Participant characteristics

Table 2 presents participant characteristics. No studies started during pregnancy. All women were included based on specific inclusion criteria for level of depressive symptoms. Four studies used the EPDS,^{Reference Goodman, Prager, Goldstein and Freeman55–Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58} one study used the EPDS or a DSM-II-R major depressive disorder diagnosis,^{Reference Murray, Cooper, Wilson and Romaniuk52} one study used a diagnosis of major depressive disorder,^{Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} and one study used the Beck Depression Inventory or DSM-IV major depressive episode or dysthymia diagnosis^{Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59,Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60} as inclusion criteria. The mean age of the mothers at inclusion ranged from 27.7 to 32 years and mean age of the infant between 1 and 7 months. Two studies included primiparous mothers only,^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Goodman, Prager, Goldstein and Freeman55} whereas the other five included both primiparous and multiparous mothers.^{Reference Horowitz, Bell, Trybulski, Munro, Moser and Hartz56,Reference Horowitz, Murphy, Gregory, Wojcik, Pulcini and Solon57,Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59–Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61,Reference Gilmer, Buchan, Letourneau, Bennett, Shanker and Fenwick66} In most studies, the majority of participants were White, married/living with partner, and with a medium or long education. All studies were relatively small, ranging from 42 to 190 participants.

Table 2 Participant characteristics

EPDS, Edinburgh Postnatal Depression Scale; BDI, Beck Depression Inventory.

Intervention characteristics

Table 3 presents intervention characteristics, assessment times and outcomes. All included studies offered individual home visits initiated postpartum. No studies offered a group-based intervention. Most interventions focused on supporting the parent–child relationship through for example video feedback, coaching or therapy. Four studies offered alternative treatment such as progressive muscle relaxation, home visits or phone calls to the control group.^{Reference Goodman, Prager, Goldstein and Freeman55,Reference Horowitz, Murphy, Gregory, Wojcik, Pulcini and Solon57,Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59,Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} One study also provided CBT for both the intervention and control group.^{Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} Control condition was usual care in two studies^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Horowitz, Bell, Trybulski, Munro, Moser and Hartz56} and waiting list in one study where participants received usual care while waiting.^{Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58} Besides one intervention lasting 7.5 months,^{Reference Horowitz, Bell, Trybulski, Munro, Moser and Hartz56} all interventions were relatively short (3–15 weeks) and the intensity ranged between weekly visits to one visit a month.

Table 3 Intervention characteristics, assessment times and outcomes

Goodman and colleagues (2015) examined the effects of Perinatal Dyadic Psychotherapy among 42 mothers recruited from postpartum units of three hospitals located in the USA.^{Reference Goodman, Prager, Goldstein and Freeman55} The study was a pilot study to examine a novel dual-focused mother–infant intervention aimed at promoting maternal mental health and improving the relationship between mother and infant. The intervention was derived from the mutual regulation model by Tronick,^{Reference Tronick67} and integrates clinical strategies of supportive psychotherapy, parent–infant psychotherapy, the touchpoints model of child development^{Reference Brazelton and Sparrow68} and the newborn behavioral observation.^{Reference Goodman, Prager, Goldstein and Freeman55,Reference Nugent, Keefer, Minear, Johnson and Blanchard69}

Horowitz and colleagues (2001) examined the effects of an interactive coaching intervention among 117 mothers from greater Boston in the USA.^{Reference Horowitz, Bell, Trybulski, Munro, Moser and Hartz56} The aim of the intervention was to improve responsiveness between infant and mother. The intervention is based on Beck's cognitive model of depression,^{Reference Beck70} Sameroff's transactional model of child development^{Reference Sameroff and Zeanah71} and Rutter's model of developmental risk and resilience.^{Reference Horowitz, Bell, Trybulski, Munro, Moser and Hartz56}

Horowitz and colleagues (2013) examined the effects of the behavioural coaching intervention communicating and relating effectively on 134 mother–infant dyads in the USA.^{Reference Horowitz, Murphy, Gregory, Wojcik, Pulcini and Solon57} The intervention aimed to improve the mother–infant relationship. The intervention is based on cognitive–behavioural family therapy theory.^{Reference Horowitz, Murphy, Gregory, Wojcik, Pulcini and Solon57}

Letourneau and colleagues (2011) examined the effects of home-based peer support among 60 mothers in Canada.^{Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58} The home-based peer support included mother–infant interaction teaching as an important element and the intervention aimed to improve interactions between mothers and their infants. The intervention is not based on any specific theory, but the authors refer to concepts from studies on mothers with postpartum depression and peer support.^{Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58}

Murray and colleagues (2003) examined the effects of three different, but related, interventions: (a) CBT, (b) psychodynamic therapy and (c) non-directive counselling in one trial in the UK.^{Reference Murray, Cooper, Wilson and Romaniuk52} In total, 190 first-time mothers were included and almost equally distributed among the three intervention groups and the single control group. The interventions all aimed to improve the mother–child relationship. The CBT intervention was a modified form of interaction guidance treatment. In the psychodynamic therapy intervention, the mother's representations were used to explore her own attachment history. The non-directive counselling intervention provided the mothers with an opportunity to talk about their feelings about current concerns.^{Reference Murray, Cooper, Wilson and Romaniuk52}

