This randomized clinical trial used a full 2³ (2×2×2) factorial design based on the multiphase optimization strategy (MOST) framework to test and compare three intervention approaches: (1) a family-strengthening intervention, (2) trauma-focused child mental health services, and (3) asset-based economic empowerment via Child Development Accounts (CDAs).

Multicomponent interventions are often tested as “bundles,” making it difficult to disentangle the contributions of individual components. The MOST is a systematic method that allows for the testing of the independent effects of different intervention components to identify their most optimal combination [15]. While multiarm randomized controlled trials are costly, often limited in arms, and require larger sample sizes, the MOST uses a factorial design. By comparing combinations of experimental conditions, the study can have a greater number of conditions without a substantial increase in sample size. This is advantageous in LMICs, where it is essential to implement only the most efficacious aspects of a program. The MOST consists of three phases: (1) preparation, (2) optimization, and (3) evaluation of the optimized intervention in a randomized controlled trial [16]. This study falls within the optimization phase.

This study includes 3 factors—the intervention approaches—each with 2 levels (delivered/“on” or not delivered/“off”). This is a full 2×2×2 factorial design, meaning that the study has 8 different experimental conditions or groups (2³=8) [17]. Child-parent dyads (n=439) were randomly assigned to 1 of 8 groups (about 54 dyads per group), which determined the combination of interventions they received. Given this design, half of the sample (n≈220) received each intervention (eg, families assigned to groups 1‐4 received the family-strengthening intervention and were compared with families from groups 5‐8; Table 1).

This trial has three study aims:

The study also aims to conduct a secondary analysis to estimate the preliminary costs (operations and personnel) of each intervention component relative to usual care alone.

The sample consists of 439 caregiver-child dyads living in 3 cities in Azerbaijan: Baku, Sumgait, and Ganja. Caregivers and children must meet the eligibility criteria to participate and commit to study participation to be enrolled. Eligible children (1) are 7 to 12 years old at baseline, (2) have a history of institutionalization (for at least 1 month), and (3) have been reunited with a caregiver (biological parent or kin relative) within a year before the start of the study.

Eligible caregivers (1) are a biological parent or kin relative (eg, grandparent) who serves as the primary caregiver;, (2) reunified with the child before the study, and (3) are at least 18 years old at baseline. Children or caregivers with significant cognitive, behavioral, and/or mental health impairment (eg, severe developmental disorder) that may interfere with their ability to consent to, benefit from, or participate safely in the program are excluded from participation. More than 1 eligible child in a family may be enrolled in the study. Enrollment requires consent from the caregiver and assent from the child (see Multimedia Appendix 1 for sample consent and assent forms).

Power calculations were conducted using the G*Power software [18,19]. To account for the repeated-measures design, power estimates were modeled across a range of effect sizes at 2 assumed autocorrelation values (ρ). Analyses were performed for a 2-tailed test with a type 1 error rate of 0.05 and desired power of 0.80, assuming a final analyzed sample size of 360 child participants (180 participants per intervention type) after a projected 10% attrition from an initial planned sample of 400 child participants (200 participants per intervention type). The results indicate that for continuous outcomes, 80% power is achieved for effect sizes between 0.09 and 0.12. The study is sufficiently powered to detect small-to-medium effects [20] for continuous outcomes over the entire follow-up period. In the MOST design, the study is powered based on the intervention component with the smallest effect size—in this case, the family-strengthening component.

The intervention components were chosen based on the findings from a qualitative study [7] conducted in Azerbaijan from 2014 to 2016 with children aged 8 to 16 years who had a history of institutionalization and their caregivers (n=47). In-depth interviews were conducted with children, caregivers, and experts working with these families. Three primary areas for intervention were identified: addressing children’s mental health needs, strengthening family relationships, and tackling financial challenges during reunification [21,22].

