Self-Esteem in Youth Exposed to Childhood Adversity

This randomized clinical trial investigates the efficacy of a transdiagnostic ecological momentary intervention for improving self-esteem in youth with low self-esteem and exposure to childhood adversity.

2) Adversity: a. Childhood trauma: Prior exposure to at least one form of childhood trauma defined as moderate or severe physical (score ≥ 10), sexual (score ≥ 8) and/or emotional (score ≥ 13) abuse, emotional (score ≥ 15), and/or physical (score ≥ 10) neglect, according to established severity categories of the Childhood Trauma Questionnaire (CTQ) [1][2][3] , and/or b.Peer bullying: Exposure to moderate or severe peer bullying, measured with the Retrospective Bullying Questionnaire (RBQ) (score of frequency of bullying in one or more ways "sometimes" or more often and/or classified the experience as "quite serious" or "extremely serious") 4 , and/or c.Parental conflict: A score of moderate or severe parental conflict, measured with the Childhood Experiences of Care and Abuse Questionnaire (CECA.Q) section Parental Conflict (frequency score of "regularly" or "often" and/or a severity score of "serious" or "violence") 5 .3) Self-esteem below average (measured with the Rosenberg Self-Esteem Scale (RSES) using the cut-off (RSES total score <26) applied in previous research on psychological interventions for reducing self-esteem 6,7 .4) Willingness to participate in the SELFIE intervention.5) Ability to give written informed consent.6) Parental consent for minors.Exclusion criteria 1) Insufficient command of Dutch 2) Psychiatric symptoms due to an organic cause eMethods 1. SELFIE interven on descrip on The SELFIE interven on was delivered by trained psychologists within a 6-week period in addi on to treatment as usual (TAU) to individuals allocated to the experimental condi on.The one-day training for SELFIE therapists (mental health professionals working at clinical sites) consisted of delivering background informa on on the SELFIE trial (i.e.study protocol, theore cal framework of self-esteem), making oneself familiar with the interven on guideline, prac cing exercises, and learning how to apply the smartphone applica on when delivering the interven on.When delivering the interven on, regular inter-and supervision was offered by a clinical psychologist.
The interven on consisted of three face-to-face sessions (each for around 60 minutes), delivered by SELFIE therapists on a fortnightly basis, , three email contacts with SELFIE therapists (again, delivered every other week as shown in eTable 2), and an EMI administered through a smartphone-based app (i.e., the PsyMate® app) for adap ve real-me and real-world transfer of interven on components tailored to person, moment, and context, delivered over the 6-week interven on period.Face-to-face sessions and emails introduced principles and techniques to be con nued by the par cipant by use of the EMI in the week therea er.Specifically, the EMI translated the training from face-to-face sessions into individuals' daily lives based on three types of delivery schemes (i.e., enhancing tasks, consolida ng tasks, interac ve tasks (see below for further detail)).Contact with SELFIE therapists offered an opportunity to reflect on progress and discuss any issues par cipants faced with the EMI.Due to the COVID-19 pandemic, most of the face-to-face sessions were offered through a secure and encrypted video conferencing system.The interven on was based on principles of EMIs [8][9][10][11][12][13][14][15] , and a guided self-help approach using principles of cogni ve-behavioral therapy (CBT), aimed at modifying cogni ve bias (selec ve percep on and selec ve memory) inherent to nega ve self-esteem (by ac vely searching for posi ve informa on and other experiences), building a compe ve posi ve self-esteem, and developing and prac cing a new behavioral repertoire guided by therapists using modeling and shaping as addi onal important therapeu c techniques 16,17 .Delivering the interven on in individuals' daily lives, and enabling youth to benefit from this interven on in a given moment and context, when most needed (e.g. in moments of low self-esteem) was the key goal of the 6-week SELFIE interven on.
Therefore, in the first introductory session, par cipants either received a study smartphone with the app already installed or were asked to install it on their smartphone by the SELFIE therapist, who explained the SELFIE interven on in detail and asked the par cipant to complete interven on exercises on the app to address the self-selected goals the par cipant wanted to work on in the 6-week interven on period.The app offered par cipants 'enhancing', 'consolida ng', and 'interac ve' tasks (see Table 2 in main manuscript).
