Abstract
This co-registered Cochrane and Campbell systematic review compares the effectiveness of parent-infant psychotherapy (PIP) in improving the mental health of parents, the parent-infant relationship and infant emotional wellbeing with that of no treatment, treatment as usual and alternative treatments. The review summarises findings from eight randomised controlled trials (RCTs), with a total of 846 participants.
Compared with no treatment or treatment-as-usual, PIP participants show an increase in the number of infants securely attached; fewer infants with an avoidant or disorganised attachment style; and more infants moving from insecure to secure attachment. This was maintained at follow up.
PIP is neither more nor less effective than no treatment or treatment-as-usual in improving maternal mental health, and reflective functioning; infant stress and development; or the quality of the mother-infant interaction.
Parent-infant psychotherapy (PIP) is a dyadic intervention that works with parent and infant together, with the aim of improving the parent-infant relationship and promoting infant attachment and optimal infant development. PIP aims to achieve this by targeting the mother's view of her infant, which may be affected by her own experiences, and linking them to her current relationship to her child, in order to improve the parent-infant relationship directly.
Abstract
BACKGROUND
Parent-infant psychotherapy (PIP) is a dyadic intervention that works with parent and infant together, with the aim of improving the parent-infant relationship and promoting infant attachment and optimal infant development. PIP aims to achieve this by targeting the mother's view of her infant, which may be affected by her own experiences, and linking them to her current relationship to her child, in order to improve the parent-infant relationship directly.
OBJECTIVES
To assess the effectiveness of PIP in improving parental and infant mental health and the parent-infant relationship. To identify the programme components that appear to be associated with more effective outcomes and factors that modify intervention effectiveness (e.g. programme duration, programme focus).
SEARCH METHODS
We searched the following electronic databases on 13 January 2014: Cochrane Central Register of Controlled Trials (CENTRAL, 2014, Issue 1), Ovid MEDLINE, EMBASE, CINAHL, PsycINFO, BIOSIS Citation Index, Science Citation Index, ERIC, and Sociological Abstracts. We also searched the metaRegister of Controlled Trials, checked reference lists, and contacted study authors and other experts.
SELECTION CRITERIA
Two review authors assessed study eligibility independently. We included randomised controlled trials (RCT) and quasi-randomised controlled trials (quasi-RCT) that compared a PIP programme directed at parents with infants aged 24 months or less at study entry, with a control condition (i.e. waiting-list, no treatment or treatment-as-usual), and used at least one standardised measure of parental or infant functioning. We also included studies that only used a second treatment group.
DATA COLLECTION AND ANALYSIS
We adhered to the standard methodological procedures of The Cochrane Collaboration. We standardised the treatment effect for each outcome in each study by dividing the mean difference (MD) in post-intervention scores between the intervention and control groups by the pooled standard deviation. We presented standardised mean differences (SMDs) and 95% confidence intervals (CI) for continuous data, and risk ratios (RR) for dichotomous data. We undertook meta-analysis using a random-effects model.
MAIN RESULTS
We included eight studies comprising 846 randomised participants, of which four studies involved comparisons of PIP with control groups only. Four studies involved comparisons with another treatment group (i.e. another PIP, video-interaction guidance, psychoeducation, counselling or cognitive behavioural therapy (CBT)), two of these studies included a control group in addition to an alternative treatment group. Samples included women with postpartum depression, anxious or insecure attachment, maltreated, and prison populations. We assessed potential bias (random sequence generation, allocation concealment, incomplete outcome data, selective reporting, blinding of participants and personnel, blinding of outcome assessment, and other bias). Four studies were at low risk of bias in four or more domains. Four studies were at high risk of bias for allocation concealment, and no study blinded participants or personnel to the intervention. Five studies did not provide adequate information for assessment of risk of bias in at least one domain (rated as unclear).
Six studies contributed data to the PIP versus control comparisons producing 19 meta-analyses of outcomes measured at post-intervention or follow-up, or both, for the primary outcomes of parental depression (both dichotomous and continuous data); measures of parent-child interaction (i.e. maternal sensitivity, child involvement and parent engagement; infant attachment category (secure, avoidant, disorganised, resistant); attachment change (insecure to secure, stable secure, secure to insecure, stable insecure); infant behaviour and secondary outcomes (e.g. infant cognitive development). The results favoured neither PIP nor control for incidence of parental depression (RR 0.74, 95% CI 0.52 to 1.04, 3 studies, 278 participants, low quality evidence) or parent-reported levels of depression (SMD - 0.22, 95% CI -0.46 to 0.02, 4 studies, 356 participants, low quality evidence). There were improvements favouring PIP in the proportion of infants securely attached at post-intervention (RR 8.93, 95% CI 1.25 to 63.70, 2 studies, 168 participants, very low quality evidence); a reduction in the number of infants with an avoidant attachment style at post-intervention (RR 0.48, 95% CI 0.24 to 0.95, 2 studies, 168 participants, low quality evidence); fewer infants with disorganised attachment at post-intervention (RR 0.32, 95% CI 0.17 to 0.58, 2 studies, 168 participants, low quality evidence); and an increase in the proportion of infants moving from insecure to secure attachment at post-intervention (RR 11.45, 95% CI 3.11 to 42.08, 2 studies, 168 participants, low quality evidence). There were no differences between PIP and control in any of the meta-analyses for the remaining primary outcomes (i.e. adverse effects), or secondary outcomes.
Four studies contributed data at post-intervention or follow-up to the PIP versus alternative treatment analyses producing 15 meta-analyses measuring parent mental health (depression); parent-infant interaction (maternal sensitivity); infant attachment category (secure, avoidant, resistant, disorganised) and attachment change (insecure to secure, stable secure, secure to insecure, stable insecure); infant behaviour and infant cognitive development. None of the remaining meta-analyses of PIP versus alternative treatment for primary outcomes (i.e. adverse effects), or secondary outcomes showed differences in outcome or any adverse changes.
We used the Grades of Recommendation, Assessment, Development and Evaluation Working Group (GRADE) approach to rate the overall quality of the evidence. For all comparisons, we rated the evidence as low or very low quality for parental depression and secure or disorganised infant attachment. Where we downgraded the evidence, it was because there was risk of bias in the study design or execution of the trial. The included studies also involved relatively few participants and wide CI values (imprecision), and, in some cases, we detected clinical and statistical heterogeneity (inconsistency). Lower quality evidence resulted in lower confidence in the estimate of effect for those outcomes.
AUTHORS' CONCLUSIONS
Although the findings of the current review suggest that PIP is a promising model in terms of improving infant attachment security in high-risk families, there were no significant differences compared with no treatment or treatment-as-usual for other parent-based or relationship-based outcomes, and no evidence that PIP is more effective than other methods of working with parents and infants. Further rigorous research is needed to establish the impact of PIP on potentially important mediating factors such as parental mental health, reflective functioning, and parent-infant interaction.
Plain language summary
PARENT-INFANT PSYCHOTHERAPY FOR IMPROVING PARENT AND INFANT WELL-BEING
Background
Parent-infant psychotherapy (PIP) is intended to address problems in the parent-infant relationship, and problems such as excessive crying and sleeping/eating difficulties. A parent-infant psychotherapist works directly with the parent and infant in the home or clinic, to identify unconscious patterns of relating and behaving, and influences from the past that are impeding the parent-infant relationship. Parents may be referred to this service (e.g. by a general practitioner in the UK) or may self refer to privately run services. The intervention is delivered to individual dyads but can also be delivered to small groups of parents and infants.
Review question
This review examined whether PIP is effective in improving the parent-infant relationship, or other aspects of parent or infant functioning, and to identify the programme components that appear to be associated with more effective outcomes and factors that modify intervention effectiveness (e.g. programme duration, programme focus).
Study characteristics
We searched electronic databases and identified randomised controlled trials (RCTs, where participants are randomly allocated to one of two or more treatment groups) and one cluster randomised trial (where prisons rather than participants were used as the unit of randomisation), in which participants had been allocated to a receive PIP versus a control group, and which reported results using at least one standard measure of outcome (i.e. an instrument which has been tested to ensure that it reliably measures the outcome under investigation).
Evidence is current to 13 January 2014.
We identified eight studies with 846 randomised participants comparing either PIP with a no-treatment control group (four studies) or comparing PIP with other types of treatment (four studies).
Key results
The studies comparing PIP with a no-treatment control group contributed data to 19 meta-analyses of the primary outcomes of parental mental health (depression), parent-infant interaction outcomes of maternal sensitivity (i.e. the extent to which the caregiver responds in a timely and attuned manner), child involvement and parent positive engagement, and infant outcomes of infant attachment category (the infant's ability to seek and maintain closeness to primary caregiver - infant attachment is classified as follows: ‘secure’ infant attachment is a positive outcome, which indicates that the infant is able to be comforted when distressed and is able to use the parent as a secure base from which to explore the environment. Infants who are insecurely attached are either ‘avoidant’ (i.e. appear not to need comforting when they are distressed and attempt to manage the distress themselves); or ‘resistant’ (i.e. unable to be comforted when distressed and alternate between resistance and anger). Children who are defined as ‘disorganised’ are unable to produce a coherent strategy in the face of distress and produce behaviour that is a mixture of approach and avoidance to the caregiver); and the secondary outcomes of infant behaviour and infant cognitive development (i.e. intellectual development, including thinking, problem solving and communicating).
In our analyses, parents who received PIP were more likely to have an infant who was securely emotionally attached to the parent after the intervention; this a favourable outcome but there is very low quality evidence to support it.
The studies comparing PIP with another model of treatment contributed data to 15 meta-analysis assessments of primary outcomes, including parental mental health, parent-infant interaction (maternal sensitivity); infant attachment and infant behaviour, or secondary infant outcomes such as infant cognitive development. None of these comparisons showed differences that favoured either PIP or the alternative intervention.
None of the comparisons of PIP with either a control or comparison treatment group showed adverse changes for any outcome.
We conclude that although PIP appears to be a promising method of improving infant attachment security, there is no evidence about its benefits in terms of other outcomes, and no evidence to show that it is more effective than other types of treatment for parents and infants. Further research is needed.
Quality of the evidence
The included studies were unclear about important quality criteria, had limitations in terms of their design or methods, or we judged that there was risk of bias in the trial. This lower quality evidence gives us less confidence in the observed effects.
1 Background
1.1 DESCRIPTION OF THE CONDITION
Infant regulatory disturbances, such as excessive crying, feeding or sleeping difficulties, and bonding/attachment problems represent the main reasons for referral to infant mental health clinics (Keren 2001). The Copenhagen Child Cohort Study (6090 infants) found a population prevalence of such regulatory problems (including emotional and behavioural, eating and sleeping disorders) in children aged 1.5 years in the region of 18% (Skovgaard 2008; Skovgaard 2010). Some regulatory disturbances are stable over time with as many as 49.9% of infants and toddlers (aged 12 to 40 months) showing a continuity of emotional and behavioural problems one year after initial presentation (Briggs-Gowan 2006). Problems of this nature are also significant predictors of longer-term difficulties. For example, infant regulatory problems have a strong association with delays in motor, language and cognitive development, and continuing parent-child relational problems (DeGangi 2000a; DeGangi 2000b). Difficult temperament, non-compliance and aggression in infancy and toddlerhood (aged one to three years) are associated with internalising and externalising psychiatric disorders at five years of age (Keenan 1998). Insecure and disorganised attachment in infancy is also associated with poorer outcomes in childhood across a range of domains such as emotional, social and behavioural adjustment, scholastic achievement and peer-rated social status (Berlin 2008; Granot 2001; Sroufe 2005a; Sroufe 2005b), particularly in the case of disorganised attachment, which is a significant predictor of significant later psychopathology (Green 2002).
Infant regulatory and attachment problems can best be understood in a relational context, and disturbances to the parent-child relationship and parental psychosocial adversity are significant risk factors for infant emotional, behavioural, eating and sleeping disorders (Skovgaard 2008; Skovgaard 2010). Early research in the field of infant mental health and developmental psychology has highlighted the significant role that the infant's primary carer plays in regulating the infant (Beebe 2010; Sroufe 1997; Tronick 1989; Tronick 1997), but one systematic review found only modest correlations between ‘maternal sensitivity’ and infant attachment security (De Wolff 1997), prompting a search for more specific predictive factors. Research has focused on the specific nature or quality of the attunement or contingency between parent and infant (Beebe 2010), and the parent's capacity for what has been termed ‘maternal mind-mindedness’ (Meins 2001) or ‘reflective function’ (Slade 2001).
Beebe developed the term ‘mid-range contingency’ to refer to interaction in which both the parent's self regulation and the interactive regulation between parent and infant is in the mid-range (Beebe 1988). Parents with low interactive tracking (e.g. resulting from withdrawal due to postnatal depression) are more likely to have infants who are insecurely attached, as are parents who have high interactive tracking (i.e. due to excessive vigilance resulting from anxiety) (Beebe 2010).
Parental reflective function refers to the parent's capacity to understand the infant's behaviour in terms of internal feeling states, and is strongly associated with maternal parenting behaviours, such as flexibility and responsiveness, while low maternal reflective function is associated with emotionally unresponsive maternal behaviours (withdrawal, hostility, intrusiveness) (Kelly 2005; Slade 2001; Slade 2005). Maternal reflective function is also associated with more optimal infant outcomes such as a greater use of the mother as a ‘secure base’ (i.e. the infant can be comforted by the primary carer when distressed, and able to explore the world in the presence of the carer when not distressed) (Kelly 2005). There is also a significant association between parental ‘mind-mindedness’ (the parent's capacity to interpret what their child is thinking and feeling accurately) and later development, including attachment security at 12 months (Meins 2001).
Research has also highlighted a number of ‘atypical’ parenting behaviours that can be present during the postnatal period, including affective communication errors (e.g. mother positive while infant distressed), disorientation (frightened expression or sudden complete loss of affect) and negative-intrusive behaviours (mocking or pulling infant's body) (Lyons-Ruth 2005). A meta-analysis of 12 studies found a strong association between disorganised attachment at 12 to 18 months of age and parenting behaviours characterised as ‘anomalous’ (i.e. frightening, threatening, looming), dissociative (haunted voice, deferential/timid) or disrupted (failure to repair, lack of response, insensitive/communication error) (Madigan 2006). These atypical parenting practices were identified in parents described as ‘unresolved’ with regard to previous trauma (Cicchetti 2006a; Cicchetti 2010; Jacobvitz 1997). However, disturbances to the mother-infant relationship are common and are associated with a range of maternal problems, including postnatal depression (Murray 2003; Timmer 2011; Toth 2006), personality disorder (Crandell 2003; Newman 2008), psychotic disorders (Chaffin 1996), substance misuse (Suchman 2005; Tronick 2005), and domestic violence (Lyons-Ruth 2003; Lyons-Ruth 2005).