A pilot feasibility study was carried out from 2017 to 2019 to adapt and pretest the family-strengthening intervention with 20 reunified families. The adapted intervention was pilot-tested in partnership with 2 local organizations: SOS Children’s Villages-Azerbaijan (community-based service provider) and the National Mental Health Centre (NMHC; research and delivery of mental health services). This study uses an adapted version of SAFE Children, a family-based preventive intervention originally developed and tested in the United States for at-risk families in low-income urban neighborhoods [23]. SAFE Children was selected because of its strong strengths-based approach and its demonstrated acceptability to families of varying sizes and compositions—a critical feature when working with families affected by institutional care, many of whom do not live in traditional 2-parent households. The research team adapted SAFE Children to the cultural context of Azerbaijan and trained 10 local facilitators (eg, social workers, family support workers) to deliver the adapted version, named “Bütöv Ailə” (Families Together in Azerbaijani). The program consists of 12 weekly multifamily group sessions focused on improving parent-child relationships, supportive parenting practices, communication skills, and child behavior management, with additional content addressing unique challenges faced by families during reunification after institutional care (eg, stigma, fear of abandonment).

The study is informed by theories of family resilience [24,25] and emotional regulation [26,27]. Developmental theory suggests that adversity and trauma experienced in early childhood may interfere with typical developmental trajectories, which can result in emotional and behavioral problems [28,29]. Yet, with appropriate resources and support, children may develop resiliency and overcome disadvantages [30,31]. In addition, asset theory [32] posits that assets—such as savings, homeownership, education (human capital), stable housing, and small business ownership—unlike cash benefits, which can be quickly depleted, provide significant long-term psychological benefits beyond economic outcomes, such as reducing financial stress and enhancing aspirations and hopes for the future [33,34].

Therefore, it is hypothesized that strengthening children’s individual capacities and external resources can balance the effects of adversity and improve mental health outcomes. Enhancing children’s coping skills, strengthening child-parent relationships, reducing parental stress, and improving family economic well-being by incentivizing savings and building assets will result in children demonstrating fewer symptoms of (1) internalizing problems (depressive or anxious mood), (2) externalizing problems (aggressive, delinquent, or disruptive behaviors), and (3) posttraumatic stress (Textbox 1).

This study uses an adapted version of the family-strengthening intervention “SAFE Children” to improve child-parent relationships, enhance supportive parenting strategies, and prevent emotional and behavioral problems among at-risk children [23]. SAFE Children was initially designed for at-risk families from low-income US neighborhoods and uses a multi-family group approach to build social support. Including multiple families is an efficient service delivery model, fostering support among participants and improving parent-child interactions directly [35].

Families participated in 12 weekly sessions, with themes including identifying family strengths, developmental expectations, communication skills, behavior management, parental distress management, and conflict resolution. The adapted intervention incorporates topics that address unique issues faced by families during reunification (eg, attachment issues, children’s feelings of resentment, parental guilt, and shame). In response to the findings of the pilot study, sessions were offered on weekends to accommodate families’ schedules. The intervention developer and coinvestigator (DB-S) oversaw the training and intervention adaptation process to ensure fidelity to its core components. The intervention is delivered in multifamily groups of 5 families. Children and caregivers attend sessions together, separating into child- and caregiver-specific activities only when required by the curriculum. To strengthen family engagement and reinforce skills at home, all household members involved in child-rearing are encouraged to participate, including siblings who do not meet study eligibility criteria.

Families were connected to clinicians from the NMHC to receive child mental health services. Services included a clinical assessment performed by a psychiatrist, the development of an individual care plan, and assignment to a mental health clinician—typically a child psychologist—who provided services to the child and caregiver. The trauma-focused child mental health services were grounded in the Attachment, Regulation, and Competency framework [36]. All clinicians delivering this intervention received additional training in trauma-focused mental health approaches, including Attachment, Regulation, and Competency, and participated in biweekly group supervision delivered by a bilingual licensed psychologist from the United States. The clinicians delivering this intervention were not involved in conducting research activities. As a public entity, the NMHC provides all mental health services to the public free of charge. Children typically attended individual sessions once a week (30‐45 min), with parent or caregiver involvement incorporated as clinically indicated. Up to 12 sessions were offered; however, services could be extended based on individual treatment plans and clinical needs.

