Optimizing infant development and minimizing risk of infant mental health issues has become a public health priority internationally (World Health Organization, 2014). Governments and institutions worldwide are directing significant capital into programs and agencies seeking to enhance children's early environments. However, to date, the bulk of the attachment-based intervention research has focused on promoting infant attachment security via enhancement in maternal sensitivity, with comparatively less focus on mitigating risk for the development of disorganized attachment. Disorganized attachment is highly prevalent in groups with known social risks. A meta-analysis by Cyr, Euser, Bakermans-Kranenburg, and van IJzendoorn (Reference Cyr, Euser, Bakermans-Kranenburg and van IJzendoorn2010) demonstrated that children living in high- versus low-risk conditions (e.g., poverty an being parented by an adolescent) are at significantly increased risk of developing disorganized attachment (d = 0.77). Further, numerous studies have demonstrated that infants and children with disorganized attachment are at particular risk for psychopathology, stress dysregulation, and poor cognitive performance early in life (Bernard & Dozier, Reference Bernard and Dozier2010; Fearon, Bakermans-Kranenburg, van IJzendoorn, Lapsley, & Roisman, Reference Fearon, Bakermans-Kranenburg, van IJzendoorn, Lapsley and Roisman2010; Hertsgaard, Gunnar, Erickson, & Nachmias, Reference Hertsgaard, Gunnar, Erickson and Nachmias1995; Luijk et al., Reference Luijk, Saridjan, Tharner, van IJzendoorn, Bakermans-Kranenburg, Jaddoe and Tiemeier2010; Lyons-Ruth & Jacobvitz, Reference Lyons-Ruth, Jacobvitz, Cassidy and Shaver2016; Madigan, Brumariu, Villani, Atkinson, & Lyons-Ruth, Reference Madigan, Brumariu, Villani, Atkinson and Lyons-Ruth2016; Spangler & Grossmann, Reference Spangler and Grossmann1993; van IJzendoorn, Schuengel, & Bakermans-Kranenburg, Reference van IJzendoorn, Schuengel and Bakermans-Kranenburg1999). Recent work also indicates that disorganized attachment is an important predictor of disturbances in later adolescent and young adult development, including borderline features, suicidality, and dissociation (Carlson, Reference Carlson1998; Lyons-Ruth, Bureau, Holmes, Easterbrooks, & Brooks, Reference Lyons-Ruth, Bureau, Holmes, Easterbrooks and Brooks2013; Obsuth, Hennighausen, Brumariu, & Lyons-Ruth, Reference Obsuth, Hennighausen, Brumariu and Lyons-Ruth2014; but see Haltigan & Roisman, Reference Haltigan and Roisman2015). As a result, there has been a growing call for the development of interventions that target reductions in the prevalence of disorganized attachment.

Several randomized controlled trials (RCTs) have been conducted to assess whether attachment-based interventions can reduce the prevalence of disorganization (Bernard et al., Reference Bernard, Dozier, Bick, Lewis-Morrarty, Lindhiem and Carlson2012; Cicchetti, Rogosch, & Toth, Reference Cicchetti, Rogosch and Toth2006; Heinicke et al., Reference Heinicke, Goorsky, Moscov, Dudley, Gordon, Schneider and Guthrie2000; Juffer, Bakermans-Kranenburg, & van IJzendoorn, Reference Juffer, Bakermans-Kranenburg and van IJzendoorn2005; Moran, Pederson, & Krupka, Reference Moran, Pederson and Krupka2005; Moss et al., Reference Moss, Dubois-Comtois, Cyr, Tarabulsy, St.-Laurent and Bernier2011; Smyke, Zeanah, Fox, Nelson, & Guthrie, Reference Smyke, Zeanah, Fox, Nelson and Guthrie2010). These intervention programs have reduced the rate of attachment disorganization among infants with a wide range of risk characteristics, including poverty, maternal depression, maltreatment, and institutional rearing. Although all of these studies were based on an intervention model in which changes in parent–infant interaction were viewed as the mechanism through which infant attachment disorganization would be reduced, few studies have extended the design to statistically assess whether change in interaction is the mechanism contributing to change in child outcome.