1.2 DESCRIPTION OF THE INTERVENTION
Since the mid-1990s, a range of interventions (e.g. home visiting and parenting programmes) have been developed to address developmental problems in the infant, and problems in the parent-infant relationship, with a view to promoting optimal infant development. These have mostly targeted the parent and used a range of techniques in their delivery (discussion, role play, watching video vignettes, and homework), with the aim of changing parenting behaviours and attitudes. Parent-infant psychotherapy (PIP) in contrast is a dyadic intervention (or triadic if both parents are involved) that involves targeting the parent-infant relationship (i.e. it is delivered to parent and infant together). A parent-infant psychotherapist works by listening and observing the interaction, identifying the concerns and worries, and helping the parent observe and find different ways to relate to their baby. This work may take place in the home, clinic or hospital setting, and aims to address a wide range of problems that can arise during the antenatal and postnatal periods. The intervention is usually delivered to individual dyads but can also be delivered to small groups of dyads.
PIP focuses on improving the parent-infant relationship and infant attachment security by targeting parental internal working models (i.e. representational world - see below) and by working directly with the parent-infant relationship in the room. The approach is essentially psychodynamic in that it involves identifying unconscious patterns of relating, and the earliest approach, developed by Selma Fraiberg (Fraiberg 1980), focused primarily on the mother's ‘representational’ world (‘representation-focused’ approach) or the way in which the mother's current view of her infant was affected by interfering representations from her own history. The aim of such therapy was to help the mother to recognise the ‘ghosts in the nursery’ (i.e. the unremembered influences from her own past) and to link them to her current functioning, in order to improve the parent-infant relationship directly, thereby facilitating new paths for growth and development for both mother and infant (Cramer 1988). Fraiberg's model has been further developed and evaluated by others (e.g. Lieberman 1991; Toth 2006), and representational and behavioural approaches have been combined (Cohen 1999a). For example, ‘Watch, Wait and Wonder’ (WWW) is an ‘infant-led’ PIP that involves the mother spending time observing her infant's self initiated activity, accepting the infant's spontaneous and undirected behaviour, and being physically accessible to the infant (behavioural component). The mother then discusses her experiences of the infant-led play with the therapist with a view to examining the mother's internal working models of herself in relation to her infant (representational component) (Cohen 1999a). PIP may also work with the father or other primary carer, or with two parents together.
The duration of the intervention depends on the presenting problems but typically ranges from five to 20 weeks, usually involving weekly sessions. Parents may be referred to this service by a clinician (e.g. general practitioner (GP) or health visitor in the UK) or may self-refer to privately run services. PIP services typically target infants less than two years of age at the time of referral. This reflects the importance of the first two years of life in terms of children's later development (as described above).
1.3 HOW THE INTERVENTION MIGHT WORK
The logic model underpinning representational forms of PIP is that changes to the mother's representations (internal working models) will improve the mother's sensitivity and behaviour towards her infant (e.g. Lieberman 1991), thereby reducing distorted projections and making it possible for her to see the infant as someone with a ‘mind of their own’. Maternal sensitivity is strongly associated with more optimal parent-infant interaction, which, in turn, is associated with infant attachment security (De Wolff 1997). Secure attachment is associated with resilience and optimal social functioning (Sroufe 2005a; Sroufe 2005b), while both insecure (Berlin 2008; Granot 2001; Lecce 2008; Sroufe 2005a; Sroufe 2005b), and disorganised attachment are associated with a range of compromised outcomes (Green 2002; Lyons-Ruth 2005). The addition of behavioural components provides opportunities for parent and infant to interact, which then become the focus of exploratory discussions between therapist and parent, aimed once again at changing maternal representations about the infant (Cohen 1999a). The empathic relationship between the therapist and parent also plays a key role in helping parents to revise their internal working models (Toth 2006).
1.4 WHY IT IS IMPORTANT TO DO THIS REVIEW
Parent-infant interaction is a significant factor in infant mental health (Fonagy 2002), and problems with the parent-infant relationship are common (Keren 2001). Government policy internationally is increasingly emphasising the importance of early intervention and the need to develop empirically derived models that can support vulnerable parents and their children, and this reflects an increased recognition at policy level that both health and social inequalities have their origins in early parent-infant interaction (Field 2010), and that the social gradient in children's access to positive early experiences needs to be addressed (Marmot 2010).
There is a growing body of evidence pointing to the effectiveness of PIP in terms of improving both parental functioning (Cohen 1999a; Cohen 2002), and fostering secure attachment relationships in young children (Toth 2006), and there is some evidence to suggest that different forms of the therapy may be differentially effective for parents with different types of attachment insecurity (Bakermans-Kranenburg 1998).
However, to date, there has been only one ‘thematic’ summary of the evidence about the effectiveness of PIP (Kennedy 2007), which did not involve a systematic search for evidence. As such, there is a need for a systematic review to identify whether this unique method of working has benefits for parents (both mothers and fathers) and infants, and whether the outcome is affected by the duration or content of the intervention.
2 Objectives
To assess the effectiveness of PIP in improving parental and infant mental health and the parent-infant relationship. To identify the programme components that appear to be associated with more effective outcomes and factors that modify intervention effectiveness (e.g. programme duration, programme focus).
3 Methods
3.1 CRITERIA FOR CONSIDERING STUDIES FOR THIS REVIEW
3.1.1 Types of studies
We pre-specified our methods for this review in the protocol (Barlow 2013).
We included randomised controlled trials (RCTs) and quasi-randomised controlled trials (quasi-RCTs) in which participants had been randomly allocated to an experimental or a control group, the latter being a waiting-list, no treatment, treatment-as-usual (normal service provision) or a placebo control group. We defined quasi-RCTs as trials where allocation was done on the basis of a pseudorandom sequence, for example, odd or even hospital number, date of birth or alternation (Higgins 2011). We also included studies comparing two different therapeutic modalities (i.e. without a control group).
3.1.2 Types of participants
We included studies involving parent-infant dyads in which the parent was experiencing mental health problems, domestic abuse or substance dependency, with or without the infant showing signs of attachment or dysregulation problems, or both attachment and dysregulation problems. We included all infants irrespective of the presence of problems such as low birthweight, prematurity or disabilities. We included studies targeting infants and toddlers in which the mean age of the infant participants was 24 months or less at the point of referral. We included studies targeting all parents (i.e. including fathers, birth parents, adoptive and kinship parents, but not foster parents).
3.1.3 Types of interventions
We included studies that had evaluated the effectiveness of PIP programmes in which the intervention met all of the following criteria: underpinned by a psychodynamic model that involved making unconscious patterns of relating by targeting the parent-therapist transference; parental internal working models or representations (i.e. the way in which the mother's current view of her infant was affected by interfering representations or unremembered influences from her own history, and to link them to her current functioning, in order to improve the parent-infant relationship directly, thereby facilitating new paths for growth and development for both mother and infant); delivered jointly to both parent and infant with a focus on the parent-infant relationship/interaction and aimed primarily at improving infant attachment security, socio-emotional functioning, or both, via the parent-infant relationship/interaction; delivered by a parent-infant psychotherapist/specialist on a dyadic basis or to dyads in groups, in any setting (clinic, hospital or home), over any period of time.
We also included studies of PIPs that used additional components (i.e. provided they still met the core inclusion criteria). We excluded studies of interventions that were delivered to the parent or parents alone (e.g. interpersonal psychotherapy (IPT) or that were dyadic but primarily psychoeducational or based on other therapeutic models (e.g. behavioural or cognitive behavioural therapy (CBT)). We excluded studies of stand-alone video-interaction guidance interventions but not studies in which video feedback had been incorporated into a PIP that met the above criteria.
3.1.4 Types of outcome measures
We extracted data for the following outcomes at both post-intervention and follow-up, provided they had been measured using a standardised parent-report or independent observation of the type listed as examples for each outcome below.
Primary outcomes
Parent outcomes
Parental mental health including, for example, depression* (e.g. Beck Depression Inventory (BDI) (Beck 1961); anxiety (e.g. Beck Anxiety Inventory (BAI) (Beck 1988); parenting stress (e.g. Parenting Stress Index (PSI) (Abidin 1983).
Parent-infant relationship outcomes
Parent-infant interaction including, for example, Child-Adult Relationship Experimental-Index (CARE-Index) (Crittenden 2001), Emotional Availability Scales (EAS) (Biringen 1993), Parent-Child Early Relational Assessment (PC-ERA) (Clark 1985), Dyadic Parent-Child Interaction Coding System (DPICS) (Robinson 1981), Nursing Child Assessment of Feeding Scale (NCAFS) (Barnard 1978a), or the Nursing Child Assessment Teaching Scale (NCATS) (Barnard 1978b); Maternal Sensitivity Scale (MSS) (Ainsworth 1974), Atypical Maternal Behavior Instrument for Assessment and Classification (AMBIANCE) (Bronfman 1999), or Frightened/Frightening (FR) Coding System (Main 1992).
Infant outcomes
Infant emotional well-being including, for example, infant attachment* measures such as the Strange Situation Procedure (SSP) (Ainsworth 1971), Preschool Measure of Attachment (PMA) (Crittenden 1992), or other measures of emotional and behavioural adjustment such as the Infant and Toddler Social and Emotional Assessment (ITSEA) (Carter 2000), Eyberg Child Behaviour Inventory (ECBI) (Eyberg 1978), Behaviour Screening Questionnaire (BSQ) (Richman 1971), and the Child Behaviour Questionnaire (CBQ) (Rutter 1970).
Adverse effects
Adverse effects of interventions were included as an outcome, including a worsening of outcome on any of the included measures.
Secondary outcomes
Parent outcomes
Parental reflective function including, for example, Parent Development Interview (PDI) (Slade 2004).
Infant outcomes
Infant stress including, for example, salivary or urinary cortisol measured in standardised units such as micrograms per decilitre (μg/dL) or nanograms per millilitre (ng/mL). Infant development including, for example, social, emotional, cognitive and motor development using the Bayley Scales (Bayley 1969).
3.2 SEARCH METHODS FOR IDENTIFICATION OF STUDIES
3.2.1 Electronic searches
We searched the following databases on 13 January 2014 without restrictions on language, date or publication status. We applied RCT filters where appropriate (see Appendix 1). Cochrane Central Register of Controlled Trials (CENTRAL) 2014 (Issue 1), part of The Cochrane Library.
Ovid MEDLINE, 1950 to 10 January 2014. EMBASE (Ovid), 1980 to January week 1 2014. CINAHL (EBSCO), 1982 to current. PsycINFO (Ovid), 1806 to January week 1 2014. BIOSIS Citation Index (ISI), all available years. SSCI (Web of Science), 1970 to current. ERIC (ProQuest), 1966 to current. Sociological Abstracts (ProQuest), 1952 to current.
We also searched the metaRegister of Controlled Trials (mRCT) on 20 January 2014 (Appendix 2).
3.2.2 Searching other resources
We contacted authors and experts in the field to identify unpublished studies. We handsearched reference lists of articles identified through database searches for further relevant studies. We also examined bibliographies of systematic and non-systematic review articles to identify relevant studies.
3.3 DATA COLLECTION AND ANALYSIS
3.3.1 Selection of studies
Two review authors (CB and JB) screened the titles and abstracts of studies identified by the searches to assess whether they met the inclusion criteria and independently assessed full copies of papers that appeared to meet the inclusion criteria. We resolved any uncertainties by discussion with a third review author (NM).
3.3.2 Data extraction and management
Two review authors (CB and SL) independently extracted data using an identical data extraction form and entered the data into Review Manager 5 software (RevMan 2012). Where data were not available in the published trial reports, we contacted study investigators to supply missing information. We extracted information regarding: study design, measurement of delivery fidelity, participant characteristics, primary and secondary outcome measures, and data.
3.3.3 Assessment of risk of bias in included studies
Two review authors (CB, SL) carried out ‘Risk of bias’ assessments using The Cochrane Collaboration ‘Risk of bias’ assessment tool (Higgins 2011). We resolved differences by consultation with a third review author (JB or NM). We assessed each trial in the following areas: sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, incomplete outcome data, selective reporting and other sources of bias, including whether there was any assessment of the distribution of confounders. Where there was insufficient information in the trial report to make a judgement, we contacted trial investigators for further information.
We assessed all domains as being at low, high or unclear (uncertain) risk of bias.
We used ‘Risk of bias’ assessments in the synthesis of data, to interpret findings for specific outcomes and to comment on the quality of the evidence.
Sequence generation
We assessed the method used to generate the allocation sequence to identify whether it had produced comparable groups.
Allocation concealment
We assessed the method used to conceal allocation sequence to assess its adequacy in terms of whether the intervention schedules could have been foreseen in advance of, or during, recruitment.
Blinding
We assessed whether any steps were taken to blind participants and personnel, and to blind outcome assessors to the intervention that participants received.
Incomplete outcome data
We assessed whether incomplete data was dealt with adequately, and how data on attrition and exclusions were reported in comparison with the total number of participants randomised. Where studies did not report intention-to-treat (ITT) analyses, we attempted to obtain missing data by contacting the study authors.
Selective outcome reporting
We assessed whether any attempt had been made to reduce the possibility of selective outcome reporting.
Other sources of bias
We examined baseline or pre-treatment means, where available, to assess whether there was any imbalance in terms of participant characteristics that were strongly related to outcome measures and that could cause bias in intervention effect estimates (Higgins 2011, Chapter 8.14.1.2).
We assessed whether the study was free of other problems that could have put it at a high risk of bias (e.g. contamination).
3.3.4 Measures of treatment effect
Dichotomous outcome data
For dichotomous endpoint measures, we present the number of parents or infants who showed an improvement as a proportion of the total number of parents/infants treated. We calculated risk ratios (RR) by dividing the risk in one group with the risk in the other group, and present these with 95% confidence intervals (CI) and standard deviations.
Not all dichotomous measures indexed relative risks of improvement over time and, for some measures, we provided the relative risk of a positive state (e.g. secure attachment) at post-intervention.
Continuous outcome data
For the meta-analyses of continuous outcomes, we estimated the mean differences (MDs) between groups. In the case of continuous outcome measures, where data were reported on different scales, we analysed data using the standardised mean difference (SMD), calculated by dividing the MD in post-intervention scores between the intervention and control groups by the standard deviation. We presented the SMDs and 95% CIs for all meta-analyses and individual outcomes from individual studies (i.e. where no meta-analysis was undertaken).
Where the above data were not available, we present significance levels reported in the paper.