In enhancing tasks, new interven on components were introduced and prac ced, some of which were modified and extended over the interven on period.Consolida ng tasks asked par cipants to prac ce previously learned components of enhancing tasks on a daily basis.For these tasks, par cipants were reminded by the app between 1-3 mes per day (varying by interven on week).During the interven on period, Ecological Momentary Assessment (EMA), a structured dairy technique, was used to assess momentary self-esteem, affect, and pleasantness of ac vi es and events, six mes a day, on days 3, 4, and 5 on each of the six interven on weeks using a me-based design with stra fied random sampling (i.e., with EMAs scheduled at random within set blocks of me) to allow for interac ve tasks.
Interac ve tasks were provided based on their EMA ra ngs of (posi ve and nega ve) affect, momentary self-esteem, and pleasantness of ac vi es and events.The threshold for triggering interac ve tasks was either high (opera onalized as a ra ng of 4 or higher on a 7-point Likert scale ranging from 1 to 7) or low (opera onalized as a ra ng of 3 or lower on a 7-point Likert scale ranging from 1 to 7) posi ve affect, momentary self-esteem, and/or pleasantness of ac vi es, on items of established and validated EMA measures (with a ra ng of 4 or higher being used as the triggering threshold, as this reflects the midpoint on the 7-point Likert scale, with any score equal to or above indexing high ra ngs on this scale).When par cipants provided high ra ngs, they were directed to the exercise 'posi ve datalog' and asked to add more successes (week 1) and/or posi ve quali es (week 3) to their datalog.When par cipants provided low ra ngs (a score of 3 or lower), they were directed to the exercise 'posi ve datalog' to show them all the successes they had previously entered.In the fi h week of the interven on period, par cipants were addi onally asked whether they had received cri cism since the last beep.If they answered affirma ve, they were directed to the exercise 'a cri cal look at cri cism' to offer them a technique to deal with cri cism.
Par cipants could discon nue the interven on at any me upon request without nega ve consequences.A er comple on of the interven on period, or at drop-out, par cipants deleted the app on their phones and had no longer access to the app.
The Dutch version of the 25-item CTQ enquires about five types of childhood trauma (physical, emotional and sexual abuse, and physical and emotional neglect).All five types of trauma are covered with five items, rated on a 5-point Likert scale (1=Never true; 5=Very often true).Cut-off scores for including participants on childhood trauma were as follows: a score of ≥ 10 on physical abuse, a score of ≥ 8 on sexual abuse, a score of ≥ 13 on emotional abuse, a score of ≥ 15 on emotional neglect, and/or a score of ≥ 10 on physical neglect, according to established severity categories of the CTQ [1][2][3] ).
The RBQ consisted of nine items that assess physical, verbal and indirect bullying in primary and secondary schools.Participants were included in the study when they reported the frequency of bullying in one or more ways (as "sometimes" or more often and/or classified the experience as "quite serious" or "extremely serious") on the RBQ 4 .
The CECA.Q, section Parental Conflict is an interview-based measure to assess frequency and severity of parental conflict 5 .Cut-off scores for this study were frequency scores of "regularly" or "often" and/or a severity score of "serious" or "violence".
Moreover, the inclusion criterion of low self-esteem was operationalized as a total score of <26 on the RSES 6,7,[16][17][18] based on previous research on psychological interventions for reducing self-esteem 6,7 .The RSES is a widely used 10-item self-report measure to assess global self-esteem with good reliability and validity 6,19 .All ten items were rated on a 4-point Likert scale (1=strongly agree; 4=strongly disagree).
The Cronbach's alpha of the RSES at screening was .78.
Data on basic socio-demographic characteristics were collected on age, gender, level of education, and employment status, and self-ascribed ethnicity and race (i.e., Moroccan, Turkish, Surinamese, other) to allow for a basic epidemiological characterization of the sample based on evidence on the basic risk factor epidemiology of mental disorders.