Some prior studies have approached the issue of mechanism of change by assessing whether there are concomitant changes in maternal sensitivity in the intervention group compared to controls, and changes in maternal sensitivity have been demonstrated (Bakermans-Kranenburg, van IJzendoorn, & Juffer, Reference Bakermans-Kranenburg, van IJzendoorn and Juffer2003; Cicchetti et al., Reference Cicchetti, Rogosch and Toth2006; Dozier et al., Reference Dozier, Peloso, Lindhiem, Gordon, Manni, Sepulveda and Levine2006; Lyons-Ruth, Connell, Grunebaum, & Botein, Reference Lyons-Ruth, Connell, Grunebaum and Botein1990; Moran et al., Reference Moran, Pederson and Krupka2005; Osofsky et al., Reference Osofsky, Kronenberg, Hammer, Lederman, Katz, Adams and Hogan2007). In their meta-analysis of attachment-based interventions, Bakermans-Kranenburg et al. (Reference Bakermans-Kranenburg, van IJzendoorn and Juffer2003) reported that the studies with the largest effect sizes demonstrating changes in maternal sensitivity on account of the intervention (d = 0.40) were also the most effective in enhancing children's attachment security (d = 0.45), suggesting that changes in maternal sensitivity may be mediating the intervention effects on infant security.

To our knowledge, however, only one RCT has statistically evaluated possible processes contributing to changes in secure attachment. Cicchetti et al. (Reference Cicchetti, Rogosch and Toth2006) conducted an RCT in maltreating families in which families were assigned to infant–parent psychotherapy (IPP; Lieberman, Weston, & Pawl, Reference Lieberman, Weston and Pawl1991), an intensive psychoeducational parenting intervention (PPI; Olds, Henderson, Tatelbaum, & Chamberlin, Reference Olds, Henderson, Tatelbaum and Chamberlin1986), or to treatment as usual via the Department of Social Services. The results revealed that children in the IPP and PPI groups evinced higher rates of secure attachment from the pre- to posttest assessments, compared to the control group. The authors examined mediators of intervention efficacy relevant to each of the interventions (i.e., IPP or PPI) to determine whether specific agents of change mediated associations between pre- and posttest assessments of attachment. For IPP, changes in maternal representations and maternal sensitivity were examined as mediators, with no significant mediation shown. For PPI, changes in parenting attitudes and social support, as well as decreases in child-rearing stress, were examined as agents of change, again with no significant mediation shown. Thus, mechanisms of change in the IPP and PPI interventions could not be identified. Cicchetti et al. (Reference Cicchetti, Rogosch and Toth2006) thus encouraged further examination of mediators of intervention efficacy among at-risk dyads, in relation to both secure and disorganized attachment, to advance our understanding of how parent-focused interventions prevent attachment disorganization.

Maternal sensitivity is a robust predictor of child attachment security (De Wolff & van IJzendoorn, Reference De Wolff and van IJzendoorn1997; Verhage et al., Reference Verhage, Schuengel, Madigan, Fearon, Oosterman, Cassibba and van IJzendoorn2016), but it is not a strong predictor of attachment disorganization. Specifically, in the large NICHD Study of Early Child Care and Youth Development (NICHD Early Child Care Research Network, 1997), the association between sensitivity and disorganization was nonsignificant (r = .06), and in a meta-analysis by van IJzendoorn et al. (Reference van IJzendoorn, Schuengel and Bakermans-Kranenburg1999), it was significant, but weak (r = .10). However, more anomalous forms of parenting, as coded by protocols for frightened, frightening, and quality, of maternal behaviour (see below), have been robustly associated with disorganized attachment (r = .34; k = 12, N = 851; Madigan, Bakermans-Kranenburg, et al., Reference Madigan, Bakermans-Kranenburg, van IJzendoorn, Moran, Pederson and Benoit2006).

Among parents who have been exposed to multiple stressors or traumatic experiences, sensitive or responsive behaviors can become mixed with indicators of fear, threat, disorientation, role confusion, and withdrawing behavior. These maternal responses have relatively low base rates in normative samples, but become much more prevalent under conditions of poverty, exposure to violence or maltreatment, or in the context of maternal psychopathology (Lyons-Ruth, Bronfman, & Parsons, Reference Lyons-Ruth, Bronfman and Parsons1999; Madigan, Moran, & Pederson, Reference Madigan, Moran and Pederson2006). Thus, assessing frightened, frightening, or disrupted maternal communication may be an important next step in evaluating mechanisms of change in at-risk samples. In addition, in the Study of Early Child Care and Youth Development and in high-risk longitudinal studies (Dutra, Bureau, Holmes, Lyubchik, & Lyons-Ruth, Reference Dutra, Bureau, Holmes, Lyubchik and Lyons-Ruth2009; NICHD Early Child Care Research Network, 2001; Shi, Bureau, Easterbrooks, Zhao, & Lyons-Ruth, Reference Shi, Bureau, Easterbrooks, Zhao and Lyons-Ruth2012; Sroufe, Egeland, Carlson, & Collins, Reference Sroufe, Egeland, Carlson and Collins2005), quality of maternal behavior has been a stronger predictor of long-term outcomes over time than infant attachment, making it critical to assess the effectiveness of intervention in changing disturbed maternal behavior as well as infant disorganized attachment.

