Effectiveness of the Famílias Fortes program was evaluated by a two-arm cluster randomized controlled trial (RCT) conducted in 60 Social Assistance Reference Centers (SARC) and similar municipal services, hereafter referred to globally as “social services,” from 12Footnote 1 municipalities designated by the Federal Government. These SARC are public centers responsible for delivering social programs, benefits and projects to low-income families and individuals in the community, including registering and monitoring the families benefited by Bolsa Família, a a social welfare program provided by the government to extremely low-income families. The existing SARCs were randomized in each city between intervention and control groups. Each SARC invited 15 families to participate in the study from the list of attended families and considering the inclusion criteria (have at least one child/adolescent between 10 and 14 years old; adult and teenagers living in the same house; availability to attend the 7 program meetings; live up to 1 km away from the social centers) and exclusion criteria (families with very high needs or challenges such as drug addiction and family breakdown). If the SARCs had more than 15 families interested, the first to volunteer were included in the study.

The families (made up of parents or legal guardians and child/adolescent) in the intervention group attended the Famílias Fortes program for seven weeks in person, while those from the control group were put on a waiting list to participate in the program after the end of the study. Data collection occurred at two time points: before the intervention (November/December 2021), and 6 months after implementation (May/June 2022).

Families answered self-completed, anonymous, virtual questionnaires via Android smartphone app or Internet link provided by the researchers. Parent and adolescent data were integrated through linkage considering the family confidential code. For any family code typing errors, the Levenshtein algorithm was used to pair the subjects, as described in previous studies [31, 32].

The Famílias Fortes program (PFF-BR 10–14) is the Brazilian adaptation of the Strengthening Families Programme (SFP-UK 10–14), developed in the United Kingdom by the Oxford Brookes University [26]. PFF-BR 10–14 consists of seven consecutive weekly 2-hour in-person meetings. Caregivers and children meet separately in the first hour and spend the second hour together in family activities. All meetings include debates, games, and interactive activities and a snack is offered at the end. Some sessions have the support of videos depicting situations of daily family life. The themes of these meetings are: (1) support goals and dreams; (2) admire family members; (3) family moments; (4) understanding family values; (5) strengthen family communication; (6) family and pressure from friends; and (7) putting it all together. Highly structured, the program is guided by the instructor, caregiver, and youth manual, all available at https://www.gov.br/mdh/pt-br/navegue-por-temas/familia/familias-fortes-1.

The professionals who delivered the program (called implementers) were typically SARC employees, such as psychologists and social workers, who had been previously serving the participating families and had undergone specific training for program implementation. Program methodology training for implementers was supervised and monitored by the MMFDH team. It was conducted online via a digital platform developed specifically for distance learning courses, the training was configured as a 25-hour course divided into 3 modules: (1) Introduction and theoretical grounds of the Famílias Fortes; (2) Training to conduct the Famílias Fortes meetings, and (3) Practices for the follow-up and closing of the Famílias Fortes.

Importantly, the program’s cultural adaptation was carried out in 2014 and 2015 by the Federal University of Brasilia [27]. In this process, European everyday examples were replaced by activities typical to Brazilian everyday life. The branding, vocabulary, and presentation format of some activities were also adapted without changes in content and core elements. Evaluations of this culturally adapted version [33, 34] concluded that the program was sufficiently attractive, culturally relevant, with acceptable goals, and compatible with the needs of Brazilian families in vulnerable contexts (the target audience of the SARC).

Adolescents and caregivers each answered a self-completed, virtual, anonymous questionnaire via an Android smartphone app or Internet link made available by the researchers during data collection, without any participation of the implementers. The two questionnaires, one for parents/guardians and the other for adolescents, were built based on international instruments developed for evaluating the Strengthening Families Program (SFP 10–14) [35] in combination with instruments previously employed in effect evaluation studies of drug use prevention programs in Brazil [36, 37]. Such instruments were extracted from questionnaires widely used in several national and international studies on drugs such as the World Health Organization questionnaire used by the Brazilian Center for Drug Information (CEBRID) [38] and that used by the Substance Abuse and Mental Health Services Administration [39].

In this study, parental outcomes (family violence, parenting styles, and children’s exposure to parental drug use) and adolescent outcomes (perceived drug risk and lifetime drug use) were assessed both at baseline and at 6-month follow-up.

Family violence was assessed using the WorldSAFE Core Questionnaire scale [40] translated and validated into Brazilian Portuguese [41]. Questionnaire questions are grouped into 5 subscales: Nonviolent Discipline, Moderate Verbal Discipline, Severe Verbal Discipline, Moderate Physical Discipline, and Severe Physical Discipline. The instrument asks how often caregivers have used specific disciplinary tactics, with responses rated on a 3-point scale: never, 1–2 times, and ≥ 3 times in the previous three months. The Moderate Verbal Discipline and Severe Verbal Discipline subscales were grouped together and its corresponding score ranged from 0 to 16, in which the higher the value the higher the degree of verbal discipline imposed. The Moderate Physical Discipline and Severe Physical Discipline subscales were also grouped together and the score ranged from 0 to 24, in which the higher the value the greater the degree of physical discipline effected. Nonviolent Discipline score ranged from 0 to 10, and the higher the value, the greater the degree of nonviolent discipline employed.