Compliance with, fidelity to, and acceptance of the SELFIE interven on
Fidelity to the SELFIE intervention protocol was based on a) SELFIE therapists' personal ratings of core components delivered in the sessions, and b) independent ratings of core components delivered in the sessions of a selection (n=72 sessions, 31,7% of all delivered sessions) of audio recordings of the face-to-face sessions (using maximum variation sampling to include audio recordings of all three sessions, and all SELFIE therapists).In addition, user experience and acceptance of, as well as satisfaction with the SELFIE intervention was assessed using a self-report measure that included 31 items rated on a 7-point scale, ranging from a rating of 1 referring to 'not at all', via a rating of 4 referring to 'moderate', to a rating of 7 referring to 'very much'.Also, the Working Alliance Inventory (WAI) 20 was completed by the participant (WAI-P) and the SELFIE therapist providing the SELFIE intervention (WAI-T) including 12 items rated on a 5-point Likert scale.We used the mean score of the 12 items for patient and therapist ratings to obtain measures of participant-and therapist-rated working alliance.

Outcomes
Blinded assessors collected data on outcomes before randomization (at 'baseline'), at the end of the 6-week intervention period ('post-intervention'), and at 6-month follow-up ('follow-up').Participants aged 16 years or older were compensated for their time and travel expenses were fully reimbursed.

Primary outcome
The primary outcome was global self-esteem, measured with the RSES 18 , which is a widely used measure to assess global self-esteem with good reliability and validity 6,19 .The RSES consists of ten items rated on a 4-point Likert scale ranging from 'strongly agree' to 'strongly disagree'.The level of global self-esteem, operationalized as the total score of the RSES, was compared between the experimental and the control condition at post-intervention and 6-month follow-up.The reliability estimate of the RSES total score in the current study was acceptable (Cronbach's α = .78).

Secondary outcomes
Secondary outcomes were positive/negative self-esteem, schematic beliefs of self, emotional wellbeing, psychological distress, general psychopathology, clinical symptoms, subjective quality of life, and functioning (assessed with non-EMA measures) as well as momentary self-esteem, momentary affect, and momentary resilience (assessed with EMA measures), which all reflect distinct outcomes of interest as an intervention target of the SELFIE intervention to allow for detecting signals of efficacy and, thereby, inform further optimization of the intervention and subsequent research.

Non-EMA measures
Positive/negative self-esteem.Positive and negative self-esteem were measured with the Self-Esteem Rating Scale (SERS), which is a 20-item rating scale to assess these two dimensions of self-esteem separately with good reliability and validity 21 .All items were rated on a 7-point Likert scale.The reliability estimate of the 10 items to measure positive self-esteem was .86,while for the 10 items to measure negative self-esteem the Chronbach's α was .84.

Schematic beliefs of self
The Brief Core Schema Scale (BCSS) was used as an established 12-item measure of positive and negative schematic beliefs of self 22 .The internal consistency of the positiveand negative-self schema scales were .76 and .71,respectively.
Emotional well-being.Emotional well-being was assessed using the Positive and Negative Affect Scale (PANAS) 23 , consisting of 20 items.The internal consistency of the positive and negative affect scales was .78 and .90,respectively.
Psychological distress.Psychological distress was measured with the Kessler Psychological Distress Scale (K10), which is a widely used and well-validated in youth 24,25 .The K10 is a 10-item questionnaire to assess psychological distress in the last month on a scale from 1='never' to 5='always'.We used the sum score for this measure.The internal consistency of the K10 was Cronbach's α =.86.
General psychopathology.The revised Symptom Checklist (SCL-90-R) was used as a reliable and valid measure to assess general psychopathology in youth 25,26 .The measure consisted of 90 items, all rated on a 5-point scale.The total sum score of the SCL-90-R was used.The Alpha value for internal consistency of this measure was .97.
Clinical symptoms.We used the 24-item version of the Brief Psychiatric Rating Scale (BPRS) 27,28 as a validated interviewer measure to assess clinical symptoms of psychopathology in youth 25 .All items were rated on a 7-point scale.The intensity of psychopathological symptoms was indicated by the BPRS total score (range 24 to 168).We conducted five reliability meetings in order to assess the interrater reliability for the BPRS scores.The Intra-Class Correlation (ICC) coefficients ranged from .82 to .96.