3.3.5 Unit of analysis issues
Cluster-randomised trials
The randomisation of clusters can result in an overestimate of the precision of the results (with a higher risk of a Type I error) where their use has not been compensated for in the analysis. One study employed cluster-randomisation (Sleed 2013a, mother and baby units in prisons). We explored the impact of the inclusion of data from this study in the meta-analyses by imputing a set of intraclass correlation coefficients (ICC) (0.2, 0.4, 0.6, 0.8) (i.e. we were unable to identify external evidence for a reasonable ICC and for this reason, we selected multiple values for ICCs). We calculated the inflated standard errors that accounted for clustering by multiplying the original standard errors with the square root of the associated design effects (see Higgins 2011, Chapter 16.3.6).
Studies with multiple treatment groups
For studies where there was more than one active intervention and only one control group, we selected the intervention that most closely matched our inclusion criteria and excluded the others (see Higgins 2011, Chapter 16.5.4).
3.3.6 Dealing with missing data
We assessed missing data and attrition for each included study and reported them in the ‘Risk of bias’ tables. Where appropriate, we contacted authors to supply data missing from included studies. Where missing data could not be provided, we have reported this and the available data only (i.e. we used no imputation).
3.3.7 Assessment of heterogeneity
We assessed clinical heterogeneity by examining the extent of between-trial differences, including methods, populations, interventions, whether the delivery of the intervention was monitored (to ascertain fidelity), and outcomes.
We assessed statistical heterogeneity using the I2 statistic. The importance of the observed I2 value is dependent on the magnitude and direction of effects and strength of evidence for heterogeneity (e.g. P value from the Chi2 test, or a CI for I2 statistic) (Higgins 2002), and we have interpreted an I2 greater than 50% as evidence of substantial heterogeneity.
Where we performed a Chi2 test of heterogeneity, we interpreted a significance level less than 0.10 as evidence of heterogeneity. We carried out the tau2 estimates for each meta-analysis and presented Chi2 values, P values, tau2 statistic and I2 statistic.
3.3.8 Assessment of reporting biases
Due to the small number of included studies, we were unable to assess reporting biases. For more information on how we will assess reporting bias in future updates of this review, see Barlow 2013 and Appendix 3.
3.3.9 Data synthesis
We undertook meta-analysis where there was sufficient clinical homogeneity in the intervention delivered, the characteristics of the study participants (e.g. age or the definition of ‘at risk’ participants), and the outcome measures. We made the decision about combining data post hoc based on the categories of interventions, participants, and outcomes identified in the reviewed literature. We combined data using a random-effects model. We calculated overall effects using inverse variance methods.
All analyses included all participants in the treatment groups to which they were allocated, whenever possible.
While we attempted to combine data where at all possible, there were some circumstances where it was not possible; for example, where some studies reported the same outcome using different formats (e.g. dichotomous and continuous), we have carried out two separate analyses. For single outcomes, we presented the individual effect sizes and 95% CIs.
3.3.10 Subgroup analysis and investigation of heterogeneity
We explored possible reasons for heterogeneity by scrutinising the studies to assess the extent of between-trial differences (e.g. age of infant, presenting problems, programme duration, programme focus, and whether or not the intervention was delivered as intended).
We had planned to carry out additional subgroup analyses (Barlow 2013), but there were too few included studies in each meta-analysis to do this (see Appendix 3 for details).
3.3.11 Sensitivity analysis
We conducted a sensitivity analysis using a fixed-effect model and a random-effects model. We intended to re-analyse the data excluding studies on the basis of design (Barlow 2013), but there were too few included studies in each meta-analysis to do this (see Appendix 3 for details).
‘Summary of findings’ tables
We presented the findings of our review in ‘Summary of findings’ tables (Summary of findings table 1; Summary of findings table 2), which provide a transparent and simple tabular format of the review's primary outcomes that may be important to parents and decision makers. For the comparison of PIP versus control, we presented findings for the outcomes of parent mental health (depression) at postintervention, and infant attachment at post-intervention and follow-up. For PIP versus alternative intervention, we presented findings for the outcomes of parent mental health (depression) at post-intervention and follow-up, and infant attachment at post-intervention and follow-up. We used GRADEpro software to construct the tables (GRADEpro 2014). The tables present information about the body of evidence (e.g. number of studies), the judgements about the underlying quality of evidence, key statistical results, and a grade for the quality of evidence for each outcome. We used the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) system to describe the quality of the evidence and the strength of recommendation (GRADE 2013; Guyatt 2011). We expressed the quality of evidence on a four-point adjectival scale from ‘high’ to ‘very low’. We gave evidence from RCT data initially a high quality rating but downgraded it if there was unexplained clinically important heterogeneity, the study methodology had a risk of bias, the evidence was indirect, there was important uncertainty around the estimate of effect, or there was evidence for publication bias. Therefore, it was possible for RCT data to have a very low quality of evidence if several of these concerns were present.
4 Results
4.1 DESCRIPTION OF STUDIES
4.1.1 Results of the search
Electronic searches in February 2013 and updated in January 2014 identified 2604 records (for search results, see Appendix 4). The January 2014 search included a correction to the EMBASE search strategy. We identified 16 additional records through other sources. After we removed duplicates, we screened 1937 records. We obtained and scrutinised 85 potentially relevant records, and 61 of these reports (58 studies) did not meet the inclusion criteria and we excluded them (see Characteristics of excluded studies). We included eight studies (from 19 reports of trials) (see Characteristics of excluded studies), and identified five ongoing studies (see Characteristics of ongoing studies) (Figure 1).
Study flow diagram
Our included studies were: Cicchetti 1999a; Cicchetti 2006b; Cohen 1999b; Cooper 2003a; Lieberman 1991; Robert-Tissot 1996a; Salomonsson 2011a; and Sleed 2013a. These were grouped as follows: Cicchetti 1999a comprised three reports (Cicchetti 1999b; Cicchetti 2000; Toth 2006); Cicchetti 2006b comprised three reports (Cicchetti 2006a; Cicchetti 2011a; Stronach 2013); Cohen 1999b comprised three reports (Cohen 1999a; Cohen 2000; Cohen 2002); Cooper 2003a comprised two reports (Cooper 2003b; Murray 2003); Lieberman 1991 was a single report; Robert-Tissot 1996a comprised two reports (Cramer 2002; Robert-Tissot 1996b); Salomonsson 2011a comprised three reports (Salomonsson 2011b; Salomonsson 2011c; Salomonsson 2011d); and Sleed 2013a was a single published report; further details can be found on the Anna Freud Centre's website (Anna Freud 2014).
In both studies conducted by Cicchetti et al. (Cicchetti 1999a and Cicchetti 2006b), each subsequent publication reported on a subset of the main study for which data were available, therefore some details, such as participant demographics, were reported differently in each publication. We have summarised these differences in the Characteristics of included studies table.
4.1.2 Included studies
Three studies were conducted in the USA (Cicchetti 1999a; Cicchetti 2006b; Lieberman 1991), two in the UK (Cooper 2003a; Sleed 2013a); and one each in Canada (Cohen 1999b), Switzerland (Robert-Tissot 1996a), and Sweden (Salomonsson 2011a).
Methods
All of the included studies were RCTs. Some of the studies also recruited or compared results from a non-randomised control group (Cicchetti 1999a; Cicchetti 2006b; Robert-Tissot 1996a); we have not included the results from these non-randomised arms in this review.
Design
Three studies employed a two-arm RCT design in which participants were randomised either to PIP or a control (treatment-as-usual) condition (Cicchetti 1999a; Lieberman 1991; Salomonsson 2011a), one other employed a cluster-randomised trial design in which mother and baby units in UK prisons were allocated to either the treatment (PIP) or a control standard care condition (Sleed 2013a). We included these studies in analyses of PIP versus control (control being no intervention or treatment-as-usual).
Two studies compared PIP with another treatment: one compared PIP that followed Fraiberg's model (Fraiberg 1987), and hereafter referred to as psychodynamic psychotherapy (PPT), with an ‘infant-led’ PIP called WWW (Cohen 1999b); and the second compared PIP with interaction guidance (Robert-Tissot 1996a). We included these studies in analyses of PIP versus an alternative intervention.
One study employed a three-arm parallel group design in which PIP (also called child-parent psychotherapy (CPP)) was compared with a psychoeducational parenting intervention (PPI), derived from the preventive home visiting work of David Olds and colleagues (Olds 1998), and a control community standard care condition (Cicchetti 2006b). This three-arm study contributed data to both analyses of PIP versus control and PIP versus an alternative intervention. To avoid double counting, we split the number of participants in the common group (i.e. the PIP group).
One study employed a randomised four-arm comparison of routine primary care, non-directive counselling, CBT, and psychodynamic (PIP) therapy (Cooper 2003a). For the purposes of this review, we aggregated data from the counselling and CBT arms and compared this with the PIP therapy arm, in analyses of PIP versus alternative intervention.
Participants
The eight included studies initially randomised 846 participants. The number of participants randomised in each trial ranged from 59 (Lieberman 1991) to 193 (Cooper 2003a) parent-infant dyads. The intervention was directed at mothers in all of the included studies (fathers were not excluded in Robert-Tissot 1996a, but no separate data for the intervention groups was included in the published report or available from the investigator).
In all eight included studies, the mean age of the infant participants was under 24 months at study enrolment. In one study, the infants were eight weeks old at study entry (Cooper 2003a). We included two studies which stated that the age of some the participating infants was 30 months at study entry but the mean age still met our inclusion criteria (in Cohen 1999b, the mean age in the two intervention groups was 21.5 months (SD = 6.5) in the WWW arm and 19.2 months (SD = 6.1) in the PIP (‘PPT‘) arm, range 10-30 months; in Robert-Tissot 1996a the mean age at pretreatment assessment was 15.6 months and ranged from 2 to 30 months (SD = 8.4). The report of Lieberman 1991 stated that the intervention was delivered to infants aged 12 months with infants being 11 to 14 months old at study entry. Cicchetti 1999a stated that the mean age of all the infants in the study (i.e. both the intervention and the control groups) was 20.4 months (SD = 2.38); in Cicchetti 2006b the mean age was 13.31 months (SD = 0.81). In Salomonsson 2011a the mean age of the infants was 4.4 months (SD = 2.4) in the PIP group and 5.9 months (SD = 3.8) in the control group; the maximum age of the infants at study entry was less than < 18 months. In Sleed 2013a, infants were aged 1-23 months in the intervention group with a mean age of 4.9 months (SD = 4.5; range = 0.2-23.0 months); in the control group, the mean age was 4.4 months (SD = 4.6; range = 0.1-18.5 months). Cicchetti 1999a, Cicchetti 2006b, and Cooper 2003a did not state the maximum age of the infants at enrolment.
The eight included studies investigated the effectiveness of PIP with a range of clinical groups, including depressed mothers (Cicchetti 1999a; Cooper 2003a); maltreating parents (Cicchetti 2006b); incarcerated mother-infant dyads (Sleed 2013b); parents experiencing a range of infant problems (Cohen 1999b; Robert-Tissot 1996a); mothers with concerns about their own role as mothers, well-being of the infants, or the mother-infant relationship (Salomonsson 2011a); and parents of infants at risk of anxious attachment because their mothers are emotionally unavailable (Lieberman 1991).
Two studies reported the population groups to be either Caucasians (94.5% Caucasian in Cicchetti 1999a) or European Caucasian infants (Robert-Tissot 1996a); or in a third study, American Hispanic (recently immigrated from Mexico or Central America) (Lieberman 1999). Three studies reported the proportion of participants from a minority race or ethnicity: in Cicchetti 2006b, overall, 74.1% of the randomised participants were described as minority race or ethnicity (Cicchetti 2006a) and in Sleed 2013a over one-half were white and the remaining “Asian, Mixed or other”; in Salomonsson 2011a, 11% in the PIP group and 22% in the control group were described as ‘immigrant’ (Salomonsson 2011a). Two studies did not report participant ethnicity (Cohen 1999b; Cooper 2003a).
Recruitment
Cicchetti 1999a recruited from a community sample of mothers with a history of depressive disorder via referrals from mental health professionals and through notices placed in the community. Cicchetti 2006b recruited maltreating mothers (i.e. child protection services identified infants known to have been maltreated or who were living in maltreating families with the biological mothers). Cooper 2003a recruited from hospital birth records and after screening for mood disturbances. Cohen 1999b and Robert-Tissot 1996a used referrals made by the parents or by medical mental health and child welfare professionals for feeding, sleeping, and behaviour problems. Lieberman 1991 recruited dyads from paediatric clinics at large teaching hospitals and neighbourhood health clinics, targeting recent Latino immigrants at risk of anxious attachment due to parental unavailability. Sleed 2013a liaised with the mother and baby unit staff of the women's prisons to identify mother-infant dyads who would be willing to take part. Salomonsson 2011a recruited mothers through information provided at the delivery ward of the hospital and at parenting internet sites.
Interventions
All included studies involved the delivery of PIP, which had its origins in the work of Selma Fraiberg (Fraiberg 1975; Fraiberg 1980), and was based on a psychodynamic model. Only three of the included studies used manualised programmes (Cicchetti 1999a; Cicchetti 2006b; Sleed 2013a).
Parent-infant psychotherapy
Lieberman 1991 delivered PIP in unstructured weekly sessions. There was no didactic teaching and the therapists sought to alleviate the mothers' psychological conflicts about their children through observations.
Cicchetti 1999a and Cicchetti 2006b employed manualised infant-parent psychotherapy. Mothers and their infants attended conjoint therapy sessions. Insights gained by the therapist were used to increase maternal sensitivity, responsivity, and attunement to the child.
In Cohen 1999b, in the PIP (‘PPT’) arm, the mother and infant played while the therapist's observations were used to draw the mother's attention to her infant's needs and signals. In the infant-led psychotherapy (WWW), mothers and infants played on the floor, and mothers were instructed to interact only at the infant's initiative. The therapist engaged in a parallel process of ‘watching, waiting, wondering’ about the interactions between mother and infant and did not intervene.
Salomonsson 2011a also employed a form of infant-led psychotherapy, in which the analyst received and emotionally processed the infant's distress and communicated it back to the infant in a form that the infant could assimilate, in the presence of the mother.
Robert-Tissot 1996a employed a method of PIP that was explicitly based on the work of Fraiberg. The therapist listened to the mother's concerns, anxieties and narratives, and examined the relationship between the therapist, mother and infant in terms of the core conflictual relationship between the mother and infant. An emphasis was placed on fostering a positive therapeutic alliance.
Cooper 2003a employed brief psychodynamic parent-infant therapy, which focused on the therapist exploring the mother's representations of the experience of motherhood in terms of her own experiences of being parented.
Only one study employed a group-based therapy (Sleed 2013a). “New Beginnings” was conducted over eight sessions that were delivered over a four-week period. The groups were comprised of up to six mother-baby dyads and two parent-infant psychotherapists as facilitators. The topics of each session were selected and examined in terms of their potential as triggers of the attachment relationship.
Further details about the specifics of the psychodynamic therapy, and intervention of all the included studies, can be found in the Characteristics of included studies table.