Observations of the disturbed interactions engaged in by mothers of disorganized infants led Main and Hesse (Reference Main, Hesse, Greenberg, Cicchetti and Cummings1990) to hypothesize that parental frightened or frightening behavior (FR) was tied etiologically to infant disorganization, due to the infant's experiencing “fright without solution.” That is, the caregiver was both the source of fear and the solution to fear, leading the infant to display the odd conflict behaviors characteristic of disorganization. Subsequently, Lyons-Ruth et al. (Reference Lyons-Ruth, Bronfman and Parsons1999) advanced a somewhat broader hypothesis that infant fear not directly provoked by the parent might also lead to infant disorganization if the parent is unable to respond to the infant's need for comfort and regulation of stressful arousal. These failures to comfort and regulate can be manifest in a variety of disrupted communications. For instance, the parent can give contradictory responses to the infant's cues or give self-referential responses that focus on the parent's needs rather than the infant's needs. The parent can also withdraw from the infant's cues for contact in subtle ways by responding in a cursory or reluctant manner or by failing to respond altogether. To capture this wider range of disrupted communication associated with infant disorganization, Lyons-Ruth et al. developed the Atypical Maternal Behavior Instrument for Assessment and Classification (AMBIANCE; Bronfman, Madigan, & Lyons-Ruth, Reference Bronfman, Madigan and Lyons-Ruth1992–2009). In a meta-analysis by Madigan, Bakermans-Kranenburg, et al. (Reference Madigan, Bakermans-Kranenburg, van IJzendoorn, Moran, Pederson and Benoit2006), both the narrower FR coding system developed by Main and Hesse (1992) and the broader AMBIANCE coding system (Bronfman et al., Reference Bronfman, Madigan and Lyons-Ruth1992–2009) were equivalent in their associations with infant disorganization (r _AMBIANCE = .35, r _FR = .32). In the current study, the AMBIANCE system was used in part because of strong evidence of stability over time, with the meta-analytic effect size for stability of the AMBIANCE over periods ranging from 10 to 72 months being r = .56 (k = 3, N = 203; Madigan, Bakermans-Kranenburg, et al., Reference Madigan, Bakermans-Kranenburg, van IJzendoorn, Moran, Pederson and Benoit2006). The AMBIANCE also has demonstrated strong predictive validity in that disrupted maternal communication predicts behavior problems from toddlerhood to age 20 years (Dutra et al., Reference Dutra, Bureau, Holmes, Lyubchik and Lyons-Ruth2009; Madigan, Moran, Schuengel, Pederson, & Otten, Reference Madigan, Moran, Schuengel, Pederson and Otten2007; Shi et al., Reference Shi, Bureau, Easterbrooks, Zhao and Lyons-Ruth2012), as well as altered adult amygdala volumes (Lyons-Ruth, Pechtel, Yoon, Anderson, & Teicher, Reference Lyons-Ruth, Pechtel, Yoon, Anderson and Teicher2016).

To date, one nonrandomized study has examined whether disrupted maternal communication, as assessed by the AMBIANCE, changes over the course of intervention. Benoit, Madigan, Lecce, Shea, and Goldberg (Reference Benoit, Madigan, Lecce, Shea and Goldberg2001) examined whether disrupted maternal communication was reduced in a sample of parent–child dyads referred for feeding problems. Mothers received a five-session modified version of interaction guidance (McDonough, Reference McDonough2000), which included a video-feedback component. Thirteen participants received the modified interaction guidance, and reductions in disrupted communication in this intervention group (from pre- to posttest) were compared to a convenience sample of 14 parent–child dyads, also referred for feeding problems, who received training in education techniques only (feeding-focused group). The results revealed a significant reduction in disrupted maternal communication in the modified interaction guidance intervention, but not the feeding-focused group, from pre- to posttest. This study highlighted a potential mechanism of treatment change, but did not address whether the change in the caregivers’ interactions with the child mediated measurable change in children's attachment classifications.