Subjective quality of life. Subjective quality of life was measured with the World Health Organisation
Quality of Life Instrument-Brief (WHOQOL-BREF) 25,29 .Using this questionnaire, mean scores of four domains were measured (physical health, psychological health, social relationships and environment), with reliability estimates of the domains being .62,.73,.61 and .72,respectively.
Functioning.The Social and Occupational Functioning Assessment Scale (SOFAS) 30 and the Global Assessment of Functioning (GAF) scale 31 were used as a well-validated measure of functioning in youth
For the analysis, we used the overall level of functioning rated by researchers on a scale of 0 to 100.
Momentary self-esteem.Momentary self-esteem was assessed with the following four items 38,39 : "I like myself", "I am ashamed of myself", "I am satisfied with myself", and "I doubt myself".The internal consistency of these items was α = .90 Momentary affect.Momentary affect was measured using a 5-item EMA measure for assessing negative affect and a 4-item EMA measure of positive affect [32][33][34] .Positive affect was measured with the following items: "I feel happy", "I feel satisfied", "I feel relaxed", "I feel cheerful", and "I can accept my feelings", while negative affect was assessed using the items: "I feel anxious", "I feel down", "I feel uneasy", "I feel lonely", "I feel insecure".Alpha values for internal consistency were .94and .92,respectively.
Momentary resilience.Momentary resilience was assessed with the EMA and was operationalized as negative affective recovery in response to momentary stress as well as positive affective recovery in response to event-related stress in daily life 33-35, 37, 40 .