Stein and colleagues (2018) examined the effects of video-feedback therapy among 144 mothers in the UK. The aim of the intervention was to improve parenting behaviours (attention to infant cues, emotional scaffolding and sensitivity). Both intervention and control families also received CBT with a focus on behavioural activation. The control group received progressive muscle relaxation that did not target parent practices or parent–child interaction.^{Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61}

Van Doesum and colleagues (2008) examined the effects of a mother–baby intervention among 71 mothers in the Netherlands.^{Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59,Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60} The intervention aimed to improve the mother–infant interaction, particularly maternal sensitivity. The intervention was based on video feedback, modelling, cognitive restructuring, practical pedagogical support and baby massage.^{Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59,Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60}

Outcome characteristics

All seven studies examined parent–child relationship outcomes. Measures applied to assess parent–child relationship (for example attachment or parent–child relationship) included the Emotional Availability Scales (EAS),^{Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59} Coding Interactive Behavior (CIB),^{Reference Goodman, Prager, Goldstein and Freeman55} the Nursing Child Assessment Teaching Scale^{Reference Horowitz, Murphy, Gregory, Wojcik, Pulcini and Solon57} and the Dyadic Mutuality Code.^{Reference Horowitz, Bell, Trybulski, Munro, Moser and Hartz56} Four studies assessed child development outcomes (such as socioemotional or cognitive development) with measures such as the Infant-Toddler Social Emotional Assessment (ITSEA),^{Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59} the Child Behavior Checklist (CBCL)^{Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60,Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} and Bayley Scales of Infant Development (BSID).^{Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} Outcomes were assessed by independent assessors (for example EAS, CIB, and BSID), by parents (for example ITSEA and CBCL), and/or teachers (for example CBCL). All seven studies included a post-intervention assessment, but only four studies included a follow-up assessment.^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Goodman, Prager, Goldstein and Freeman55,Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59–Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} As the development of maternal depressive symptoms is not the focus of this review, we did not include maternal depression in our analyses, although it was reported in all studies.

Risk of bias

Risk of bias assessments for each study's outcomes are displayed in online supplementary Table 1 divided into parent–child relationship and child development outcomes. Six out of seven studies provided insufficient information for one or more risk of bias domain, thus hindering a clear risk of bias judgement. Only two studies^{Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60,Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} provided an accessible a priori protocol, which made it particularly difficult to assess risk of bias caused by selective reporting. In general, risk of bias ranged between low and medium. Five studies had outcomes where one or two domains were classified as medium risk of bias.^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Goodman, Prager, Goldstein and Freeman55,Reference Horowitz, Murphy, Gregory, Wojcik, Pulcini and Solon57,Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58,Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} Only one study had outcomes with a high risk of bias in one domain: ‘incomplete outcome data addressed’.^{Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60} One study was excluded from the review because of an unacceptably high risk of bias caused by lack of assessor masking and high risk of bias in relation to ‘incomplete outcome data addressed’.^{Reference Ericksen, Loughlin, Holt, Rose, Hartley and Buultjens65}

The outcomes included in the meta-analysis on the parent–child relationship were characterised by low-to-medium and unclear risk of bias domain. In three of these studies,^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Horowitz, Bell, Trybulski, Munro, Moser and Hartz56,Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59} the risk of bias domains of the outcomes were assessed as relatively low (1–2) or unclear risk of bias. The outcomes of the remaining three studies^{Reference Goodman, Prager, Goldstein and Freeman55,Reference Horowitz, Murphy, Gregory, Wojcik, Pulcini and Solon57,Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58} were characterized by low-to-medium and unclear risk of bias.

Parent–child relationship

Seven studies reported on parent–child relationship outcomes.^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Goodman, Prager, Goldstein and Freeman55–Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59,Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} supplementary Table 2 presents the study outcomes for the individual studies. Owing to the lack of follow-up assessments, only meta-analysis of the parent–child relationship post-intervention was conducted.

Post-intervention parent–child relationship

The meta-analysis of the effect of the parenting intervention on the parent–child relationship at post-intervention included 573 participants from six studies and is presented in Fig. 2.^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Goodman, Prager, Goldstein and Freeman55–Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59} No significant effect of the parenting interventions on the parent–child relationship was found (d = 0.028, 95% CI −0.30 to 0.31) (I ² = 49.02). One study^{Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58} that provided a peer-support intervention instead of using professional providers was removed for sensitivity analysis. This did not alter the result substantially. Likewise, three studies offering alternative treatment for the control group^{Reference Goodman, Prager, Goldstein and Freeman55,Reference Horowitz, Murphy, Gregory, Wojcik, Pulcini and Solon57,Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59} were removed for sensitivity analysis. The result of the meta-analysis of the three remaining studies that provided treatment as usual^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Horowitz, Bell, Trybulski, Munro, Moser and Hartz56,Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58} was not altered substantially. All six studies were therefore kept in the analysis.