In addition to the 3 experimental interventions, all participants were eligible to receive standard public services under Azerbaijan’s national deinstitutionalization program for family reunification and reintegration. This included case management, referrals to public schools, and social support services during the transition and reintegration process—such as medical-social services, sociopedagogical support (education and development), and linkage to the general social protection system (financial and material assistance for vulnerable families). The government’s standard support focuses on case management, access to social and medical-psychological services through state institutions, and basic follow-up after reunification. The emphasis is on returning children to a family environment and connecting them to existing public services.

For each component, we will analyze data regarding participation (eligibility, enrollment, and attrition rates), intervention fidelity (via facilitator and participant rating forms completed at the end of each session), safety (number of adverse events), intervention acceptability (participant satisfaction, session attendance, and associations between satisfaction and attendance), perceived usefulness (relevance and utility), and cross-arm contamination.

Two interim analyses will be conducted when 33.3% and 66.7% of the planned sample size have completed 6-month follow-up assessments. With respect to efficacy, if any interim analysis shows statistical significance (P<.05) for all primary outcomes, the study will be stopped, indicating a significant benefit of 1 intervention over another. With respect to futility, if the effect size estimates suggest a null result is likely given the planned total sample size, the study may be stopped to conserve resources. A post hoc power analysis will assess the minimum sample size needed to achieve 80% power and reject the null hypothesis at each interim analysis. If this exceeds the planned 400 dyads for self-reported outcomes, the investigative team will discuss whether to stop the study for futility or modify the design to avoid it, such as increasing enrollment or hiring additional staff.

Families were recruited in 3 major cities (Baku, Sumgayit, and Ganja). Trained local research assistants (RAs) distributed recruitment materials to recruiters who invited families to join the study. Local research staff provided additional information to interested candidates and obtained informed consent. To minimize potential pressure on children and parents, consent was obtained by a research staff member rather than by representatives of local authorities.

After obtaining informed consent, RAs conducted a screening interview to determine participant eligibility. Potential participants were offered a choice of either US $1 or equivalent nonmonetary compensation (eg, toys, football) for completing the interview. To promote retention, participants provided their addresses, email addresses, and telephone numbers at baseline, along with contact information for at least 3 other individuals who could assist in reaching them. After completing the baseline survey, families were informed of their randomized intervention assignment.

The principal investigator (PI) conducted the randomization prior to the start of recruitment. Eight computer-generated random numbers (ranging from 0 to 100) were assigned to the 8 intervention conditions and ranked from lowest to highest to determine the allocation sequence within each city. The first 5 families recruited in a city were assigned to the first intervention condition in the sequence. After 5 families had been assigned to each intervention condition, allocation resumed at the beginning of the sequence. To reduce predictability, the random numbers were generated from a range of 0 to 100 rather than 1 to 8. A secure, web-based, password-protected database was used to maintain the randomization record.

Data collectors were masked to participants’ intervention group assignments. To foster fidelity to the intervention manuals and maintain the integrity of each separate intervention, facilitators delivered only 1 type of intervention and were not trained in the specific content of the other intervention strategy. Due to the nature of the recruitment procedures, participants could not be blinded to the existence of other intervention conditions.