eMethods 3. Sta s cal analysis
Based on a pre-specified and published statistical analysis plan (SAP) 41 , statistical analyses were performed in Stata version 16 42 , while being blinded to random allocation to experimental and control condition.The following hypotheses were tested: First, we hypothesized that, compared with the control condition (care as usual (CAU)), self-esteem will, on average, be higher in the experimental condition (SELFIE+CAU) than in the control condition across post-intervention and 6-month follow-up (primary outcome).Second, we hypothesized that, compared with the control condition (CAU), momentary self-esteem, positive self-esteem, positive schematic beliefs of self, emotional well-being, momentary resilience, momentary positive affect, functioning, and subjective quality of life will, on average, be higher and negative self-esteem, negative schematic beliefs of self, momentary negative affect, psychological distress, general psychopathology, and clinical symptoms (secondary outcomes) will be lower in the experimental condition (SELFIE + CAU).The trial was powered to detect an effect size (standardized mean difference (SMD)) of 0.3 (experimental vs. control condition), i.e., a previously reported difference in the RSES self-esteem score that can be considered as clinically relevant for lowlevel interventions (adopting a conservative strategy) 43 .Sample size calculation using power simulation in the R environment indicated that a sample size of 130 participants (65 per condition) is sufficient to test our primary hypothesis that levels of self-esteem are, on average, higher in the experimental than control condition across post-intervention and follow-up with a power of 0.87 (primary hypothesis) when testing at 2-tailed alpha = 0.05 using linear mixed modeling.To allow for 12.5% attrition at 6month follow-up (25% at 24-month follow-up), we aimed to recruit n=174 participants at baseline.In line with the intention-to-treat principle, all data of participants were used in the analysis, including data of participants who dropped out from the intervention, and those with low adherence.We fitted the model using Robust Restricted Maximum Likelihood (robust REML 44 ) in Stata 16 42 , which assumes that data is missing at random [45][46][47][48] .Therefore, potential bias due to attrition over the study period, differences between regions/centres, or levels of self-esteem at baseline were mitigated by the model.
In addition, we assessed potential bias due to missing outcomes in descriptive analyses of baseline characteristics stratified by missing data for condition and the primary outcome.In a sensitivity analysis using multiple imputation in Stata 16, we examined the main effect of condition on self-esteem measured with the RSES, including all n=174, with missing data imputed for baseline variables as well as RSES self-esteem at post-intervention and follow-up as the primary outcome (see eMethods 4).
To test the effect of SELFIE + care as usual (CAU) compared to CAU only on the primary outcome variable (self-esteem measured with the RSES), we fitted a mixed effects regression model with the primary outcome of self-esteem at post-intervention and 6-month follow-up entered as the dependent variable and self-esteem measured at baseline (grand-mean centered), condition (SELFIE + CAU vs. CAU), time (as a two-level factor) and center (as a four-level factor) as independent variables, in line with the intention-to-treat principle.Residuals within subjects were allowed to be correlated with a completely unstructured variance-covariance matrix to take within-subject clustering of repeated measures into account.The model was fitted using robust Restricted Maximum Likelihood (REML) estimation, assuming data is missing at random.The main effect of condition (β2) on self-esteem was parameterized in order to reflect the difference between the two conditions at the two follow-up points (i.e., post-intervention and 6-month follow-up), which was tested (at α = .05)by a Wald-type test with df = 1.This tested the joint null hypothesis (that there is no difference at both follow-up time points) against the alternative hypothesis (that there is, on average, a difference across the two followup time points).The alternative hypothesis allowed for deviations in either direction at both time points.We computed p-values to index statistical significance at α < .05.
To test the effect of SELFIE + CAU compared to CAU only on the non-EMA secondary outcomes (i.e., positive and negative self-esteem, positive and negative schematic beliefs of self, emotional wellbeing, psychological distress, general psychopathology, clinical symptoms, functioning, and subjective quality of life), we used separate (multilevel) linear regression models with the secondary outcomes at post-intervention and 6-month follow-up entered as dependent variables and positive and negative self-esteem, positive and negative schematic beliefs of self, emotional well-being, psychological distress, general psychopathology, clinical symptoms, functioning, and subjective quality of life at baseline, condition (SELFIE + CAU vs. CAU), time (as a two-level factor), and center (as a four-level factor), as independent variables, in line with the intention-to-treat principle.Residuals within subjects were allowed to be correlated with a completely unstructured variance-covariance matrix to take within-subject clustering of repeated measures into account.The model was fitted using robust REML.
The main effect of condition (β2) on the secondary outcome variables was parameterized in order to reflect the difference between the two conditions at the two follow-up points (i.e., post-intervention and 6-month follow-up), which were tested (at α = .05)by a Wald-type test with df=1.This tested the joint null hypothesis (that there is no difference at both follow-up time points).