In this paper, we assess both infant disorganization and maternal disrupted behaviors, using the AMBIANCE system, as part of a larger randomized intervention trial, the Compétences parentales et Attachement dans la Petite Enfance: Diminution des risques liés aux troubles de santé mentale et Promotion de la resilience project (CAPEDP; Tubach et al., Reference Tubach, Greacen, Saïas, Dugravier, Guedeney, Ravaud and Guedeney2012), which aimed to improve developmental outcomes among impoverished families in a large metropolitan area in France. Families meeting eligibility for the CAPEDP project were those burdened with one or more psychosocial risks, such as qualification for social welfare due to low income, single parenthood, or poor educational attainment. The central aim of the current report is to address the question of mechanism by assessing whether reductions in disrupted maternal communication contributes to any observed reductions in infant disorganized attachment.

The CAPEPD intervention aimed to promote the development of secure attachment, while simultaneously reducing family contextual stress, disorganized infant attachment, and infant mental health issues. The CAPEPD study used a blended intervention model that combined rigorous home-visitation from 27 weeks of pregnancy through the child's second year of life that included routine prenatal care and well-baby care, rigorous follow-up, whole-family involvement, and health education counseling (e.g., breastfeeding, and alcohol and tobacco use; Olds et al., Reference Olds, Henderson, Tatelbaum and Chamberlin1986), with enhancement of sensitive parenting and reductions in disrupted parenting behaviors via video-feedback techniques. Approximately 367 women were enrolled in the CAPEDP study and were randomly assigned to the intervention or care as usual groups. A smaller ancillary study was developed to measure the impact of treatment on maternal and child attachment outcomes. This portion of the study involved only a subsample of those included in the larger CAPEDP project because of the labor, resource, and cost-intensive nature of attachment-based measurement tools. For the ancillary study, participants were asked in sequence to be part of this component, and the first 120 families to accept were enrolled. The objective of this substudy, known as the CAPEDP—Attachment (CAPEDP-A) study, was to assess the impact of the CAPEDP intervention in terms of increasing infant attachment security, reducing infant attachment disorganization, and reducing disrupted maternal communication. To our knowledge, the CAPEDP study is the first randomized intervention study to target the disrupted maternal communications that have been shown to be the strongest maternal predictors of infant attachment disorganization (see Table 1).

Table 1. Dimensions of disrupted communication assessed by the AMBIANCE coding system

Note: Adapted from “Maternal Frightened, Frightening, or Atypical Behavior and Disorganized Infant Attachment Patterns,” by K. Lyons-Ruth, E. Bronfman, and E. Parsons, Reference Lyons-Ruth, Bronfman and Parsons1999, Monographs of the Society for Research in Child Development, 64(3, Serial No. 258), pp. 67–96. Copyright 1999 by the Society for Research in Child Development. Adapted with permission.

In their comprehensive review of studies on disorganization, Lyons-Ruth and Jacobvitz (Reference Lyons-Ruth and Jacobvitz2016) noted that a significant gap in the literature exists regarding the mechanisms through which randomized interventions bring about a reduction in disorganized attachment. Important goals of RCTs are to identify outcomes that are causally related to particular interventions and to identify potential mediating mechanisms of change related to those interventions. Although causal implications of the intervention can be evaluated through random assignment to the intervention or treatment as usual group, causality associated with potential mediators of change cannot be derived from random assignment alone, even when it is also possible to randomize participants to different levels of the potential mediators (Pearl, Reference Pearl2014). To make causal claims about the role of the mediator on the outcome, counterfactually based definitions of direct and indirect effects have been developed (Pearl, Reference Pearl2014). These definitions are called counterfactuals because they are based on counterfactual conditions such as “If A were true, would C have been true?” Statistical software has recently been developed to implement this approach (Muthén & Asparouhov, Reference Muthén and Asparouhov2015), allowing for the examination of the causal role of the mediator. These statistical advances for testing mediation in RCTs may have important implications for policy because they can help identify the active ingredients of an intervention, which may help develop more focused and possibly less expansive interventions.

A major aim of the current study is to understand one mediating pathway that may explain how disorganized attachment is reduced in an intervention trial. Specifically, we examine whether the CAPEDP intervention results in a reduction in disrupted maternal communication compared to controls, which in turn contributes to a reduction in the incidence of disorganized attachment. In sum, the current study was designed to assess the following hypotheses: (a) participation in the intervention group, but not in the control group, would reduce the likelihood of disrupted maternal communication; (b) participation in the intervention group, but not in the control group, would reduce the likelihood of disorganized attachment; and (c) the effect of intervention on reducing disorganized attachment would be mediated by reductions in disrupted maternal communication.