As presented in Table 2, the primary mental health outcomes include child-reported symptoms of depression (20-item Center for Epidemiological Studies Depression Scale for Children) [41], posttraumatic stress (8-item Children’s Revised Impact of Events Scale) [42], self-esteem (20-item Tennessee Self-Concept Scale-Second Edition) [43], attention-deficit/hyperactivity disorder (ADHD) symptoms (caregiver and teacher-reported 19-item ADHD Rating Scale-IV) [44], and general social-emotional functioning (caregiver and teacher-reported 25-item Strengths and Difficulties Questionnaire) [45]. Grades from school records and teacher-reported school behaviors (eg, completing homework, fights with other students) are also outcomes of interest. The Strengths and Difficulties Questionnaire has been validated in the Azerbaijani language [46]. Child mental health outcomes among families randomized to the mental health services were also assessed by a clinician-administered Health of the Nation Outcome Scales for Children and Adolescents (15 items) [47].

Additional child-reported measures include social support (24-item Social Support Scale for Children) [48] and parental discipline (11-item Child Discipline Index) [49]. Child cognitive functioning was assessed using selected measures from the Wechsler Intelligence Scale for Children, Fifth Edition (WISC-V) [50] and the Cambridge Neuropsychological Test Automated Battery (CANTAB) [51]. WISC-V measures included Block Design, which assesses visuospatial reasoning and problem-solving abilities, and Picture Span, which evaluates visual working memory. CANTAB measures included the Emotional Bias Task, which assesses emotional processing and interpretation bias; the Motor Screening Task, which evaluates sensorimotor function and task comprehension; Reaction Time, which measures processing speed and attention; the Stockings of Cambridge, which assesses executive functioning and planning; and Spatial Span, which measures visuospatial working memory capacity.

Caregiver-reported measures include parenting style (32-item Parenting Styles and Dimensions Questionnaire, Short version) [52]; parental mental well-being symptoms, including depression, anxiety, and stress (21-item Depression Anxiety Stress Scale) [53]; parental experiences of childhood trauma (19-item Adverse Childhood Experiences International Questionnaire) [54]; family relationships (19-item Brief Family Relationship Scale) [55], coping strategies (28-item Brief COPE) [56]; and social support (12-item Multidimensional Scale of Perceived Social Support) [57]. The caregiver survey also assessed economic well-being (eg, employment status, financial security) and saving attitudes.

The baseline questionnaire collected information on sociodemographic characteristics (eg, child and caregiver age and gender), family composition (eg, caregiver marital status and household size), and the child’s institutional history (eg, age at placement and type of institution). For participants receiving the economic intervention, bank statements provided data about the monthly savings and asset accumulation rate.

Children, their caregivers, and their teachers completed surveys at baseline, year-1, and year-2 follow-up. Child and caregiver data were collected through interviewer-administered surveys conducted separately in Azerbaijani by trained graduate-level psychology students serving as RAs. Assessments were conducted at local field offices using iPads and the cloud-based Qualtrics survey platform and required approximately 50 to 60 minutes to complete. Teacher surveys were completed independently. All surveys contained the same scales, except for demographic data that do not vary over time (eg, ethnicity, age). Except for nonverbal digital cognitive tasks (WISC and CANTAB), all instruments were translated into Azerbaijani and tested in a pilot feasibility study. The CANTAB battery consists of simple computer-administered nonlinguistic tasks considered highly precise and largely independent of cultural and language differences [58].

Intervention participation and utilization data (eg, attendance, session completion, and types of services received) were collected from intervention administrative records. Information about children’s academic performance (eg, grades) was collected from school records.

Semistructured interviews will be conducted with children, caregivers, and service providers (n=60) representing each intervention condition following the completion of intervention sessions and follow-up assessments. Interviews will explore participants’ experiences with the interventions, including perceived benefits, acceptability of intervention components, impacts on family relationships and child well-being, and barriers and facilitators to participation. Service providers (eg, mental health clinicians, family strengthening facilitators, and financial education coaches) will provide additional perspectives on implementation, training, organizational, and cultural factors affecting intervention delivery. Participants will be randomly selected from the 7 active intervention conditions within each city; the control condition will be excluded from qualitative sampling.