Duration of intervention
The duration of the therapies ranged from eight sessions (Sleed 2013a) to between 46 and 49 weeks (Cicchetti 2006b).
Site of delivery of intervention
Robert-Tissot 1996a delivered the intervention and assessed outcomes in a research clinic. Cooper 2003a delivered the intervention in the participants' homes, and Cicchetti 2006b and Lieberman 1991 delivered the intervention in the client's home and the assessment of outcome in university research facilities. Cohen 1999b delivered the intervention in a playroom at a children's mental health centre. Sleed 2013a delivered the intervention in the mother and baby units of several women's prisons in the UK. It is unclear where Salomonsson 2011a and Cicchetti 1999a delivered the intervention, although outcome assessments were made in the research clinic setting.
Monitoring treatment fidelity
The delivery of the intervention was monitored in all included studies and therapists were highly trained and supervised during the intervention.
Comparisons
Four studies compared PIP with a treatment-as-usual control condition (Cicchetti 1999a; Lieberman 1991; Salomonsson 2011a; Sleed 2013a). The control condition in Cicchetti 1999a was not described but assumed to be treatment-as-usual in which all participants were able to access other mental health treatments; and in Lieberman 1991 was not described but assumed to be normal service provision. In Sleed 2013a, mothers and babies in both groups had access to standard health and social care provision as provided by the prison service. In Salomonsson 2011a, the control condition involved access to the local child health centre responsible for check-ups from birth to six years of age, antidepressants, and brief psychotherapies.
Two studies compared PIP with alternative interventions, including a behavioural and infant-led PIP (i.e. WWW) (Cohen 1999b); and Interaction Guidance (Robert-Tissot 1996a).
One study employed a three-arm trial design in which PIP was compared with a home visitation intervention delivered over 12 months (i.e. a preventive intervention, involving home visitations scheduled weekly over a 12-month period) and a treatment-as-usual control (Cicchetti 2006b).
Cooper 2003a employed a four-arm study in which PIP was compared with a control condition (i.e. routine primary care provided by the primary healthcare team), and two alternative interventions (CBT and non-directive counselling).
Outcomes
Timing of outcome assessment
In Cicchetti 1999a, post-intervention outcomes were assessed when the child was 36 months of age. In Cicchetti 2006b, all outcomes were assessed at post-intervention (i.e. when the children were aged 26 months) (Stronach 2013), but cortisol levels were assessed at 12 months' follow-up (when the children were approximately 38 months of age) (Cicchetti 2011a). Cohen 1999b reported outcomes at postintervention and six-month follow-up (Cohen 2002). Lieberman 1991 assessed outcomes at post-intervention (24 months of age). Robert-Tissot 1996a assessed outcomes immediately post-intervention and at six-month follow-up (when the children were approximately 26 months of age). Long-term follow-up interviews at pre-adolescence were also reported (Cramer 2002), but the results were not presented separately for the intervention groups and we were unable to obtain further information. Sleed 2013a assessed outcomes immediately post-intervention (range one to 25 months of age) and two months after the end of treatment when the infants were three to 27 months of age. Cooper 2003a presented results immediately post-intervention and at nine-month, 18-month and five-year follow-up. Salomonsson 2011a reported post-intervention outcomes only.
Outcomes
The included studies reported a range of outcomes relating to parental mental health, parent-infant interactions and infant development (i.e. attachment, behaviour, cognitive, and mental development).
Primary outcomes
Parent outcome
Five studies measured maternal depression using a number of self report measures (e.g. BDI; Beck 1978) (Cicchetti 1999a; Cohen 1999b), Edinburgh Postnatal Depression Scale (EPDS) (Salomonsson 2011a), and Center for Epidemiologic Studies Depression Scale (CES-D) (Sleed 2013a). In two studies, the number of number of depressive episodes was reported using criteria from the Diagnostic and Statistical Manual of Mental Disorders - Third Edition - Revised (DSM-III-R) (Cohen 1999b; Cooper 2003a).
Cohen 1999b measured parent stress using the PSI (Abidin 1986), and parenting sense of competence using the Parenting Sense of Competence (PSOC) scale (Johnston 1989). Salomonsson 2011a reported scores on the Swedish equivalent of the PSI (the Swedish Parenting Stress Questionnaire; SPSQ) and a measure of general psychological stress, the General Severity Index (GSI) of the Symptom Checklist - 90 (SCL-90) (Derogatis 1994).
Parent-infant relationship outcomes
Six studies measured mother-infant interaction during play using video-taped sessions, which were then analysed and coded. Cohen 1999b measured dyadic reciprocity, dyadic conflict, maternal intrusiveness, and maternal unresponsiveness using the Chatoor Play Scale (Chatoor 1986). Lieberman 1991 measured maternal initiation of interaction, behaviour on reunion (goal-corrected partnership) and maternal child-rearing attitudes (control aggression, encourage reciprocity and awareness of complexity in child rearing). Robert-Tissot 1996a measured controlling or unresponsive behaviour using the CARE-Index (Crittenden 1981). Sleed 2013a used the Coding Interactive Behavior (CIB) for parent positive engagement and child involvement. Cooper 2003a assessed maternal warmth, responsiveness, and acceptance using the Global Rating Scales (GRS; Murray 1996). Salomonsson 2011a assessed sensitivity, structuring, non-intrusiveness and non-hostility; and two infant dimensions: responsiveness and involvement using the EAS.
Robert-Tissot 1996a measured maternal sensitivity using the CARE-Index (Crittenden 1981); Lieberman 1991 measured maternal empathic responsiveness (sensitivity) coded from video-taped free-play measures; and Sleed 2013a used the CIB for dyadic attunement/maternal sensitivity. Cooper 2003a assessed maternal sensitivity using the GRS (Murray 1996). Salomonsson 2011a assessed sensitivity using the EAS.
Infant outcomes
Five studies measured infant attachment (Cicchetti 1999a; Cicchetti 2006b; Cohen 1999b; Cooper 2003a; Lieberman 1991). Cicchetti 1999a used the Attachment Q-Set (AQS) (Waters 1995). Three studies assessed infant attachment using the SSP (Cicchetti 2006b; Cohen 1999b; Cooper 2003a). The SSP categorises infant attachment styles into one of four classifications: secure, insecure avoidant, insecure resistant or disorganised attachment (Ainsworth 1971). Secure attachment is a favourable outcome. Lieberman 1991 measured security of attachment using the 90-item Attachment Q-sort, which is a refinement of the original Q-set (AQS) (Waters 1985), and is an observation-based report, obtained by home visitors at follow-up only. Lieberman 1991 also assessed dyadic behaviour on reunion, proximity avoidance and resistance, behaviour on reunion and goal-corrected partnership, using the Ainsworth Interactive Behaviour Scales (modelled on the SSP but with 10-minute episodes suitable for younger children; Ainsworth 1978).
One study assessed the mean number of angry or externalising behaviours and restriction of affect using a non-standardised rating of video-taped interactions (Lieberman 1999). One study used the Child Behaviour Checklist (CBCL; Achenbach 1992) (Cicchetti 2006b). One study measured infant behaviour using the Behavioural Screening Questionnaire (BSQ), and the Rutter Parent A2 Scale (maternal report) and Parent Behavior Checklist (PBCL - teacher report) (Cooper 2003a).
Adverse effects
Three studies (Cicchetti 1999a; Cicchetti 2006b; Cohen 1999b) measured adverse effects in terms of the potential adverse impact of the intervention on attachment security (i.e. measured change from secure to insecure attachment).
Secondary outcomes
Parent outcomes
One study measured maternal representations using the Mother's Object Relations Scale (MORS) (Danis 2005; a self-report measure) and the PDI (Slade 2004; a structured interview, objective report) (Sleed 2013a); and one study measured maternal attitudes using the Egeland 1979 abbreviated version of the Maternal Attitude Scale (MAS; Cohler 1970) (Lieberman 1991).
Infant outcomes
In one study, infant symptoms were obtained by asking the mother to list the primary problems present at the time of referral for help and to rate these on a 100-point scale for severity, degree of difficulty, and perceptions of effectiveness and comfort in dealing with problems (Cohen 1999b).
Infant functioning and cognitive development were measured in Cohen 1999b using the Bayley Scales of Infant Development (BSID) I or II (Bayley 1969; Bayley 1993), to derive a Developmental Quotient (DQ); and the BSID Mental Development Index (MDI) (Bayley 1969) in Cooper 2003a. The infant's cognitive development (infant IQ) was assessed in Cicchetti 1999a using the Wechsler Preschool and Primary Scales of Intelligence (WPPSI-R; Wechsler 1989). Salomonsson 2011a measured infant functioning using the Ages and Stages Questionnaire (ASQ).
Cicchetti 2006b measured infant stress using samples of morning salivary cortisol (expressed as micrograms per decilitre; μg/dL), but the results were aggregated for both intervention groups and we were unable to obtain disaggregated data.
4.1.3 Excluded studies
We formally excluded 58 studies, details of which can be found in the Characteristics of excluded studies table. Of these, eight were RCTs but did not fit our inclusion criteria. Twenty-one were not RCTs but otherwise met at least one of our inclusion criteria. Twenty-five studies did not assess the effectiveness of PIP. In three RCTs of PIP (Lieberman 2005; Smyrnios 1993; Toth 2002), the age of the children was outside the maximum age specified in the inclusion criteria for this review.
4.2 RISK OF BIAS IN INCLUDED STUDIES
We presented the ‘Risk of bias’ tables for each included study beneath the Characteristics of included studies table. Figure 2 shows a summary of risk of bias across all studies; Figure 3 shows the results as percentages. For all studies rated as unclear risk of bias, when contacted, the study investigators were unable to provide further details.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies
4.2.1 Allocation (selection bias)
We judged three of the included studies as being at low risk of bias because adequate methods were used to generate the randomisation sequence (Cooper 2003a; Salomonsson 2011a; Sleed 2013a).
The method used to generate the randomisation sequence in Robert-Tissot 1996a was unclear. Cicchetti 1999a and Lieberman 1991 used block randomisation procedures; they assigned infants to the intervention of control group by blocks stratified for infant sex and birth order (first, later), but the method of randomisation was unclear.
In Cicchetti 2006b, we judged the risk of bias due to randomisation methods as high because some of the control participants who were originally assigned to PPI or infant-parent psychotherapy (IPP; also called CPP) declined, and the study investigators presented results for this group of decliners in the same group as the community standard care randomised group. We judged Cohen 1999b to be at high risk of bias for random sequence generation because randomisation was inadequate in one-third of the cases in which assignment was dependent on therapist caseload and availability for time of treatment; in two-thirds of the cases, assignment was done using a random numbers table.
Three studies concealed allocation sequence (Lieberman 1991; Salomonsson 2011a; Sleed 2013a). In one study, we could not obtain details about allocation concealment (Robert-Tissot 1996a), and we rated this as unclear. We judged four studies to be at high risk of bias in terms of allocation concealment: Cicchetti 1999a and Cicchetti 2006b did not use a preset allocation sequence; Cohen 1999b did not provide details about allocation concealment, and allocation of one-third of the participants was based on therapist availability (therefore, it could not have been concealed); and the allocation sequence was not concealed in Cooper 2003a.
4.2.2 Blinding (performance bias and detection bias)
Performance bias
Due to the nature of the interventions, it was not possible to blind both the participants and personnel to group allocation.
Detection bias
We rated blinding of outcome assessment as low in all of the included studies. Although it is not possible to blind participants and personnel to the allocation given the nature of the treatment, the investigators of the included studies made efforts to ensure that, where possible, those who were involved in the measurement of outcomes were blind to the initial allocation. For four studies, the blinding of outcome assessors was complete. In Robert-Tissot 1996a, the researchers who coded mother-infant interactions and analysed and coded the responses to the different questionnaires were blind to the treatment condition. In Cooper 2003a, the outcome assessors were blinded to the allocation. In Sleed 2013a, coders were blind for PDI, and CIB coding was carried out by trained coders who were blind to both the group and time point. In Cicchetti 1999a, outcome assessment was conducted with toddlers, with their mothers present, by an experimenter unaware of experimental hypotheses and the diagnostic and intervention status of participants in the study.
In four studies, blinding of outcome assessment was carried out, but it may not have been possible to blind assessors to all of the outcomes. In these cases, where for example, maternal self reports were used or where outcome assessments were directly administered, we judged that the risk of bias was low. Cohen 1999b used a range of measures (SSP, Chatoor Play Scale, and Bayley Scales) that have proven reliability and validity and that were either scored independently from video-tapes by raters blind to treatment group or administered directly to the child, which reduces the effects of reporter bias. In Cicchetti 2006b, two independent raters each coded all of the video-tapes of the individual SSP sessions, and raters were unaware of the maltreatment status and group assignment of individual mother-child dyads. It was unclear if the cortisol assays were blinded; however, as we were unable to obtain disaggregated data for this outcome, and did not use it in our analyses, we rated this study as low for blinding of outcome assessment risk of bias. In Lieberman 1991, the SSP was coded by fully trained and experienced judges. The home and laboratory measures were scored by coders who were blind to the 12-month classification and group assignment of the dyads. Different coders scored the home and laboratory measures. In the 24-month laboratory situation, the free-play measures and reunion measures were scored by different coders. Intervention process raters had no access to other outcome data. These procedures were adopted to protect the independence of the different ratings. In Salomonsson 2011a, outcome assessors were not blinded for assessments of depression (EPDS), but this was a maternal self report and unlikely to introduce bias. Two raters with substantial clinical infant experience, blind to case history, interview content and assignments, rated EAS.
4.2.3 Incomplete outcome data (attrition bias)
We judged seven of the eight included studies to be at low risk of attrition bias with attrition being well investigated and explained, and any effects on the balance of baseline characteristics in the intervention and control groups was not significant (Cicchetti 1999a; Cohen 1999b; Cooper 2003a; Lieberman 1991; Robert-Tissot 1996a; Salomonsson 2011a; Sleed 2013a).
We rated Cicchetti 2006b, as unclear for attrition bias, as we were uncertain if the decision to aggregate those participants who were randomised to receive treatment but declined into the no treatment control group, introduced a source of bias.
4.2.4 Selective reporting (reporting bias)
We judged two out of the eight included studies to be at high risk of selective outcome reporting (Cicchetti 2006b; Robert-Tissot 1996a). In both studies, several outcomes were not reported for the intervention group; in some cases, two intervention groups were combined and compared with the control group or a nonclinical sample (non-clinical being a non-randomised control group). In Cicchetti 2006b, pre-specified outcomes were reported, but intervention group data were aggregated and presented alongside randomised control group and non-maltreated non-randomised comparison groups.