Method

Participants

Pregnant women in the second trimester were recruited into CAPEDP through maternity check-ups at 10 public hospitals in the Paris area from December 2006 to March 2009. The hospitals, and specifically, the doctors and nurses who worked with this population, were asked to refer women who met the following criteria: (a) under 26 years of age; (b) first-time mothers who were less than 27 weeks pregnant; (c) able to speak French fluently; (d) living in the intervention area (Paris and its inner suburbs); and (e) presenting with one or more of the following psychosocial risk factors: having less than 12 years of education; sufficiently low enough income to qualify for French social welfare health insurance (Couverture Maladie Universelle Complémentaire), that is, with an income less than or equal to 850 euros a month or, for undocumented migrants, Government Medical Aid (Aide Médicale d’État); and intending to bring the child up without the involvement or support of the father. Women who were already actively involved in sustained treatment for mental health (e.g., addictions) or physical disorders requiring long-term follow-up were excluded from the study.

After completing baseline screening and informed consent procedures, participants were randomly assigned in a 1:1 ratio to either the CAPEDP intervention or the usual care group using a computer-generated randomization sequence, stratified by recruitment center (see Tubach et al., Reference Tubach, Greacen, Saïas, Dugravier, Guedeney, Ravaud and Guedeney2012, for additional details on randomization methods). At the maternity check-up, a total of 440 women agreed to participate in the study. Seventy-three of the 440 women (16.6% of the sample) withdrew their consent or could not be reached to schedule the baseline assessment visit.Footnote ¹ When children reached 12 months of age, all families participating in the main CAPEDP intervention trial were consecutively invited to take part in the CAPEDP-A assessment. The CAPEDP-A assessment was designed specifically to examine attachment outcomes based on intervention status. After receiving information about the study, and upon agreeing to participate with their child, mothers signed an informed consent form, and an appointment was given to them for a 2-hr assessment procedure within the following fortnight. Due to the time-intensive nature of coding attachment methodology, inclusion in the CAPEDP-A assessment was terminated when the required 120 mothers had agreed to participate with their child. However, due to technical problems with videotaping Strange Situations, only 117 participants were included in the final analyses. Mothers received 50 euros as a stipend for participating in the CAPEDP-A assessment. There were no significant differences in sociodemographic factors in those who accepted versus declined and/or were not asked to participate in the CAPEP-A substudy (see Table 2). CAPEDP-A received ethical approval from the Comité de Protection des Personnes Ile de France IV, Institutional Review Board, and from the Commission Nationale de l'Informatique et des Libertés (CNIL, 907255), with Clinical Trial Registration Number: NCT00392847.

Table 2. Comparison between general CAPEDP and CAPEDP-A risk factors at inclusion

Procedures

At the baseline assessment visit, during the 27th week of pregnancy, demographic and mental health data were collected. For the CAPEDP-A mothers, their mean age was 22.3 years (SD = 2.5; range = 16–26), 84.9% had less than 12 years of school education (15.1% had less than 9 years), 42.4% had sufficiently low income to be eligible for government medical aid (CMU or AME), and 24.4% participants declared they were planning to bring up their child without the support of the infant's father. In addition, 32.4% of the CAPEDP-A sample considered themselves to be poor; 40.3% were single; 31.9% had not planned their current pregnancy; 39.5% had been pregnant at least once before, but their previous pregnancies had been interrupted; and 52.1% were first-generation immigrants. No significant baseline differences were found between the intervention and care as usual groups on sociodemographic variables (for additional details, see Dugravier et al., Reference Dugravier, Tubach, Saias, Guedeney, Pasquet and Purper-Ouakil2013).

Following the initial baseline assessment, all participants engaged in a scheduled series of assessment visits that occurred when the infant was approximately 3, 6, 12, 18, and 24 months of age. Evaluators were not informed as to whether the family being evaluated was in the intervention group or the control group. When infants were between 12 and 17 months of age, all mothers who agreed to participate in the CAPEP-A study participated in a 2-hr attachment-focused assessment procedure.