Further, in order to test the effect of SELFIE + CAU compared to CAU only on the secondary EMA outcomes (i.e., momentary self-esteem, resilience, negative affect and positive affect), we fitted separate (multilevel) mixed effects regression models with the secondary EMA outcomes (eight assessments per day on six consecutive days each) at post-intervention and 6-month follow-up entered as dependent variables and the secondary EMA outcomes (person-mean centered) at baseline, condition (SELFIE + CAU vs. CAU), time (as a two-level factor), center (as a four-level factor), a secondary EMA outcome at baseline × time interaction, and a condition × time interaction as independent variables, in line with the intention-to-treat principle.A two-level model with time points (post-intervention, 6-month follow-up; level 1, i) nested within subjects (level 2, j) was estimated.For subject and time point, a random intercept was included, while we included random slopes for time (level 2; only with an error term, no predictors) and secondary EMA outcomes (level 2; only with an error term, no predictors).The variance-covariance matrix of these effects was set to unstructured.
Additionally, we assumed that the within-subject residuals were autocorrelated at each time point for the EMA observations.The model was fitted using robust REML.
For all above analyses, Cohen's d-type effect sizes and 95% Confidence Intervals (CIs) were constructed for the primary and secondary outcomes in order to provide an estimate of the magnitude of differences between conditions 49 .A d-type effect size was obtained by dividing the model coefficient for the difference between the two conditions by the square root of the total variance.All d-type effect sizes reported in the tables of EMA outcomes are based on the total variance estimated in the mixed model, which is deemed more appropriate in cases where there is (potential for) large heterogeneity between level-2 units (here: participants; random intercept and residual error terms [49][50][51] .For determining effect sizes in more complex models including a random slope for time, we also used the total model-based variance observed in the EMA measures at post-intervention 52 , i.e., we fitted the full model to obtain the variance estimate and then divided the coefficient for the between-group difference by the square root of this total variance.The effect sizes for these therefore represent the effect size (in case of interaction effects: the effect size due to a particular difference between two sub-groups) with reference to the total variation observed in the EMA measures at any point during the post-intervention or follow-up phases.* 30% of all delivered intervention sessions have been rated by an independent rater through the audio recordings.The to be rated sessions were selected based on creating a sample with a variety of therapists who delivered the session as well as a variety of sessions (1, 2, or 3).** rating on a scale from 1 to 7. *** rating on a scale from 1 to 5.  ‡ Breaks in blinding occurred for 3 assessments, which were then re-assigned to, and completed by, other blind assessors.Blinded assessors correctly guessed allocation to experimental and control condition for 12 of 153 assessments at post-intervention (with n=8 correctly guessing allocation to experimental condition and n=4 correctly guessing allocation to control, condition), and for 4 of 140 assessments at 6-month follow-up (n=4 correctly guessing allocation to experimental condition and n=0 correctly guessing allocation to control condition).eMethods 4. Sensi vity analysis on primary outcome at post-interven on and 6-month follow-up using mul ple imputa on.
The analysis of the primary outcome was conducted assuming that data is missing at random given variables included as covariates in the model 48 , as specified in our protocol paper 53 and preregistra on on the open science framework 41 .While these were key variables that relate both to the substan ve research ques on as well as to poten al key drivers of discon nua on (e.g., baseline values of global self-esteem measured with the Rosenberg Self-Esteem Scale (RSES) and treatment alloca on for selec ve responding and drop-out; centre for average differences in the target variables as well as siteeffects on drop-out), these reflect only a limited number of variables, and on request of a peerreviewer, we addi onally report the results of an analysis based on mul ply imputed data sets via chained equa ons.We opted for a chained equa ons approach as it allows a flexible specifica on of the imputa on model across mul ple me points.The imputa ons were conducted in Stata 16 42 , and we imputed 20 data sets.The imputa ons were started with a fixed seed value (generated via random.org)and with 500 burn-in itera ons.Con nuous variables were predicted using linear regressions, dichotomous variables using logis c regressions, and the posterior es mates of the model parameters were es mated from bootstrap samples.
The imputa on was performed separately for experimental and control condi on 48 , and used the following baseline variables (see details in Table 1): age, sex (dichotomised as male vs female), educa on (dichotomised as low vs middle and high), employment (dichotomised as ac ve (employed and student) vs. non-ac ve (unemployed)), ethnicity, study site, and medica on represen ng par cipant mix, as well as the secondary outcome variables posi ve and nega ve self-esteem measured with the Self-Esteem Ra ng Scale (SERS); general psychopathology measured with the revised Symptom Checklist (SCL-90-R); the four subscales (physical, psychological, social, environment) for quality of life, assessed with the World Health Organisa on Quality of Life Instrument-Brief The RSES entered the imputa on in three places: it was used as a baseline variable, and the postinterven on and follow-up assessments as longitudinal endpoints.The baseline variables were all used to impute any baseline missing values (see Table 1, only very few missing values) as well as for the imputa on of the RSES at post-interven on and follow-up.The post-interven on value of the RSES was also used to impute the follow-up assessment of the RSES.We decided on this set-up to represent the temporal order as well as since the overall sample size of the trial is small for models with many predictors and could easily result in models with more parameters than our sample size (especially in a by-group imputa on).Addi onally, as there are increasingly missing values on all variables across the three me points, but our main interest was the RSES, therefore using mul ple variables from each me point could lead to increased predic on error.