Fig. 2 Meta-analysis of studies reporting parent–child relationship at post-intervention.

One out of the two parent–child relationship outcomes in the study by Murray et al ^{Reference Murray, Cooper, Wilson and Romaniuk52} was based on a parent-reported questionnaire. As all other outcomes included in the meta-analysis were observational measures, this outcome was not included in the analysis. When examining relationship problems, Murray et al (2003) found significant positive effects of counselling, psychodynamic therapy and CBT compared with usual care (counselling: RR = 0.63, 95% CI 0.32–0.97); psychodynamic: RR = 0.57, 95% CI 0.28–0.92); cognitive–behavioural: RR = 0.46, 95% CI 0.20–0.81).^{Reference Murray, Cooper, Wilson and Romaniuk52}

Parent–child relationship at short-term follow-up

Two studies reported on the parent–child relationship at short-term follow-up.^{Reference Goodman, Prager, Goldstein and Freeman55,Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59} Van Doesum and colleagues^{Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59} found significant effects of the mother–infant intervention on maternal sensitivity (d = 0.82, 95% CI 0.34–1.31), maternal structuring (d = 0.57, 95% CI 0.09–1.04), child responsiveness (d = 0.69, 95% CI 0.21–1.16) and child involvement (d = 0.75, 95% CI 0.27–1.23) at short-term follow-up when the children were around 18 months old. No significant effects were found on child attachment security, maternal non-intrusiveness or maternal non-hostility.^{Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59} Goodman and colleagues likewise measured maternal sensitivity and infant involvement but found no significant effects on maternal sensitivity, infant involvement or dyadic reciprocity when the children were around 8 months old.^{Reference Goodman, Prager, Goldstein and Freeman55}

Parent–child relationship at long-term follow-up

Three studies reported on the parent–child relationship at long-term follow-up.^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60,Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} Meta-analysis was not conducted, as the study by Stein and colleagues used an active control group, whereas the control groups in the two remaining studies received either alternative treatment or treatment as usual. Kersten-Alvarez and colleagues found no effects on maternal interactive behaviour or attachment security when the children were 4 years old.^{Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60} Murray and colleagues found that none of the three interventions provided had a significant effect on child attachment at long-term follow-up (child aged 18 months) when compared with the control group.^{Reference Murray, Cooper, Wilson and Romaniuk52} Stein and colleagues found no significant effects on attachment security for children aged 2 years.^{Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61}

Child development

Four studies reported on child development outcomes. Supplementary Table 3 presents the study outcomes for the individual studies. Owing to the lack of developmental outcomes and follow-up assessments, it was not possible to conduct meta-analysis of child development.

Post-intervention child development

Two studies examined child development post-intervention.^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58} Letourneau and colleagues found no significant effect on child cognitive and socioemotional development.^{Reference Letourneau, Stewart, Dennis, Hegadoren, Duffett-Leger and Watson58} Murray and colleagues found that none of the three interventions provided had a significant effect on infant behaviour problems post-intervention.^{Reference Murray, Cooper, Wilson and Romaniuk52}

Child development at short-term follow-up

One study, van Doesum and colleagues, examined child development at short-term follow-up.^{Reference Van Doesum, Riksen-Walraven, Hosman and Hoefnagels59} They found a significant effect of home visits on child competence behaviour (d = 0.62, 95% CI 0.14–1.10), but no significant effects on externalising, internalising or dysregulated behaviour.^{Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60}

Child development at long-term follow-up

Three studies examined child development at long-term follow-up (range: 13–56 months).^{Reference Murray, Cooper, Wilson and Romaniuk52,Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60,Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61} Murray and colleagues^{Reference Murray, Cooper, Wilson and Romaniuk52} found significant positive effects of the counselling intervention (d = 0.64, 95% CI 0.22–1.05) and the psychodynamic intervention (d = 0.49, 95% CI 0.07–0.91) on emotional and behavioural problems when the children were 18 months old, but no effects of the cognitive–behavioural intervention at 18 months old. When the children were 60 months old, a positive significant effect of the cognitive–behavioural intervention on mother-rated emotional and behavioural difficulties (d = 0.61, 95% CI 0.12–1.11) was found, but not on teacher-rated emotional and behavioural difficulties. No significant effects of the two other interventions were found. None of the three interventions showed significant effects on child cognitive development at any of the two follow-up assessments. Kersten-Alvares and colleagues found no significant effects of the intervention on child self-esteem, verbal intelligence, prosocial behaviour, school adjustment and behaviour problems at long-term follow-up.^{Reference Kersten-Alvarez, Hosman, Riksen-Walraven, Van Doesum and Hoefnagels60} Stein and colleagues found no significant effects on cognitive and language development, behaviour problems, attention focusing, attentional shifting, inhibitory control and child emotion regulation.^{Reference Stein, Netsi, Lawrence, Granger, Kempton and Craske61}