Measures

AMBIANCE

The AMBIANCE (Bronfman et al., Reference Bronfman, Madigan and Lyons-Ruth1992–2009) codes disrupted caregiver behavior and communication toward the infant during videotaped caregiver–infant interactions. After a written account of all instances of disrupted maternal communication observed during the Strange Situation (Ainsworth, Blehar, Waters, & Wall, Reference Ainsworth, Blehar, Waters and Wall1978), a frequency score was derived for each of the five dimensions of the AMBIANCE scale: affective communication errors, role/boundary confusion, fearfulness/disorientation, intrusiveness/negativity, and withdrawing behavior. The AMBIANCE coding system also involves a continuous 7-point scale assessing the global level of disrupted communication in the video. The global level of disrupted communication is assigned by the coder based on the frequency and intensity of disrupted maternal communications displayed by the caregiver, where 1 = warm and sensitive communication, 3 = generally positive interaction with some evidence of disrupted communication, 5 = clear and repeated disruption in affective communication, and 7 = disrupted communication with few or no ameliorating behaviors. A binary classification is then assigned, with scores of 5 or above classified as “disrupted” and scores of less than 5 as “not-disrupted” (Bronfman et al., Reference Bronfman, Madigan and Lyons-Ruth1992–2009). For the AMBIANCE scale, four coders were trained by the AMBIANCE developers (E. Bronfman, S. Madigan, and K. Lyons-Ruth), all obtaining reliability (>80% concordance on the disrupted vs. nondisrupted classification).

Strange Situation Procedure (SSP)

The SSP (Ainsworth et al., Reference Ainsworth, Blehar, Waters and Wall1978) is a laboratory paradigm with a series of eight 3-min, increasingly stressful episodes, for 12- to 18-month-olds. The SSP is videotaped, and infant behavior is coded using four 7-point anchored scales for proximity seeking, contact maintaining, avoidance, and resistance and one 9-point scale for disorganization (Main & Solomon, Reference Main, Solomon, Greenberg, Cicchetti and Cummings1990). For the secure, avoidant, and ambivalent-resistant SSP classifications, two coders were trained by Karin Grossman and Fabienne Becker-Stoll, obtaining reliability (i.e., >85% classification concordance). Fifty percent of the SSPs were coded by a second independent rater, and intercoder reliability for the three-way classification was 85%. An expert coder who, blind to intervention group status, coded all SSPs, coded for disorganization.Footnote ² The concordance for the disorganized versus not-disorganized classification (N = 18) between the primary coder and a second trained and reliable coder was 86.3%.

The intervention program and care as usual

In France, a national mother–child support and prevention network, the Protection Maternelle et Infantile (PMI; i.e., Maternal and Infant Protection), was implemented after World War II, parallel to the creation of free public mental health services for both adults and children across the country. At present, mothers can consult PMI centers free of charge at any point in time from pregnancy to 3 years postpartum. However, PMI nurses do not receive specific training in mental health issues for mothers and children and receive little organized psychological supervision (DASES, 2003). The absence of targeted assistance with attachment-relevant and mental health issues served as the impetus for the development of the CAPEDP study. The CAPEDP is the first large-scale French intensive home-visiting program that specifically aims to enhance the life outcomes of children and families.

The CAPEDP was developed in line with international best practice criteria (Daro, McCurdy, Falconnier, & Stojanovic, Reference Daro, McCurdy, Falconnier and Stojanovic2003; Durlak & DuPre, Reference Durlak and DuPre2008; Gomby, Reference Gomby2005; Kahn & Moore, Reference Kahn and Moore2010) and was adapted to suit the particularities of the French population. The CAPEDP program had two major features that differentiated it from most other home-visiting programs targeting mental health promotion. The first feature was that it addressed child mental health promotion in families who already have free access to one of the most extensive, comprehensive, and longstanding social and health care systems in the Western world. The second differentiating feature was that qualified psychologists conducted the entire home-visiting program.

The CAPEDP intervention (Saïas et al., Reference Saïas, Greacen, Tubach, Dugravier, Marcault and Tereno2013; Tereno et al., Reference Tereno, Guedeney, Dugravier, Greacen, Saïas, Tubach and Guédeney2013; Tubach et al., Reference Tubach, Greacen, Saïas, Dugravier, Guedeney, Ravaud and Guedeney2012) sought, where possible, to act upon the major modifiable determinants of infant mental health from the third trimester of pregnancy to the child's second birthday. Psychologists visited families an average of 6 times during the prenatal period (starting from the 27th week of pregnancy), eight times in the first 3 months of the child's life, 15 times when the child was between 4 and 12 months of age, and another 15 times during the child's second year, resulting in a total of 44 home visits during the entire intervention. Each session was approximately 60 min in duration. Additional details regarding the content of the manualized CAPEDP intervention can be found elsewhere (Saïas et al., Reference Saïas, Greacen, Tubach, Dugravier, Marcault and Tereno2013).