Descriptive statistics will characterize the sample and assess the variables’ psychometric and distributional properties (eg, potential outliers). The analysis will also include process measures for intervention implementation and cost estimation of intervention delivery, following Weinstein et al [59] and Gold’s [60] guidelines. Additional analyses will be performed in conjunction with the study aims:

In the data analysis, the primary outcomes include children’s mental health and socioemotional functioning. Secondary outcomes include cognitive and social processes associated with mental health, as well as functional outcomes related to academic performance and social functioning in the school setting. Specifically, primary mental health outcomes include child-reported depressive symptoms, posttraumatic stress symptoms, and self-esteem; caregiver- and teacher-reported ADHD symptoms and general socioemotional functioning such as internalizing and externalizing problems; and clinician-administered mental health outcomes among families randomized to the mental health services condition. Secondary outcomes include child cognitive functioning, assessed using WISC-V and CANTAB tasks; child-reported social support, particularly peer support; and teacher-reported grades and school behaviors. Mediating variables include caregiver-reported parenting practices, caregiver mental health, family functioning, coping strategies, social support, saving attitudes, and household economic well-being.

The main analyses focus on treatment effects at the 1-year and 2-year follow-ups, with the 1-year follow-up used to assess initial intervention effects and the 2-year follow-up used to assess the maintenance of those effects over time. Children from the same family will be accounted for using mixed-effects models, with random effects specified to adjust for the nonindependence of observations among children nested within the same family. Missing outcome data will be handled using maximum-likelihood estimation under a missing-at-random assumption, with sensitivity analyses using multiple imputation. To address the multiple testing issue, we will control the false discovery rate using the Benjamini-Hochberg procedure [66,67]. This approach was selected instead of Bonferroni correction because it limits false discoveries while avoiding the overly conservative loss of power associated with Bonferroni adjustment when multiple related hypotheses are tested.

Given the vulnerability of children with histories of institutional care and family separation, the study incorporates several safeguards to protect participants’ safety, privacy, and autonomy. Caregiver consent and child assent are obtained separately by trained Azerbaijani-speaking research staff to reduce coercion, and families are informed that participation is voluntary and that refusal or withdrawal at any time does not affect access to usual care or public mental health services. Adverse events are monitored, reported, and reviewed for systematic trends every 2 months. Research staff and intervention facilitators are trained mental health, psychology, or social work professionals, and procedures are in place to respond to emotional distress, including a General Distress Management Protocol and referral to the National Mental Health Centre when serious mental health concerns or risks of harm to self or others arise. Confidentiality and its limits are explained during consent and assent; although local law does not mandate research staff to report suspected child abuse or neglect, the study follows Child Safeguarding procedures for suspected child-safety incidents, including review by the local Project Director and Child Safeguarding team and referral to appropriate child protection authorities for serious cases. Institutional review boards at the University of Chicago (IRB20-1018) and Azerbaijan Medical University approved this study and must approve any future amendments to the protocol.

Compensation is modest and prorated to reimburse participants for their time, thereby reducing the possibility of coercion. Eligible participants received US $10 per person for each assessment and US $10 per family for each intervention session attended to compensate for their time and transportation costs. Compensation is provided after each completed visit or session, with partial compensation available for incomplete participation.

All identifiable information is anonymized with a unique study ID number, and only the PI and local project coordinator can access deidentified data. The PI and local project director monitored local data collection and management procedures regularly. The Data and Safety Monitoring Board, an independent committee responsible for ensuring participant safety and data integrity, conducted the monitoring of research processes in collaboration with mental health professionals, a biostatistician, and representatives from the Azerbaijan Medical University.

The investigative team has shared information about this trial via ClinicalTrials.gov, as per National Institutes of Health policy, and at key conferences. The team will develop a plan for dissemination at the national and regional levels (eg, brief reports). Once all data have been deidentified, cleaned, and validated, and the main findings have been published, the investigators expect to share data with the scientific community. To safeguard the protection of participants, datasets free of identifiers will be made available to individuals who directly contact the PI.