We rated the risk of bias from selective reporting in Cicchetti 1999a as unclear. Although some outcomes were reported selectively, it was unclear if this would have introduced bias. We also rated Cohen 1999b as unclear for selective outcome reporting because the number of participants in each attachment category was not reported (although the patterns of change were reported). We were unable to obtain the attachment category data from the principal investigator, as the records were no longer available.
We judged the remaining four studies as being at low risk of reporting bias because all pre-specified outcomes were reported according to the group assignment (Cooper 2003a; Lieberman 1991; Salomonsson 2011a; Sleed 2013a).
4.2.5 Other potential sources of bias
We also examined baseline or pre-treatment means where available, to assess whether there was any imbalance in terms of participant characteristics. Although there were some differences in the baseline demographics, this was not thought to introduce any additional sources of bias and we rated all studies as low risk.
4.3 EFFECTS OF INTERVENTIONS
We summarised the results of our meta-analyses below. Data derived from single studies that could not be pooled in meta-analyses are provided in the Data and analyses section, Comparison 3 and Comparison 4.
For effect sizes of SMD, values greater than 0.70 have been treated as large; values between 0.40 and 0.70 as moderate; and values less than 0.40 but greater than 0.10 as small (Higgins 2011).
We have summarised the results below under headings corresponding to the primary and secondary outcomes outlined in the Types of outcome measures section and in the Summary of findings table 1 and Summary of findings table 2.
In the text below, numbers given are the total number of participants randomised. Where it has been possible to calculate an effect size, we have reported this with 95% CIs. Where the calculated effect size was statistically significant (P value < 0.05), we stated whether the result favours the intervention (i.e. PIP) group.
We have indicated which outcome measures were derived from participant self reports.
Comparison 1. Parent-infant psychotherapy versus control
Parent outcomes
Primary parent outcomes
1.1 Parent mental health: depression (dichotomous data) postintervention
One meta-analysis of three studies reported the number of subsequent episodes of depression post-intervention using the Diagnostic Interview Schedule - Version III - Revised (DIS-III-R; Cicchetti 1999a), DSM-III-R (Cooper 2003a), or CES-D (Sleed 2013a). The results of one meta-analysis using a random-effects model showed no significant difference between PIP and control groups (RR 0.74, 95% CI 0.52 to 1.04, 278 participants, Analysis 1.1). Heterogeneity was low (Chi2 = 2.30; degree of freedom (df) = 2; P value = 0.32; tau2 = 0.01; I2 = 13%). We rated the overall quality of the evidence as low (Summary of findings table 1).
We assessed the impact of the inclusion of data from one cluster-randomised trial (Sleed 2013a) in further sensitivity analyses (see ‘Sensitivity analyses: parent-infant psychotherapy versus control, sensitivity analyses assuming different values of intraclass correlation coefficient‘) .
1.2 Parent mental health: depression (continuous data) postintervention
Four studies reported a continuous measure of depression at post-intervention, which was measured using the CES-D Scale (Sleed 2013a), BDI (Cicchetti 1999a), and EPDS (Cooper 2003a; Salomonsson 2011a. The result shows no difference between PIP and control groups (SMD -0.22, 95% CI -0.46 to 0.02, 356 participants, Analysis 1.2). Heterogeneity was low (Chi2 = 3.75; df = 3; P value = 0.29; tau2 = 0.01, I2 = 20%). We rated the overall quality of the evidence as low for depression at postintervention (Summary of findings table 1).
We assessed the impact of the inclusion of data from one cluster randomised trial (Sleed 2013a) in further sensitivity analyses (see ‘Sensitivity analyses: parent-infant psychotherapy versus control, sensitivity analyses assuming different values of intraclass correlation coefficient‘) .
Secondary parent outcomes
There were no studies comparing PIP and a control intervention reporting secondary parent outcomes that could be pooled in a meta-analysis.
Parent-infant relationship outcomes
Primary parent-infant relationship outcomes
1.3 Maternal sensitivity meta-analysis: post-intervention
Four studies reported maternal sensitivity post-intervention, which was measured using the CIB scale (dyadic attunement) (Sleed 2013a); EAS (sensitivity subscale) (Salomonsson 2011a); video-taped interaction rated using the sensitivity dimension from the Murray 1996 scales (Cooper 2003a); and maternal empathic responsiveness/empathy post-intervention, which was measured using video-taped free play episodes and coded using a non-standardised nine-point scale based on body orientation, posture, facial expressions, and timing and content of responses to provide a measures of empathic responsiveness/empathy (Lieberman 1991). There was no statistically significant difference between PIP and control groups (SMD - 0.13, 95% CI -0.64 to 0.38, 283 participants, Analysis 1.3). There were significant levels of heterogeneity (Chi2 = 13.00; df = 3; P value = 0.005; tau2 = 0.21; I2 = 77%). We assessed the impact of the inclusion of data from one cluster-randomised trial (Sleed 2013a) in further sensitivity analyses (see ‘Sensitivity analyses: parent-infant psychotherapy versus control, sensitivity analyses assuming different values of intraclass correlation coefficient‘).
1.4 Child involvement meta-analysis: post-intervention
Two studies reported child involvement post-intervention, which used the CIB scales (i.e. involvement of the child subscale) (Sleed 2013a), and the EAS (i.e. infant involvement subscale) (Salomonsson 2011a). There was no statistically significant difference between intervention and control groups (SMD -0.01, 95% CI -0.32 to 0.30, 163 participants, Analysis 1.4). There was no evidence of heterogeneity (Chi2 = 0.39; df = 1; P value = 0.53; tau2 = 0, I2 = 0%).
We assessed the impact of the inclusion of data from one cluster-randomised trial (Sleed 2013a) in further sensitivity analyses (see ‘Sensitivity analyses: parent-infant psychotherapy versus control, sensitivity analyses assuming different values of intraclass correlation coefficient‘). The impact of the inclusion of data from one cluster randomised trial was assessed following adjustment for clustering (i.e. using an imputed ICC: 0.2, 0.4, 0.6, 0.8), and the result remains largely unchanged (i.e. slightly wider 95% CIs due to the inflated standard errors). The results showed no significant difference between PIP and control groups (SMD -0.07, 95% CI -0.46 to 0.32 if ICC = 0.2; SMD -0.09, 95% CI -0.50 to 0.33 if ICC = 0.4; SMD -0.09, 95% CI -0.52 to 0.33 if ICC = 0.6; SMD -0.10, 95% CI -0.53 to 0.33 if ICC = 0.8, Analysis 5.4).
1.5 Maternal positive engagement meta-analysis: post-intervention
Three studies measured maternal positive engagement at post-intervention, which used the CIB scale (i.e. interactions where the parent looked and talked to her baby positively, did not appear depressed, and was enthusiastic in engaging with her baby) (Sleed 2013a); the ‘structuring’ domain of the EAS (i.e. assesses the extent to which the mother provided appropriate guidance and suggestions during videotaped mother and baby interactions) (Salomonsson 2011a), and maternal engagement using coded video-taped free play sessions in a laboratory session rated on a seven-point scale for frequency, quality and persistence of effort to initiate interaction with the child (Lieberman 1991). There was no statistically significant difference between PIP and control groups (SMD -0.16, 95% CI -0.46 to 0.15, 216 participants, Analysis 1.5). Heterogeneity was low (Chi2 = 2.53; df = 2; P value = 0.28; tau2 = 0.02; I2 = 21%).
We assessed the impact of the inclusion of data from one cluster-randomised trial (Sleed 2013a) in further sensitivity analyses (see ‘Sensitivity analyses: parent-infant psychotherapy versus control, sensitivity analyses assuming different values of intraclass correlation coefficient‘).
Infant outcomes
Primary infant outcomes
Infant emotional well-being
1.6 Infant attachment category: post-intervention
Two studies assessed attachment classification (i.e. secure, avoidant, resistant, disorganised) using the SSP at post-intervention (Cicchetti 1999a; Cicchetti 2006b). We conducted meta-analyses for each type of infant attachment style. The data showed a statistically significant difference favouring the PIP group for the outcome of secure attachment (RR 8.93, 95% CI 1.25 to 63.70; P value = 0.03, 168 participants, Analysis 1.6). There was substantial heterogeneity (Chi2 = 3.71; df = 1; P value = 0.05; tau2 = 1.54; I2 = 73%) and we rated the overall quality of the evidence as very low (Summary of findings table 1).
We found that significantly fewer infants were avoidant in the PIP group compared with the control group (RR 0.48, 95% CI 0.24 to 0.95, P value = 0.03, 168 participants, Analysis 1.6). There was low heterogeneity (Chi2 = 0.06; df = 1; P value = 0.81; tau2 = 0; I2 = 0%). Significantly fewer infants in the PIP group compared with the control group were disorganised (RR 0.32, 95% CI 0.17 to 0.58, 168 participants, Analysis 1.6). We rated the overall quality of the evidence as low (Summary of findings table 1). There was low heterogeneity (Chi2 = 0.26; df = 1; P value = 0.61; tau2 = 0; I2 = 0%). However, there were no statistically significant differences between the PIP group and the control group at post-intervention for the resistant category (RR 0.69, 95% CI 0.16 to 2.97, 168 participants, Analysis 1.6). There was low heterogeneity (Chi2 = 0.16; df = 1; P value = 0.69; tau2 = 0, I2 = 0%).
1.7 Infant attachment category: follow-up
Two studies reported attachment category (i.e. secure, avoidant, resistant, disorganised) using the SSP at follow-up (Cicchetti 2006b; Cooper 2003a). There was a statistically significant difference favouring the PIP group compared with the control group for the number of infants securely attached, that is, more infants in the PIP group were securely attached by the end of the intervention (RR 3.30, 95% CI 1.82 to 6.00, P value < 0.0001, 129 participants, Analysis 1.7). We rated the overall quality of the evidence as low for secure infant attachment (Summary of findings table 1). Significantly more infants in the control group were avoidant (RR 0.33, 95% CI 0.15 to 0.76, P value = 0.81, 129 participants, Analysis 1.7). However, there were no differences between the groups in the proportions of infants classified as resistant (RR 0.57, 95% CI 0.11 to 3.07, 129 participants, Analysis 1.7) or disorganised (RR 0.80, 95% CI 0.29 to 2.19, 129 participants, Analysis 1.7). We rated the overall quality of the evidence as very low for the disorganised infant attachment outcome (Summary of findings table 1). There was low heterogeneity for the secure, avoidant and resistant categories (tau2 = 0.00; I2 = 0%, Analysis 1.7). There was substantial heterogeneity for the disorganised category (Chi2 = 2.04; df = 1; P value = 0.15; tau2 = 0.27; I2 = 51%, Analysis 1.7).
1.8 Infant attachment change: post-intervention
Two studies reported whether participants had changed attachment category by the end of the intervention (i.e. immediately post-intervention) (Cicchetti 1999a; Cicchetti 2006b). Significantly more infants in the PIP group had moved from insecure at pre-intervention to secure at post-intervention (RR 11.45, 95% CI 3.11 to 42.08; P value = 0.0002, 168 participants, Analysis 1.8). There were moderate levels of heterogeneity (Chi2 = 1.61; df = 1; P value = 0.21; tau2 = 0.39; I2 = 38%). There were more infants who were secure at pre-intervention and remained secure at post-intervention (stably secure) in the PIP groups, but this was not statistically significant (RR 2.28, 95% CI 0.41 to 12.56, 168 participants, Analysis 1.8). There were moderate levels of heterogeneity (Chi2 = 1.46; df = 1; P value = 0.23; tau2 = 0.66; I2 = 31%).
There was no statistically significant difference between the number of participants whose attachment category changed from secure at pre-intervention to insecure at post-intervention (RR 0.09, 95% CI 0.01 to 1.56, 168 participants, Analysis 1.8). (Note: it was not possible to assess heterogeneity in this meta-analysis because no participants changed from secure to insecure categories; therefore, statistical heterogeneity is not estimable.)
Although more infants in the control group were insecure at pre- and postintervention (stably insecure), there was no significant difference in children who were stably insecure (RR 0.56, 95% CI 0.26 to 1.22, 168 participants, Analysis 1.8). There were high levels of heterogeneity (Chi2 = 3.15; df = 1; P value = 0.08; tau2 = 0.22; I2 = 68%).
1.9 Infant problem behaviours, angry or externalising behaviours meta-analysis: follow-up
Two studies reported angry or externalising behaviour at follow-up, which was measured using the CBCL (externalising subscale) (Cicchetti 2006b); and the BSQ, modified for use with this age group (Cooper 2003a). There was no statistically significant difference between the two groups (SMD 0.22, 95% CI -0.34 to 0.77, 131 participants, Analysis 1.9). There was substantial heterogeneity (Chi2 = 1.91; df = 1; P value = 0.17; tau2 = 0.08; I2 = 48%).
Secondary infant outcomes
1.10 Infant cognitive development meta-analysis: post-intervention
Two studies measured infant cognitive development post-intervention using the Weschler Preschool Scales of Intelligence (WPPSI; full scale) (Cicchetti 1999a); and the ASQ: SE (maternal report) (Salomonsson 2011a). There was no statistically significant difference between PIP and control groups (SMD -0.15, 95% CI -0.82 to 0.51, 172 participants, Analysis 1.10). There were significant levels of heterogeneity (Chi2 = 4.76; df = 1; P value = 0.03; tau2 = 0.18; I2 = 79%).
Adverse effects
Other than the potential for the therapy to have adverse effects on the attachment of the infant (see 1.6 Infant attachment category: post-intervention; 1.7 Infant attachment category: follow-up; and 1.8 infant attachment change: pre- to postintervention), no studies comparing PIP with a control group reported adverse effects.
Comparison 2. Parent-infant psychotherapy versus alternative (non-parent-infant psychotherapy) intervention
Parent outcomes
Primary parent outcomes
2.1.1 Parent mental health: depression (continuous data): postintervention
Note: in Cohen 1999b, PPT refers to the standard PPT, that is, PIP and WWW to the infant-led PIP.
Two studies reported continuous measures of depression post-intervention, which was measured using the BDI (self report) (Cohen 1999b); and the EPDS (Cooper 2003a). There was no statistically significant difference between PPT and WWW (SMD 0.23, 95% CI -0.49 to 0.95, 172 participants, Analysis 2.1). There were significant levels of heterogeneity (Chi2 = 4.36; df = 1; P value = 0.04; tau2 = 0.21; I2 = 77%). We rated the quality of the evidence as very low (Summary of findings table 2).
2.1.2 Parent mental health: depression (continuous data): follow-up
Two studies reported a continuous measure of depression at follow-up, which was measured using the EPDS at follow-up (Cooper 2003); and the BDI at six-month follow-up (Cohen 1999b). There was no statistically significant difference between PIP and non-PIP at 18 months (SMD 0.13, 95% CI -0.23 to 0.5, 160 participants, Analysis 2.1) and levels of heterogeneity were low (I2 = 2%). We rated the quality of the evidence as low (Summary of findings table 2).