The CAPEDP home-visiting program used an intervention manual based on Services Intégrés pour la Périnatalité et la Petite Enfance (Integrated Services for Perinatal Health and Early Childhood), a Canadian adaptation of the Nurse Family Partnership intervention program. The manual proposes 39 different intervention brochures to be used during home visits. Each brochure addresses a specific health or mental health topic, based on a common theme of promoting quality mother–child relationships. The intervention manual also drew from Weatherston's guidelines concerning the most practical aspects of home-visiting, as well as best practice recommendations for home visiting from the Florida State University Center for Prevention and Early Intervention Policy. The manual offers a series of reference points for addressing different topics at different periods during the intervention (prenatal, 0–6 months, 6–15 months, and 15–24 months). Each intervention was based around four themes: the family and their social and cultural network; the mother's needs and health; creating a safe and stimulating environment for the baby; and the baby's development. The psychologists were instructed to adapt their interventions to the needs of each family being visited, as well as to encourage the family to make the most of available PMI centers, social services, and community resources in general.

Home-visiting psychologists were also provided with a set of items that they could discuss with the families being visited: (a) family information brochures, each addressing a specific topic and designed to facilitate discussion and to be left with the family; (b) a series of six DVDs including short films on pregnancy, childcare, and child development, which were used by the home-visiting psychologists to facilitate forming a working alliance with each family, particularly in the antenatal period; and (c) a comprehensive document, collated by the research team, on promoting infant emotional development and mother–child attachment quality. Furthermore, the home-visiting team systematically proposed, with the mothers’ approval, to film short sequences of everyday interactions between the mothers and their children: bath time, mealtime, play, changing nappies, and so on. During subsequent visits, mothers would watch the video and discuss what they saw with the home-visiting psychologist.

Specific tools were developed to support the intervention team concerning attachment issues (Tereno et al., Reference Tereno, Guedeney, Dugravier, Greacen, Saïas, Tubach and Guédeney2013). All home-visiting interveners were psychologists. Each home visitor was provided with two manuals, each aimed at structuring their interventions, with one focusing on attachment issues and the other on oppositional behavior. A brochure drafted by the research team and focusing on the emotional development of young children was also provided to intervention families with the aim of making them aware of the importance of early attachment bonds. Video feedback was used to encourage parents to reflect on their parenting practices and experiences. The overall aim was to increase maternal sensitivity (detecting infants’ signals, interpreting them appropriately, and providing adequate and rapid responses to infants’ demands) and mentalizing skills (identifying their own and their infants’ intentions and emotional states). Furthermore, the intervention sought to decrease disrupted maternal communication (detecting, preventing, and repairing disrupted maternal affective communication) and infant disorganized behavior (detecting and emotionally repairing disorganized behavior). When the attachment assessment took place at about the children's first birthday, families in the intervention group had on average received 29 home visits.

All home-visiting psychologists received specific training before the beginning of the intervention. This 6-day program involved didactical training on particular subjects as well as pedagogical activities to develop intervention skills. Principal topics addressed included (a) perinatal health and mental health, (b) the art of home visiting, (c) the specific contents of the CAPEDP intervention manual, and (d) the inclusion process. Specific training was also provided on the use of video for helping parents reflect on their parenting experiences and practices and promoting attachment security in infants from 6 to 9 months old and handling oppositional behaviors in children 12–15 months old. To ensure treatment fidelity, all home-visiting psychologists received weekly individual supervision by senior clinicians, and group supervision by the principal investigator on a fortnightly basis. A trained and highly skilled project manager supervised fidelity to the intervention manual (see Saias et al., Reference Saias, Lerner, Greacen, Simon-Vernier, Emer and Pintaux2012, for further details on fidelity).

Care as usual

On a nationwide basis, France provides community-based mother–child support and prevention services with no out-of-pocket payment, known as the Protection maternelle et infantile (PMI; Mother and Child Protection Services). In addition to having direct access to these PMI centers from the beginning of pregnancy through to their children's third birthday, families can also access free community mental health services for both parents and children. Furthermore, local government also provides family benefits aimed at helping families raise their children, on the condition that they agree to bring them in for regular health check-ups and compulsory vaccinations. PMI nurses also make home visits to families identified at the maternity hospital as being in particularly difficult circumstances, although very few families end up receiving more than one such home visit. Although PMI nurses receive little specific training on mental health prevention, they can refer families directly to local community child and adolescent mental health services, which provide home visits if deemed necessary for the child's mental health or safety.