We es mated the average difference in RSES self-esteem between condi ons across post-interven on and follow-up as the primary outcome via employing a mixed model trea ng post-interven on and follow-up data as nested observa ons within individuals, with a separate random effect error term for me.This devia on from the main analysis model (see also eTable 7 below), which calculated predicted differences based on model-es mated coefficients, was necessary since the integra on of mul ply imputed data sets for these opera ons is not fully resolved at this point, and therefore to define a model that most-closely resembled the model test in a single model es mated from the mul ply imputed data.Based on this model, the RSES was, on average, higher in the experimental than control condi on across post-interven on and follow-up me points (B=2.35,95% CI 1.13-3.57,p<0.001; original model: B=2.32, 95% CI 1.14-3.50,p<0.001).eTable 7 further shows the results of re-es ma ng the main analysis model on the imputed data sets compared to the original parameter es mates.This sensi vity analysis showed broadly similar findings compared with our a priori planned main analysis for the primary outcome.While the 95% CIs were wider for the imputed data than for the original analysis, none of the es mated parameters showed a substan al shi .Specifically, the findings from the sensi vity analysis that the evaluated average difference in RSES levels between the two condi ons (i.e., the primary outcome in the main analysis) was in the expected direc on and the es mated coefficients were of similar size, indica ng similar effect sizes.When inspec ng findings at each me point separately, higher levels of global self-esteem in the experimental than control condi on were observed at post-interven on and follow-up, with point es mates for between-condi on differences being of similar size in the main and sensi vity analysis.Further, findings at post-interven on showed that levels of global self-esteem were higher in the experimental than control condi on, with a very similar 95% CI that did not include 0 (B=2.74,95% CI 1.34-4.14;adjusted es mate of main analysis B=2.83; 95% CI 1.46-4.20;see Table 3).The only difference between main and sensi vity analysis was that the analysis based on mul ply imputed data showed a slightly smaller between-condi on es mate at follow-up, with the 95% CI being wider and poten ally not different from "0" (1.58, 95% confidence interval -0.12 to 3.28; adjusted es mate of main analysis 1.81; 95% confidence interval 0.38 to 3.23; see Table 3).eTable 7. Parameter es mates for the mixed model analysis for the primary outcome, le column based on n=174 with 348 observa ons a er mul ple imputa on; right column presents es mates from originally es mated mixed model (see Table 3).Building on an ecological interven onist causal model approach 13,66 , we adopted a strategy that targets candidate mechanisms, demonstrates that these are modifiable in vivo by an Ecological Momentary Interven on, and examines subsequent effects on the outcome.We have previously argued that the benefits of this approach, applied in the context of randomized controlled trials, are at least two-fold: first, this approach allows us to inves gate important criteria for establishing causality such as associa on, temporal order, experimental evidence, and sole plausibility outside the research lab, in daily life.Second, it provides robust evidence whether the experimental manipula on method (i.e., the EMI) reflects an efficacious technique for preven on and interven on in individuals' daily living environments.Given previous research suggested that self-esteem is an important candidate mechanism in pathways to adverse mental health outcomes 67,68 (including in daily life 38,39,[69][70][71] ) and provided evidence in support of 'associa on' as one important criterion for establishing causality, selfesteem reflected a promising interven on mechanism for the current trial adop ng an ecological interven onist causal model approach.Notably, our findings may be viewed par ally suppor ve of this approach, as, first, they provide experimental evidence on significant and sustained improvements in the primary outcome of global self-esteem of moderate effect size, corroborated by effect sizes sugges ve of beneficial effects in secondary outcomes on posi ve and nega ve self-esteem, schema c self-beliefs, and, importantly, momentary self-esteem in daily life.Second, while no improvements were evident in observer-rated symptoms and func oning, effect sizes signaled beneficial effects for self-reported general psychopathology and quality of life as important subsequent outcomes.In short, our findings provide experimental evidence on modifying self-esteem as target mechanism (in vivo) and important subsequent outcomes.Overall, the current study reflects the first confirmatory trial of the ecological interven onist causal model approach and, coupled with our recently published EMIcompass trial 72 , contributes more generally to demonstra ng that the ecological interven onist causal model approach reflects an important transla onal strategy from mechanism to interven on and from the lab/clinic to daily living environments that may substan ally contribute to ecological transla on in public mental health provision.In a next step, we now need to formally test, in secondary (media on) analyses of the current trial, including data on long-term effects at 18-and 24-month follow-up, the temporal order of experimental manipula on/exposure (i.e., SELFIE), puta ve mechanism (i.e., self-esteem), and subsequent outcomes (i.e., general psychopathology, quality of life), are genuinely transdiagnos c in nature.This applies in par cular to clinical outcomes that require measures with sufficient bandwidth cu ng across dimensions and categories of diagnos c en es and capture relevant variance across clinical stages (e.g., in terms of frequency, severity etc.).While important work to address this challenge of iden fying transdiagnos c symptom measures is under way 73 , current studies need to draw on already available and validated scales.While the BPRS and K10 have been recommended for capturing clinical outcomes in youth 25 , which informed our choice as outcome measures in the current study, it has not been designed to cover all relevant domains of early pluripotent or extended transdiagnos c phenotypes and their symptoma c outcomes.We urgently need to address this challenge for the field of preven on and early interven on to be able to progress to target transdiagnos c mechanisms and intervene at an early point to prevent enduring adult mental health problems in youth.