Parenting styles were evaluated by the Scale of Parental Demands and Responsiveness [42], translated and validated into Portuguese [43]. Each instrument item (six items composing the demanding dimension; and ten items the responsive dimension) is assessed by a 3-point Likert scale, where values closer to three indicate greater perceived demand and responsiveness (ranging from 0 to 12 and 0 to 20, respectively). Parenting style dimensions were established using the median split procedure following the methodology proposed by previous studies [42,43,44]. Caregivers who scored at or above the median for demandingness or responsiveness were classified as high in demandingness or responsiveness, whereas caregivers with scores at or below the median were classified as low in demandingness or responsiveness. Parenting styles were organized into four categories combining these two factors: authoritative (scoring high on demandingness or responsiveness), authoritarian (scoring high on demandingness and low on responsiveness), indulgent (scoring low on demandingness and high on responsiveness), and negligent (scoring low on demandingness and responsiveness). Prior psychometric evaluation of the scale in a sample of Brazilian adolescents showed good fit: χ2 = 1518.249, p < 0.001, RMSEA = 0.050, CFI = 0.940, TLI = 0.929, WRMR = 2.377 [45].

Child exposure to parental drug use was measured by means of a block of questions that assessed the degree to which caregivers exposed their child to drug use (alcohol, cigarette, marijuana, and cocaine), such as “Do you get drunk near your child?,” and similarly to social drinking, smoking and use of illegal drugs, whose response options could be “never, sometimes, and always.”

Adolescent drug use was assessed as lifetime drug use (alcohol, cigarette, marijuana, inhalants, and binge drinking) and risk perception (alcohol, cigarette, and marijuana) measured by questions drawn from the CEBRID [38] and SAMHSA [39] questionnaires. Lifetime drug use (yes X no) was assessed through questions such as “Have you ever tried any alcoholic beverages?” Perceived risk was measured by one question for each drug with Likert-type response options ranging from no risk to high risk, such as “How risky is it for someone your age to smoke cigarettes once or twice?”

Covariates included age, gender, race, and socioeconomic status (SES) of the caregivers and adolescents. SES was assessed using the Brazilian Association of Research Enterprises (ABEP) scale [46], which considers the schooling of the head of the household and goods and services used by the household. ABEP score ranged from 1 to 100 points, with categories ranging from highest to lowest according to the cut-off points established in literature: high (45–100), medium [29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44], medium-low [17,18,19,20,21,22,23,24,25,26,27,28] and low (0–16) [46].

Program effectiveness analysis was performed according to two paradigms: Intention to Treat (ITT) and Per Protocol (PP). The ITT model considered all families who participated in the study regardless of number of meetings attended and whether or not they answered the questionnaire at follow-up. The PP model, in turn, considered only those families who had participated in at least five program sessions or more and who answered the baseline and 6-month follow-up questionnaires, aiming to ensure program evaluation among people who joined it, similar to another study [21]. Attending at least five sections was considered the protocol for MMFDH dissemination and not an arbitrary decision of the researchers.

Considering the longitudinal hierarchical structure of the data, a multilevel mixed-effects modeling with repeated measures approach was adopted. Changes in intervention outcomes use over time was evaluated by three-level mixed-effects models (level 1: repeated time observations nested within the subject; level 2: subject clustered within SARC; level 3: SARC). Models were calculated using SARC and subjects as a random effect, and explanatory variables (experimental group, time of assessment, the interaction between group and time), controlling variables (gender, age, SES), and study outcomes as fixed effects. Unadjusted estimations were obtained when considering the effect of only one independent (predictor) variable and adjusted estimations were controlled for covariates (sex, age, and socioeconomic status). Standard errors allow for intragroup correlation, that is, the observations are independent across groups (clusters) but not necessarily within groups. All models were fitted with STATA 17 program generalized linear mixed models (GLMM).

Mixed-effects models are powerful tools for analyzing cluster randomized trials that handle two sources of non-independence: the clustered observations and the repeated measures within each subject over time [47]. Additionally, mixed-effects models deal with missing data using maximum-likelihood estimation analyzing all available outcome data, regardless of whether an individual has complete data, making these models consistent with an ITT analysis [48]. In our sample, we assumed the missing data mechanism as missing at random (MAR), that is, when the probability of missing data on a variable is related to other variables measured in the model rather than the variable with missing values itself. Missing data stemmed mostly from failure to answer the follow-up assessment questionnaire and not because single items remained unanswered or due to dropping out. Mixed model for repeated measures under MAR assumption produces valid and unbiased estimations and additional methods for handling missing data, such as multiple imputation, are generally unecessary [47]. For attrition analysis, we compared families whose data from the two time-points had been matched with families who answered only the baseline questionnaire.