Data analyses

Chi-square tests (categorical measures: i.e., organized vs. disorganized) were examined in Statistical Package for the Social Sciences (SPSS version 22.0) to test our hypotheses that mothers’ participation in the intervention group, but not in the control group, would reduce the likelihood of (a) the mother's displaying disrupted maternal communication and (b) the infant displaying disorganized attachment. The third mediational hypothesis was tested in MPlus 7.2 (Muthén & Muthén, Reference Muthén and Muthén1998–2012). Using counterfactual definitions, the total effect of the intervention on the outcome can be decomposed into two components (Muthén & Asparouhov, Reference Muthén and Asparouhov2015; Pearl, Reference Pearl2014). The first component is the pure natural direct effect (PNDE), which describes the effect of the intervention if it was administered but its effect on the mediator was blocked. Stated differently, it is the effect of the intervention on the outcome if the mediator is kept at the level it would have been in the absence of exposure to the intervention. Thus, the PNDE captures the effect of the intervention that is not accounted for by the mediator. The second component is the total natural indirect effect (TNIE), which describes the change in the outcome among those not exposed to the intervention that is due to the difference in the level of the mediator between those who are exposed to the intervention versus those who are not. Thus, this component indexes the effect of the intervention on the outcome that is explained by the average difference in the mediator between the control and the intervention groups.

Results

Preliminary analyses

As expected due to randomization, there were no significant differences between the intervention and control groups on demographic variables, including maternal age, and child age and gender (see Table 3).

Table 3. Sample demographics across intervention and control groups

^a There were no significant differences in sociodemographic variables between the intervention and care as usual groups.

Intervention effects

Disrupted maternal behavior

At the attachment outcome assessment at 13 months, 40% of mothers were classified as having disrupted maternal communication and 57% were classified not disrupted. There were significant differences between the intervention and the control groups on classifications of disrupted maternal communication, χ² (1) = 4.80, p < .05; odds ratio (OR) = 0.43; 95% confidence interval (CI) [0.20, 0.92]. In the intervention group, the percentage of disrupted maternal communication was 31.7% compared to 51.9% in the control group.

Disorganized mother–child attachment

Of mother–child dyads, 13.6% dyads were classified as disorganized attachment relationships. However, there were significant differences between the intervention and the control groups on classifications of disorganized attachment, χ² (1) = 4.44, p < .05; OR = 0.31; 95% CI [0.10, 0.96]; see Table 4. In the intervention group, the percentage of disorganized attachment was 7.7% compared to 21.2% in the control group.

Table 4. Distribution of disrupted versus nondisrupted AMBIANCE classifications and disorganized versus organized Strange Situation classifications by treatment group

Mediation model

Next, we tested a mediation model through which intervention status predicted infant disorganized attachment in the SSP via disrupted maternal communication. In this analysis, we controlled for infant age and gender, as well as maternal age. There was an association between intervention status and disrupted maternal communication, as well as between disrupted maternal communication and infant attachment. We then computed the PNDE and the TNIE to allow us to make causal inferences about the role of the mediator on the outcome (Muthén & Asparouhov, Reference Muthén and Asparouhov2015). We report the odds ratio and their 95% CI, which can be used to determine their statistical significance (i.e., it is significant if the 95% CI does not include 1).

The results indicated that the CAPEDP intervention led to significantly lower odds of children showing attachment disorganization as the total effect was significant, OR = 0.32, 95% CI [–0.05, 0.69]. This total effect can be decomposed into two components: the effect of the intervention on the outcome that is not explained by the mediator (PNDE) and the effect of the intervention on the outcome that is explained by the mediator (TNIE). The PNDE was significant, OR = 0.45, 95% CI [–0.06, 0.96], Cohen d = 0.44, indicating that the intervention led to decreased odds of children having disorganized attachment via mechanisms other than disrupted maternal communication. The TNIE was also significant, OR = 0.71, 95% CI [0.44, 0.97], Cohen d = 0.63, indicating that disrupted maternal communication partially explained why the intervention led to lower attachment disorganization. A more detailed understanding of the role of disrupted maternal communication can be obtained by examining the proportion of the total effect of the intervention on the outcome (b = –0.12 before its conversion into an odds ratio) that is explained by the indirect effect of the mediator on the outcome (b = –0.03 before its conversion into an odds ratio). Specifically, this ratio indicated that disrupted maternal communication explained 25% (–0.03/–0.12) of the effect of the intervention on disorganized attachment.

Because the mediator and the outcome were measured contemporaneously, we also assessed the reverse mediation model, entering disorganization status as mediator and disrupted communication as outcome, and controlling for infant age and gender, and maternal age. As expected, the model did not show significant mediation. Specifically, in this model, the total natural indirect effect was not significant, OR = 0.79, 95% CI [0.56, 1.03]. These results demonstrate discriminant validity, further supporting the model proposed here: disrupted maternal communication partially mediates the relation between treatment and disorganization.