eTable 1 . 2 . 3 . 5 . 6 . 4 .
Eligibility Criteria eMethods 1. SELFIE Intervention Description eTable Overview of SELFIE Intervention Procedure eMethods 2. Measures eMethods 3. Statistical Analysis eTable Diagnoses at Baseline eTable 4. Correlations Among Primary and Secondary Outcomes at Baseline eTable Percentiles and Range of Baseline Characteristics of the Intention-to-Treat Sample eTable Compliance With, Fidelity to, and Acceptance of the SELFIE Intervention eMethods Sensitivity Analysis on Primary Outcome at Postintervention and Follow-Up Using Multiple Imputation eFigure.Response Trajectories in the Primary Outcome (RSES Self-Esteem) eAppendix 1 SELFIE in Context of Previous and Future EMI Research eAppendix 2 Targeting Self-Esteem in Ecological Interventionist Causal Models and Transdiagnostic Intervention Research eReferences This supplementary material has been provided by the authors to give readers additional information about their work.eTable 1. Eligibility criteria Inclusion criteria 1) Aged between 12 and 26 years old.

eTable 2 . 1 Face
Overview of SELFIE interven on procedureWeek ng the SELFIE interven on Short face-to-face closing session

(
WHOQOL-BREF); the Posi ve and Nega ve Affect Scale (PANAS) subscales for posi ve and nega ve affect; clinical symptoms of psychopathology measured with the Brief Psychiatric Ra ng Scale (BPRS); the four subscales of posi ve and nega ve schema c beliefs of self and others measured with the Brief Core Schema Scale (BCSS); and func oning, assessed with both the Global Assessment of Func oning (GAF) scale and the Social and Occupa onal Func oning Assessment Scale (SOFAS).

eAppendix 2
Targe ng self-esteem in ecological interven onist causal models and transdiagnos c interven on research which will, in turn, allow for clarifying the role of self-esteem as transdiagnos c target mechanism and/or outcome.The evalua on of interven ons targe ng transdiagnos c mechanisms across various diagnos c en es and severity levels of mental health problems face the challenge of selec ng outcome measures that © 2023 Reininghaus U et al.JAMA Psychiatry.

eTable 6b .
Compliance with, fidelity to, and acceptance of the SELFIE intervention by using own vs. study smartphone ‡ Note: S.D., standard deviation, all ratings on a scale from 1 to 7. ‡ Of n=85 participants randomized to the experimental condition, n=10 used a study smartphone, n=67 used their own smartphone, and n=8 did not start using a smartphone for SELFIE intervention delivery.Missing values in user experience, satisfaction and acceptance questionnaire, n=16.

Table 6c
Observer-rated secondary outcomes at post-intervention and 6-month follow-up, excluding assessments for which blinded assessors correctly guessed random allocation to experimental and control condition.‡ Adjusted for centered baseline values, region and group status; SE, Standard Error; N, number of participants; CI, confidence interval; BPRS, Brief Psychiatric Rating Scale; SOFAS, Social and Occupational Functioning Assessment Scale; GAF, Global Assessment of Functioning.