Secondary parent outcomes
There were no studies comparing PIP and an alternative intervention reporting secondary parent outcomes that could be pooled in a meta-analysis.
Parent-infant relationship outcomes
Primary parent-infant relationship outcomes
2.2.1 Maternal sensitivity: post-intervention
Three studies reported maternal sensitivity post-intervention, which was measured using the Chatoor Play Scale - subscale dyadic reciprocity (Cohen 1999b), the Crittenden CARE-Index (Robert-Tissot 1996a), and the GRS of video-taped mother-infant interactions (Cooper 2003a). There was no statistically significant difference between PIP and non-PIP groups (SMD 0.20, 96% CI -0.27 to 0.66, 248 participants, Analysis 2.2). There were marginally significant levels of heterogeneity (Chi2 = 5.81; df = 2; P value = 0.05; tau2 = 0.11; I2 = 66%).
2.2.2 Maternal sensitivity: follow-up
Two studies reported maternal sensitivity at six-month follow-up using the Chatoor Play Scale - subscale dyadic reciprocity (Cohen 1999b), and the Crittenden CARE-Index (Robert-Tissot 1996a). There was no statistically significant difference between the PIP and the non-PIP groups (SMD 0.05, 95% CI -0.69 to 0.80, 133 participants, Analysis 2.2). There were significant levels of heterogeneity (Chi2 = 4.6; df = 1; P value = 0.03; tau2 = 0.23; I2 = 78%).
Infant mental health and development
Primary infant outcomes
2.3 Infant attachment category: post-intervention
2.3.1 Attachment category secure: post-intervention
Two studies comparing PIP versus non-PIP reported attachment status, which was measured using the SSP (Cicchetti 2006b; Cohen 1999b). There was no statistically significant difference in the numbers of infants categorised as having secure attachment between PIP and non-PIP groups (RR 0.83, 95% CI 0.32 to 2.13, 102 participants, Analysis 2.3). There was a high level of heterogeneity (Chi2 = 2.6; df = 1; P value = 0.11; tau2 = 0.3; I2 = 62%) and we rated the overall quality of the evidence as very low (Summary of findings table 2).
2.3.2 Attachment category avoidant/resistant/disorganised: postintervention
Only one study comparing PIP versus non-PIP reported the proportion of infants with avoidant, resistant or disorganised attachment status at post-intervention (Cicchetti 2006b). There was no statistically significant difference between PIP and non-PIP groups in the numbers of infants categorised as avoidant (RR 4.60, 95% CI 0.20 to 105.62, 36 participants); or disorganised (RR 0.79, 95% CI 0.34 to 1.82, 36 participants, Analysis 2.3). No infants were categorised as resistant. We rated the overall quality of the evidence for the disorganised outcome as low (Summary of findings table 2).
2.4 Infant attachment category: follow-up
Two studies comparing PIP versus non-PIP reported the proportion of children with secure, avoidant, resistant or disorganised attachment status at follow-up (Cicchetti 2006b; Cooper 2003a). There was no significant difference between the groups for the secure category (RR 1.61, 95% CI 0.83 to 3.12, 136 participants, Analysis 2.4); this subgroup had moderate levels of heterogeneity (Chi2 = 1.79; df = 1; P value = 0.18; tau2 = 0.11; I2 = 44%) and we rated the overall quality of the evidence as very low. There was no significant difference between the groups for the avoidant category (RR 0.76, 95% CI 0.35 to 1.64, 136 participants, Analysis 2.4); resistant category (RR 1.48, 95% CI 0.32 to 6.98, 136 participants, Analysis 2.4) or disorganised category (RR 0.82, 95% CI 0.30 to 2.22, 136 participants, Analysis 2.4). There were high levels of heterogeneity for the disorganised category (Chi2 = 2.79; df = 1; tau2 = 0.33; P value = 0.09; I2 = 64%). For the disorganised infant attachment outcome, we rated the overall quality of the evidence as very low (Summary of findings table 2).
2.5 Infant attachment category change: pre- to post-intervention
2.5.1 Attachment change insecure to secure: pre- to post-intervention
Two studies comparing PIP versus non-PIP reported change in attachment category from insecure to secure pre- to post-intervention, and from secure to insecure pre-to post-intervention (Cicchetti 2006b; Cohen 1999b). There were no statistically significant differences between the PIP and non-PIP groups for change from insecure to secure (RR 0.48, 95% CI 0.06 to 4.10, 102 participants; I2 = 76%, Analysis 2.5) or secure to insecure (RR 1.59, 95% CI 0.49 to 5.13, 102 participants, Analysis 2.5).
2.5.2 Attachment change stable secure and stable insecure: pre- to postintervention
Two studies reported change in attachment from stable secure pre- to postintervention and from stable insecure pre- to post-intervention (Cicchetti 2006b; Cohen 1999b). There were no significant differences between intervention groups for the proportion of children who were stably secure (RR 1.59, 95% CI 0.28 to 8.93, 102 participants, Analysis 2.5) or stably insecure (RR 1.07, 95% CI 0.70 to 1.64; tau2 = 0; I2 = 0%, Analysis 2.5).
2.6 Infant problem behaviours
2.6.1 Angry or externalising behaviours: follow-up
Three studies comparing PIP versus non-PIP reported angry or externalising behaviours at follow-up, which was measured using the Mental Scales of the BSID (i.e. the Bayley Infant Behavior Rating Scale, subscale Emotion Regulation) at six-month follow-up (Cohen 1999b); the subscale of the CBCL for maternally reported externalising behaviours at 12-month follow-up (Cicchetti 2006b); and the BSQ, modified for use with the study's age group at six-month follow-up (Cooper 2003a). There was no statistically significant difference between groups (SMD -0.05, 95% CI -0.39 to 0.28,184 participants, Analysis 2.6). There was no evidence of heterogeneity (Chi2 = 1.07; df = 2; P value = 0.58; tau2 = 0; I2 = 0%).
Secondary infant outcomes
2.7 Infant cognitive development: follow-up
Two studies reported infant cognitive development at follow-up, which was measured using the Mental Scales of the BSID I or II to derive a DQ (Cohen 1999b), and the BSID MDI (Bayley 1969) in Cooper 2003a. There was no statistically significant difference between PIP and non-PIP groups (SMD 0.10, 95% CI -0.25 to 0.46, 162 participants, Analysis 2.7). There was no evidence of heterogeneity (Chi2 = 0; df = 1; P value = 0.95; tau2= 0; I2 = 0%).
2.8 Infant cortisol regulation; post-intervention and follow-up
Salivary cortisol samples were obtained by the principal investigators of Cicchetti 2006b and reported in Cicchetti 2011a, which described collection of morning saliva cortisol samples at pre-intervention, mid-intervention, post-intervention (26 months of age) and 12-month follow-up (38 months of age). We were unable to obtain disaggregated data from the maltreated PIP and non-PIP intervention groups. The study report stated that there were no differences in cortisol levels between the aggregated maltreated intervention groups (PIP and non-PIP) versus the community standard control group. By 12-month follow-up, infants in the maltreated intervention group had normalised cortisol levels (i.e. they were not different from a normal community sample).
Adverse effects
Other than the potential for the therapy to have adverse effects on the attachment of the infant (i.e. the extent to which infants who were securely attached at baseline were categorised as insecure at post-intervention or follow-up) (see 2.3 Infant attachment category: post-intervention; 2.4 Infant attachment category: follow-up; and 2.5 infant attachment change: pre- to post-intervention), none of the comparisons of PIP with a comparison treatment group reported adverse events.
Comparison 3. Parent-infant psychotherapy versus control (single-study results only)
We did not perform meta-analyses for the following outcomes as each outcome was only reported by one study.
Parent outcomes
Primary parent outcomes
3.1 Parent mental health
There was no statistically significant difference between PIP and the control group for any dichotomous measure of parental mental health depression assessed at nine-month, 18-month or five-year follow-up (DSM-III-R interview, Analysis 3.1; Cooper 2003a); or for any continuous measure of parental mental health depression assessed using the EPDS at nine-month, 18-month or five-year follow-up (Analysis 3.2; Cooper 2003a); or for parental stress assessed using the SPSQ and general maternal psychological stress measured using the GSI (SCL-90) at post intervention (Analysis 3.3; Analysis 3.4; Salomonsson 2011a).
3.2 Secondary parent outcomes
There was no statistically significant difference between PIP and the control group for parental reflective function at post-intervention using the PDI (Analysis 3.5; Sleed 2013a); maternal representations at post-intervention using the MORS object relations scale (Analysis 3.6; Sleed 2013a); or maternal attitudes using the maternal attitudes scale at post-intervention (Analysis 3.7; Lieberman 1991).
Parent-infant relationship outcomes
Primary and secondary parent-infant relationship outcomes
3.3 Parent-infant interaction
One trial assessed maternal reports of mother-infant relationship problems and behaviour management problems at post-intervention (Cooper 2003a). There were no statistically significant differences between the groups for either outcome, for the total number of mothers reporting problems, or of the mothers who had reported problems before treatment and who were still reporting problems after the treatment period (i.e. controlling for reported problems prior to treatment did not make a difference) (Analysis 3.8). There was no statistically significant difference for maternal intrusiveness post-intervention between PIP and control groups using the EAS scales (Analysis 3.9; Salomonsson 2011a). However, there were statistically significant findings favouring the PIP group for a well-adapted dyadic relationship using the Parent-Infant Relationship Global Assessment Scale (PIR-GAS) postintervention (SMD -0.58, 95% CI -1.05 to -0.12, P value = 0.01, Analysis 3.9; Salomonsson 2011a), and dyadic behaviour on reunion at post-intervention (SMD - 1.37, 95% CI -1.98 to -0.77, P value < 0.00001, Analysis 3.9; Lieberman 1991).
Infant outcomes
Primary infant outcomes
3.4 Attachment security
There was a significant difference in the intervention group for stability in infants classified as disorganised at both pre- and post-intervention using the SSP (RR 0.15, 95% CI 0.04 to 0.59, P value = 0.007) (note: disorganised attachment style is a negative outcome, this result favours the PIP group) and for infants who had changed to a secure attachment (RR 5.20, 95% CI 1.26 to 21.42, Analysis 3.10) (Cicchetti 1999a). There was no statistically significant difference between the groups for infants who changed from an initially disorganised classification to another insecure classification (Analysis 3.10). Using other measures of attachment security (Q-sort) at post-intervention, one study found that infants in the intervention group showed significantly less proximity avoidance (SMD -1.18, 95% CI -1.77 to -0.59, P value < 0.0001, Analysis 3.11), but no statistically significant difference for contact resistance (Lieberman 1991). There were no significant differences for attachment security between groups for the Q-Scale reported at postintervention (Analysis 3.12; Cicchetti 1999a).
3.5 Infant problem behaviours
There was no statistically significant difference between PIP and control group at post-intervention for any measure of infant problem behaviours (Analysis 3.13) at follow-up using the scale for angry or externalising behaviours, that is the mean number of instances when a toddler hit, kicked, bit or yelled at mother (Lieberman 1991), or long-term (five years) follow-up based on maternal reports (Rutter A2 report) or teacher-report (PBCL) (Cooper 2003a). There was no significant difference for restriction of affect measured on a seven-point scale post-intervention (Lieberman 1991), or total problem and internalising behaviours at 12-month follow-up using the CBCL total and internalising scales (Analysis 3.14; Cicchetti 2006b).
3.6 Infant cognitive development meta-analysis
There were no statistically significant results for infant cognitive development at 18-months follow-up using the BSID MDI (Cooper 2003a), or long-term follow-up (measured at five years using the McCarthy Scales of Children's Abilities) (Analysis 3.15; Cooper 2003a). One study reported infant performance IQ and verbal IQ postintervention (Analysis 3.16; Cicchetti 1999a). The infants of mothers who had experienced a further episode of depression had significantly different full-scale and verbal IQ scores at post-intervention favouring the PIP group (using the WPPSI-R, full scale IQ: SMD -0.99, 95% CI -1.77 to -0.22; verbal IQ: SMD -0.89; 95% CI -1.66 to -0.12, 30 participants, Analysis 3.17).
Adverse effects
None of the comparisons of PIP with a control group in the above single-study analyses showed adverse changes attributable to PIP.
Comparison 4. Parent-infant psychotherapy versus alternative (non-parent-infant psychotherapy) intervention (single-study results only)
Note: in Cohen 1999b, PPT refers to the standard form, that is, PIP and WWW to the infant-led PIP.
Parent outcomes
Primary parent outcomes
4.1 Parent mental health (depression)
There were no statistically significant results for any measures of parent mental health (depression) in any of our analyses of single-study data at post-intervention or follow-up.
One study reported the number of mothers experiencing subsequent episodes of depression post-intervention using the DSM-III-R at nine-month follow-up, 18-month follow-up and five-year follow-up (Analysis 4.1; Cooper 2003a).
One study assessed depression using the EPDS at nine-month follow-up and at longterm follow-up (five years) (Cooper 2003a). Neither analyses favoured PIP or alternative intervention (Analysis 4.2).
4.2 Parenting stress
One study reported parenting stress using the PSI at post-intervention and follow-up (Cohen 1999b). None of the analyses for the parent or child domain of this scale showed a statistically significant difference between PIP and alternative intervention (Analysis 4.3).
Secondary parent outcomes
4.3 Parenting sense of competence
One study reported parent sense of competence using the PSOC scale at postintervention and follow-up (Cohen 1999b). None of the analyses for total satisfaction or total lack of efficacy favoured either PIP or alternative intervention (Analysis 4.4).
Parent-infant interaction
Primary and secondary parent-infant relationship outcomes
4.4 Parent-infant interaction: - maternal responsiveness, intrusiveness, and dyadic conflict
There were no statistically significant results for maternal responsiveness or intrusiveness measured at post-intervention or follow-up, using the Chatoor Play Scale (Cohen 1999b). However, there was a statistically significant difference for dyadic conflict measured by WWW that favoured the alternative intervention (SMD 0.54, 95% CI 0.01 to 1.07, P value = 0.04, 67 participants, Analysis 4.5).
4.5 Mother-infant relationship or infant behaviour management problems: post-intervention and follow-up
One study assessed maternal reports of mother-infant relationship problems and behaviour management problems at post-intervention (Cooper 2003a). There were no statistically significantly differences between the groups for the total number reporting problems and those who had initially reported problems and who were still reporting problems after therapy.
There was no statistically significant difference between the groups (WWW and PPT in Cohen 1999b) for maternal perception of problem severity, how effective the mothers felt in dealing with the problem or comfort in dealing with problems. There was no significant difference between PIP and non-PIP groups for either outcome at post-intervention or follow-up (Analysis 4.6; Analysis 4.7).
Infant mental health and development
Primary outcomes
4.6 Infant attachment category: change from initially insecure (avoidant - A, resistant - C or disorganised - D): pre- to post-intervention
Using data from Cohen 1999b, which assessed infant attachment category using the SSP, there were no statistically significant differences between the PIP or non-PIP groups for: 1. the change from initially insecure avoidant or resistant to disorganised; or 2. infants who were initially categorised as disorganised and changed to avoidant or resistant (organised but insecure) (Analysis 4.8). However, 3. insecure (avoidant or resistant) or disorganised to secure and organised (resistant) significantly favoured the WWW group (RR 0.35, 95% CI 0.13 to 0.99, P value = 0.05, 66 participants, Analysis 4.8); and 4. no change or less secure, favoured the PPT group (RR 1.35, 95% CI 1.02 to 1.79, P value = 0.04, 66 participants, Analysis 4.8).
There were no statistically significant differences between the groups for participants who had retained post-intervention gains or moved to a secure or organised attachment (Analysis 4.8).
4.7 Infant behaviours
One study found no statistically significant differences between the groups for angry or externalising behaviours at post-intervention, which was measured using the Mental Scales of the BSID (i.e. the Bayley Infant Behavior Rating Scale, subscale Emotion Regulation) (Cohen 1999b). In addition, there were no statistically significant differences between the groups in the total infant behaviour and internalising behaviour at six-month follow-up measured using the CBCL (Cicchetti 2006b), or for long-term follow-up using the PBCL teacher report at five years, or the Rutter A2 maternal report at five years (Cooper 2003a) (Analysis 4.9).
Secondary outcomes
4.8 Infant cognitive development
There were no statistically significant differences between the groups for infant cognitive development (infant competence measured using the BSID MDI at postintervention (Cohen 1999b), or at long-term (five-year) follow-up using the McCarthy General Cognitive Index (GCI) (Cooper 2003a) (Analysis 4.10), or when infant cognitive development was assessed in infants with and without developmental delay at post-intervention using the BSID MDI (Cohen 1999b) (Analysis 4.11).
Adverse effects
None of the comparisons of PIP with a treatment group in the above single study analyses showed adverse changes attributable to PIP, other than a difference between PIP (‘PPT‘) and an alternative intervention (infant-led WWW) for change in infant attachment (see 4.6 Infant attachment category: change from initially insecure). From pre- to post-intervention, more infants in the PPT group stayed in the same attachment category or became less secure from pre- to post-intervention (RR 1.35, 95% CI 1.02 to 1.79, P value = 0.04, 66 participants, Analysis 4.8; Cohen 1999b).
Sensitivity analyses: parent-infant psychotherapy versus control, sensitivity analyses assuming different values of intraclass correlation coefficient
We conducted sensitivity analyses assuming different values of ICC (i.e. using an imputed ICC: 0.2, 0.4, 0.6, 0.8) in the cluster-randomised trial (Sleed 2013a). The overall result of the analysis was robust to the use of imputed ICCs.
Parent outcomes
Primary parent outcomes
5.1 Parent mental health: depression (dichotomous data) postintervention
There were no significant differences between PIP and control groups (RR 0.69, 95% CI 0.46 to 1.03 if ICC = 0.2; RR 0.67, 95% CI 0.44 to 1.03 if ICC = 0.4; RR 0.67, 95% CI 0.43 to 1.03 if ICC = 0.6; RR 0.66, 95% CI 0.42 to 1.03 if ICC = 0.8, Analysis 5.1).
5.2 Parent mental health: depression (continuous data) postintervention
There were no significant differences between PIP and control groups (SMD -0.22, 95% CI -0.47 to 0.03 if ICC = 0.2; SMD -0.24, 95% CI -0.52 to 0.05 if ICC = 0.4; SMD -0.24 with 95% CI -0.53 to 0.05 if ICC = 0.6; SMD -0.24 with 95% CI -0.53 to 0.05 if ICC = 0.8, Analysis 5.2).
Secondary parent outcomes
5.3 Maternal sensitivity meta-analysis: post-intervention
There were no significant differences between PIP and control groups (SMD -0.16, 95% CI -0.65 to 0.32 if ICC = 0.2; SMD -0.19, 95% CI -0.67 to 0.29 if ICC = 0.4; SMD -0.22, 95% CI -0.69 to 0.26 if ICC = 0.6; SMD -0.23, 95% CI -0.71 to 0.25 if ICC = 0.8, Analysis 5.3).
Parent-infant relationship outcomes
Primary and secondary parent-infant relationship outcomes
5.4 Child involvement meta-analysis: post-intervention
There were no significant differences between PIP and control groups (SMD -0.07, 95% CI -0.46 to 0.32 if ICC = 0.2; SMD -0.09, 95% CI -0.50 to 0.33 if ICC = 0.4; SMD -0.09, 95% CI -0.52 to 0.33 if ICC = 0.6; SMD -0.10, 95% CI -0.53 to 0.33 if ICC = 0.8, Analysis 5.4).
5.5 Maternal positive engagement meta-analysis: post-intervention
There were no significant differences between PIP and control groups (SMD -0.23, 95% CI -0.55 to 0.09 if ICC = 0.2; SMD -0.25, 95% CI -0.58 to 0.08 if ICC = 0.4; SMD -0.26, 95% CI -0.60 to 0.07 if ICC = 0.6; SMD -0.27, 95% CI -0.60 to 0.07 if ICC = 0.8, Analysis 5.5).
Sensitivity analyses: parent-infant psychotherapy versus control, fixed-effect versus random-effects meta-analyses
Overall, there was little difference between point estimates for analyses using fixed-effect or random-effects meta-analyses with the exception of the following outcomes.
Parent outcomes
Primary parent outcomes
6.1 Parent mental health: depression (dichotomous data) postintervention
A fixed-effect meta-analysis found a statistically significant difference between PIP and control groups on the depression outcomes favouring PIP at post-intervention (fewer depressed parents in PIP group: RR 0.72, 95% CI 0.53 to 0.99 for the dichotomous depression outcomes; lower depression scores in PIP group: SMD - 0.21, 95% CI -0.42 to 0.00 for the continuous depression outcomes; analyses not shown).
Infant mental health and development
Primary outcomes
6.2 Infant attachment category: change from initially insecure
Using a fixed-effect meta-analysis, there was a statistically significant difference in children who remained stably insecure at post-intervention favouring the control group (RR 0.57, 95% CI 0.38 to 0.87; analyses not shown).
5 Discussion
5.1 SUMMARY OF MAIN RESULTS
Overall
We included eight studies (846 participants), four compared PIP with control groups only (Cicchetti 1999a; Lieberman 1991; Salomonsson 2011a; Sleed 2013a), and four compared PIP with another treatment. of the four comparing PIP with another treatment, one compared a representative PPT with an ‘infant-led’ PIP (WWW) (Cohen 1999b); one compared PIP with interaction guidance (Robert-Tissot 1996a); one comprised three arms permitting a comparison of PIP with both a no-treatment community control group and a psychoeducational parent training programme (Cicchetti 2006b); and one employed a randomised four-arm comparison of PIP, CBT, non-directive counselling, and routine primary care (in which, for the purposes of this review, we aggregated data from the counselling and CBT arms (non-psychodynamic interventions; Cooper 2003a).
All eight studies provided post-intervention data. Three studies provided follow-up data at six and 12 months (Cicchetti 2006b; Cohen 1999b; Cooper 2003a), and one study provided long-term (five-year) follow-up data (Cooper 2003a).
When describing the null findings for most of the outcomes synthesised in the review, it should be noted that these provide no evidence of an effect (rather than evidence of no effect) and may be due to low statistical power given the small number of included studies and the imprecision in the random-effects variance component. Where we report statistical significance, the statistic is reliable but it cannot be inferred that the finding is clinically important or that it should be the sole factor taken into account in decision-making.
Parent-infant psychotherapy versus control group
Six studies contributed data to the PIP versus control comparisons producing 19 meta-analyses measuring the primary outcomes of: parental depression (both dichotomous and continuous data at post-intervention); measures of the parent-infant relationship, including the primary outcomes of maternal sensitivity, parent engagement and child involvement at post-intervention; primary infant outcomes of attachment category at post-intervention and follow-up (secure, avoidant, disorganised, resistant), attachment change (insecure to secure, stable secure, secure to insecure, stable insecure); and infant behaviour at follow-up. The studies also provided data for analyses of the secondary outcomes of infant cognitive development at post-intervention.
The results showed no clear evidence of improvements in incidence of parental depression or parent-reported levels of depression, and no clear evidence of a difference for parent-infant interaction (i.e. maternal sensitivity, child involvement and parent engagement). There were improvements in the proportion of infants securely attached at post-intervention and a reduction in the number of infants with an avoidant attachment at post-intervention; there were also significantly fewer infants with disorganised attachment at post-intervention. Four meta-analyses were produced with follow-up data and showed significantly more PIP infants who were categorised as secure, and significantly fewer infants who were avoidant. There was no clear evidence of a difference favouring either the intervention group or the controls group for resistant attachment, and no evidence of effectiveness in reducing the numbers of infants with a disorganised attachment style. There was an increase in the proportion of children moving from insecure at pre-intervention to secure attachment at post-intervention favouring PIP. There were no differences between the groups for infant behaviour, or infant cognitive development (see Summary of findings table 1).
Parent-infant psychotherapy versus alternative treatment
Four studies contributed data to the PIP versus alternative treatment analyses producing 15 meta-analyses measuring the primary outcomes of parent mental health (depression); the parent-infant interaction outcome of maternal sensitivity, and the primary infant outcomes of attachment category (secure, avoidant, resistant, disorganised) and attachment change (insecure to secure, stable insecure), although meta-analysis was not possible for infants who were stable secure, or those who changed from secure to insecure (i.e. because no events occurred in the PIP group) or for infant behaviour. Meta-analysis was possible for the secondary outcome of infant cognitive development. None of the meta-analyses for primary or secondary outcomes of PIP versus alternative treatment at post-intervention or follow-up showed significant differences in outcome between PIP and alternative treatment interventions (see Summary of findings table 2).
Adverse effects
Overall, this review found little evidence of adverse effects. Three of the included studies assessed whether PIP had adverse effects in terms of infant attachment security (i.e. the extent to which infants who were securely attached at baseline were categorised as insecure at post-intervention or follow-up) (Cicchetti 1999a; Cicchetti 2006b; Cohen 1999b). In our meta-analyses, none of the comparisons of PIP with either a control or comparison treatment group showed adverse changes for infant attachment or any other outcome. However, results of an analysis from only one study showed a difference between PIP and an alternative intervention (infant-led WWW) for adverse effects (Cohen 1999b). From pre- to post-intervention, more infants in the PIP group stayed in the same attachment category or became less secure from pre- to post-intervention.
5.2 OVERALL COMPLETENESS AND APPLICABILITY OF EVIDENCE
The data presented here are complete for the published studies that were identified. We were also able to identify five ongoing studies evaluating the effectiveness of PIPs that have not yet reported, and the results from these will be included at the next update.
In terms of the applicability of the evidence, there was some variation in the type of PIP being evaluated, including two types of infant-led psychotherapy (Cohen 1999b; Salomonsson 2011a), in addition to the standard representational types of therapy. The duration of intervention also varied (i.e. ranging from eight sessions in Sleed 2013a to up to 49 weeks in Cicchetti 2006b), as did the content and focus. For example, the focus varied from child attachment in the context of maltreating families (Cicchetti 2006b), maternal depression or mothers' feelings of failure in bonding with their infant (Cicchetti 1999a; Cohen 1999b; Cooper 2003a), parental psychosocial functioning and the parent-child relationship (e.g. Cooper 2003a; Lieberman 1991; Sleed 2013a), and functional and behavioural problems in children (Cohen 1999b; Robert-Tissot 1996a). The target populations were diverse, including depressed women, anxiously attached dyads, maltreating parents and incarcerated women, and there was little consistency in the outcomes measured or clarity about the relationship between outcomes and programme mechanisms. This variability was manifest in high levels of statistical heterogeneity for some of the meta-analyses that we conducted, including some of the measures of attachment that we found to be significant.
5.3 QUALITY OF THE EVIDENCE
We included eight studies comprising 846 initially randomised participants. Our risk of bias estimates show that overall the quality of the included studies was poor (Figure 2; Figure 3). Many studies had limitations in their design or implementation or were unclear about important quality criteria, including randomisation, and were at high risk for allocation concealment, sequence generation and blinding. It should be noted that all of the studies were judged at high risk of performance bias because it is not possible to blind participants and personnel in studies of this nature. Many studies did not provide adequate information for assessment of these domains. In two studies, the intervention groups were combined (Cicchetti 2006b; Robert-Tissot 1996a). In all cases of uncertainly, we attempted to obtain further information or disaggregated data from the trial investigator, but, where this was not available, it was because the investigator no longer had access to historical trial data or was unable to provide additional information.
None of the included studies conformed with the CONSORT (Consolidated Standards of Reporting Trials) reporting requirements for an RCT (Schulz 2010). We rated the overall quality of the evidence as low or very low for the key outcomes of parental depression and secure or disorganised infant attachment (Summary of findings table 1; Summary of findings table 2), resulting in lower confidence in the estimate of effect for those outcomes. Where we downgraded the evidence, it was because there was risk of bias in the trial, out of six estimates of potential bias (random sequence generation, allocation concealment, incomplete outcome data, selective reporting, blinding of participants and personnel, blinding of outcome assessment), only two studies obtained scores of four or more. The included studies involved relatively few participants and wide CIs (imprecision), although it is accepted that large-scale community studies of this type would be very difficult to perform due to the one-to-one nature of the intervention. The results of many meta-analyses had moderate to high levels of statistical heterogeneity (inconsistency). One explanation for this may be the clinical heterogeneity between the trials, with trials having different aims, participants (depressed parent, infants at risk of abuse or insecure attachment, prison populations, community samples), methods of delivery (i.e. individual or group-based), and primary outcome measures, for example, parent (maternal) mental health (depression) and infant attachment. This clinical heterogeneity that was present in the eight included studies, alongside the limited available evidence, precluded us from reaching any firm conclusions about the effectiveness of PIPs.
5.4 POTENTIAL BIASES IN THE REVIEW PROCESS
Potential biases in the review process were limited. We estimated the SMD by calculating the treatment effect for each outcome in each study by dividing the MD in post-intervention scores for the intervention and treatment groups by the pooled standard deviation. However, it should be noted that random allocation does not guarantee equality of means between groups at pre-test, and also that post-test standard deviation may be inflated by a differential response to intervention, and may underestimate the effect size attributable to the intervention.
In addition, we did not take into account the possibility of ICC that can occur in group interventions. Although we corrected for unit analysis issues arising from cluster-randomisation, we did not investigate further the clustering effect of individually randomised trials with group delivered therapies. This could mean that we have overestimated the significance of the findings.
We contacted the study investigators to provide missing data, but where this was not provided, we did not choose to impute missing data for two reasons. In terms of dichotomous data, there was no clear judgement about the value of performing sensitivity analyses based on best-case/worst-case scenarios or their alternative. For continuous data, we would have required access to the raw data (i.e. in order to impute missing data by using the last-observation-carried-forward). Therefore, the analysis was limited to outcomes for which published data were available.
We had planned to carry out additional subgroup analyses to explore the programme components that appeared to be associated with more effective outcomes, and factors that modified intervention effectiveness, but there were too few included studies in each meta-analysis to do this. There were similarly too few studies to conduct sensitivity analyses to examine the impact of study design or quality.
5.5 AGREEMENTS AND DISAGREEMENTS WITH OTHER STUDIES OR REVIEWS
To date, there has been only two systematic reviews of parent-infant (Singleton 2004) or attachment-based (Bakermans-Kranenburg 2003) therapies.
The first included 25 studies of parent-infant therapies with children from birth to 36 months of age, but included studies that were not RCTs, alongside diverse populations (e.g. low birthweight babies, low-income families, infants with cerebral palsy) and diverse interventions (e.g. infant massage, home visiting, PIP) (Singleton 2004). The results showed significant improvements in infant mental health (d = 0.28, where d is the effect size (Cohen 1977)), parent-infant relationship (d = 0.56), parent ability (d = 1.65), but not infant development (d = -0.06) with significant heterogeneity for all meta-analyses.
The second review included 70 studies assessing the effectiveness of a diverse range of attachment interventions, including PIP, video-interaction guidance, social support, etc. (Bakermans-Kranenburg 2003). The results showed that attachment-based interventions appeared to be effective in reducing insensitive parenting (d = 0.33) and infant attachment insecurity (d = 0.20). The most effective interventions used a moderate number of sessions and a clear-cut behavioural focus in families with, as well as without, multiple problems. Interventions that were more effective in enhancing parental sensitivity were also more effective in enhancing attachment security, which supports the notion of a causal role of sensitivity in shaping attachment. In our review, there was no difference in attachment outcomes between PIP versus alternative treatment group, and the reasons for this were unclear. Some non-psychodynamic interventions are relationship-based and it may be that interventions of any type that promote parental sensitivity result in more secure infant attachment without the need for further input that specifically targets infant attachment.
It is difficult to assess the degree of consistency between the findings of the current review and the two earlier reviews because neither review disaggregated the effects for different intervention types, and positive effects, therefore, may be due to intervention components other than PIP (Bakermans-Kranenburg 2003; Singleton 2004).
6 Authors' conclusions
6.1 IMPLICATIONS FOR PRACTICE
The results of this review suggest that parent-infant psychotherapy (PIP) is a promising model in terms of improving infant attachment security in high-risk populations, including maltreatment and prisons, but that there is currently limited evidence of benefit across many other outcomes measured, including maternal representations and parent-infant interaction. Several studies were lacking in rigour, and there was significant statistical heterogeneity affecting some of the key outcomes, suggesting the need for caution when interpreting these findings.
With regard to the effectiveness of PIP relative to other methods of working, the evidence is again inconclusive, partly as a result of the diverse interventions with which PIP was compared. While early evidence suggested that interventions for parent mental health problems, such as depression, which did not contain specific components that explicitly target parent-infant interaction were not effective, the data from this review have suggested that PIP has not yet been demonstrated to be superior to other methods of working with high-risk groups of parents, for reasons that are unclear and that there is an urgent need for further high-quality research.
None of the included studies provided data regarding the cost of implementing PIP or its cost-effectiveness relative to other methods of intervening.
6.2 IMPLICATIONS FOR RESEARCH
Further rigorous research is needed in the form of high-quality primary studies that make use of adequate methods of random allocation and that minimise bias in their conduct and reporting. There is also a need to establish whether the current findings can be replicated with other groups of parent-infant dyads, and to identify the mechanisms of change. Existing research suggests that parental reflective functioning and interactive regulation are key to the development of attachment security in infants, but the limited data available in the current review found no evidence of an impact on mediatory factors such as parent-infant interaction or reflective functioning. Future research should identify a set of core outcomes for both parents and infants that would facilitate comparisons between different studies, and that would improve our ability to assess whether this intervention has an impact in terms of key moderating variables. Further research is also needed to assess the impact of PIP on father-infant dyads or mother-father-infant triads. The included studies in this review were all conducted in high-income northern hemisphere countries, albeit with different population groups, and future studies should evaluate the impact of relationship-based interventions in lower-income countries.
There is a lack of consensus about what a core PIP model should involve, with considerable evolution of the basic model (e.g. introduction of video-interaction guidance and infant massage, etc.) in the absence of evidence regarding the benefits of the core PIP model compared with treatment-as-usual.
Future studies should aim to identify the active components involved in PIP, and compare standard PIP models with variations on the core PIP model of working, including attachment-based components for which there is already evidence of benefit with high-risk groups (e.g. video-interaction guidance).
Similarly, studies comparing alternative interventions should also include a control arm, and should focus on attachment-based interventions for which there is already evidence of benefit with high-risk groups (e.g. video-interaction guidance), in addition to variations on the core PIP model of working.
Future comparative studies should also aim to address whether different forms of therapy may be differentially effective for parents with different types of attachment insecurity (Bakermans-Kranenburg 1998), and provide cost-effectiveness data.
7 Acknowledgements
We acknowledge the financial support provided by PIP UK (see below for further information) to undertake this review, and the administrative and research support of the Cochrane Developmental, Psychosocial and Learning Problems Group and the Campbell Collaboration in undertaking this review.
This review is co-registered within the Cochrane Collaboration and a version of it appears on The Cochrane Library.
We gratefully acknowledge the additional data and information provided to us by the study investigators, Prof Christiane Robert-Tissot, Prof Alice Lieberman, Prof Lynne Murray and Prof Peter Cooper, Prof Nancy Cohen, Prof Dante Cicchetti, Dr Fred Rogosch, and Dr Michelle Sleed.
Ms Soili Larkin provided additional data extraction and data entry support to the review.
Dr Yinghui Wei carried out intraclass correlation coefficient analyses and provided statistical advice.
8 Contributions of authors
JB secured funding for the review and is the contact author and guarantor of the review. JB drafted the text of the review with CB and NM.
CB screened studies with JB and NM, extracted data with JB and NM, entered data into Review Manager 5 (RevMan 2012), characteristics of studies tables and analyses, carried out ‘Risk of bias’ assessments, drafted the analysis and results sections, edited the review, provided support to the authors in the use of Review Manager 5, and maintained the review reference management databases. CB also contacted principal investigators for further details and study characteristics.
NM screened literature searches for potentially included studies, finalised the included studies lists, provided advice about outcomes, commented and edited the text. NM also obtained additional data about the included studies.
NM and CB provided additional references and comments on the text of the protocol.
SL extracted data for the ‘Characteristics of included studies’ tables and carried out ‘Risk of bias’ assessments.
YW conducted the intraclass correlation coefficient analyses, gave advice on statistical methods, and assisted with production of the ‘Summary of findings’ tables.
9 Declarations of interest
Funding for this review has been received from PIP UK, a charitable organisation led by Andrea Leadsom (MP) to establish and evaluate parent-infant psychotherapy programmes.
Jane Barlow - editor with the Cochrane Developmental, Psychosocial and Learning Problems Group and Co-Chair of the Campbell Social Welfare Group.
Cathy Bennett - proprietor of Systematic Research Ltd, a company providing research services, and employee of that company; received a consultancy fee from the PIP UK grant for contribution to this review, as well as travel expenses for travel to work-related meetings and conferences. Cathy also received consultancy fees for other Cochrane reviews and work in evidence-based medicine. Cathy is a member of the data monitoring committee for the Barrett's Oesophagus Surveillance Study (BOSS), this work is not related to review writing.
Nick Midgley - employed by Anna Freud Centre, which offers a parent-infant psychotherapy service. One of the studies reviewed is based on work from this team; however, his role in the Centre is not connected with the parent-infant psychotherapy service in any way. Nick is a recipient of funding to conduct the review from PIP UK.
Soili Larkin - none known.
Yinghui Wei - received a consultancy fee for her work on this review.
10 Differences between protocol and review
We modified the protocol as follows. We stated that a systematic review of this topics is important as there is a need to develop empirically derived models that can support vulnerable parents and their children. We provided further details in the ‘Background section of the review to differentiate parent-infant psychotherapy (PIP) from other home visiting approaches. We clarified that we included interventions delivered to dyads in a small group setting; that psychotherapy focuses on improving infant attachment security by targeting parental internal working models (i.e. representational world) and working directly with the parent-infant relationship; that focus on the parent-infant relationship/interaction (not just maternal representations); and are aimed primarily at improving attachment security or infant socio-emotional functioning, or both. We clarified that we excluded interventions that were purely delivered to the parent (e.g. interpersonal psychotherapy), or which were primarily psycho-educational, or which were primarily based on other therapeutic models (e.g. cognitive behavioural therapy) or were focused on parent-infant interaction in a purely behavioural way. We excluded stand-alone video-interaction guidance interventions, but not studies in which video feedback was incorporated into a PIP that included the core components. We excluded psychotherapy working only with the parent. We clarified that we included birth parents (mothers and fathers), adoptive and kinship parents, but did not include foster parents. For the purposes of this review, we referred to the included children as infants irrespective of the term used to describe them in the study report (e.g. children, toddlers). We included studies of PIPs that included additional components (i.e. provided they still met the core inclusion criteria). We clarified that we recorded information about monitoring the fidelity of the intervention during the course of the trial, in order to ascertain whether or not the intervention was delivered as intended. We previously specified in our protocol that the maximum age of the infants at entry into the study should be 24 months or less; it should have stated that the The decisions about combining data were made post hoc and based on the categories of interventions, participants, and outcomes identified in the reviewed literature. For studies where there was more than one active intervention and only one control group, we had intended to select the intervention that most closely matched our inclusion criteria and exclude the others (see Higgins 2011, Chapter 16.5.4). However, because we found some head-to-head comparisons of PIP versus other interventions, we conducted analyses on intervention versus control and one intervention (PIP) versus another intervention, and we were able to include studies that used more than one intervention by splitting the participant numbers in the shared (PIP) intervention group. One study comprised three intervention arms and one control arm (Cooper 2003a). We combined the two non-PIP intervention groups into one to create two pairwise comparisons (intervention versus control and PIP versus other intervention), using statistical methods to pool the mean and standard deviation for continuous outcomes where applicable. We made a post-hoc decision based on editorial comments to conduct sensitivity analyses for a fixed-effect and a random-effects model and intraclass correlation coefficient (ICC) imputations. There was little difference between point estimates for analyses using a random-effects or a fixed-effect model, with no impact on the overall conclusions of treatment effects with the exceptions of two outcomes. We added a new subsection of “sensitivity analyses” where we summarised the results from sensitivity analyses. We included ‘Summary of findings’ tables for the outcomes of parental depression and infant attachment. We added a section to the methods of the review to describe how we constructed the ‘Summary of finding’ tables (GRADEpro 2014), and how we applied the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) system to rate the overall quality of evidence. For statistical assessment of heterogeneity, we have added the tau2 estimates for each meta-analysis. We now comment on the heterogeneity of each metaanalysis by looking at and reporting all of the following statistics: Chi2, P value, tau2, and I2 statistic. We edited the background text of this review after the publication of the protocol, in response to comments from Dr Aron Schlonsky, Campbell Collaboration Social Welfare Coordinating Group, and Dr Nick Midgley (review author).
11 Characteristics of studies
11.1 CHARACTERISTICS OF INCLUDED STUDIES
11.1.1 Cicchetti 2006b
Footnotes
AAPI: Adult-Adolescent Parenting Inventory; ANOVA: Analysis of Variance; AQS: Attachment Q-Set; ASQ: SE: Ages and Stages Questionnaire: Social-Emotional; BDI: Beck Depression Inventory; BSID: Bayley Scales of Infant Development; BSQ: Behavioural Screening Questionnaire; CARE-Index: Child-Adult Relationship Experimental Index; CBCL: Child Behavior Checklist; CBT: cognitive behavioural therapy; CES-D: Center for Epidemiologic Studies Depression Scale; CHCC: Child Health Centre Care; CIB: Coding Interactive Behavior; CPP: Child-Parent Psychotherapy; CPS: Child Protective Service;
CS: Community Standard controls; CTQ: Childhood Trauma Questionnaire; DAS: Dyadic Adjustment Scale; df: degrees of freedom; DIS-IV: Diagnostic Interview Schedule Version IV; DSM-III-R: Diagnostic and Statistical Manual of Mental Disorders - Third Edition - Revised (DSM-III-R); dyad: a mother and infant pair; EAS: Emotional Availability Scales; EPDS: Edinburgh Postnatal Depression Scale; FES: Family Environment Scale; GSI: General Severity Index; IPP: Infant-Parent Psychotherapy; IQ: intelligence quotient; ISEL: Interpersonal Support Evaluation List; ITT: intention to treat; KIA/KIDIES Profile (Kiddie-Infant Affect Profile of Child Mood and Affect; LEI: Life Event Inventory; MBQ: Maternal behaviour (Q set); MBUs: mother and baby units; MDI: Mental Development Index; MIP: mother-infant psychotherapy; MORS: Mother's Object Relations Scales; NHS: National Health Service (UK); NIMH: National Institute of Mental Health (USA); n: number; p.: page; PAAS: Perceptions of Adult Attachment Scale; PDI: Parent Development Interview; PBCL: Preschool Behaviour Checklist; PhD: Doctor of Philosophy; PIP: parent-infant psychotherapy; PIR-GAS: Parent-Infant Relationship Global Assessment Scale; PPI: psychoeducational parenting intervention; PPT: psychodynamic psychotherapy; PSI: Parenting Stress Index; PSOC: Parenting Sense of Competence; PSS: Perceived Stress Scale; RCT: randomised controlled trial; SCL-90: Symptom Checklist - 90; SD: standard deviation; SES: socioeconomic status; SPSQ: Stress and Parenting Stress Questionnaire; SSB: Social Support Behaviours; SSP: Strange Situation Procedure; TAU: treatment-as-usual; TPP: Toddler-Parent Psychotherapy; WPPSI-R: Wechsler Preschool and Primary Scales of Intelligence; WWW: Infant-Led Parent-Infant Psychotherapy ‘Watch Wait and Wonder’.
11.2 CHARACTERISTICS OF EXCLUDED STUDIES
Footnotes
IPT: interpersonal therapy; PIP: parent-infant psychotherapy; RCT: randomised controlled trial; quasi-RCT: quasi-randomised controlled trial; SD: standard deviation
11.3 CHARACTERISTICS OF ONGOING STUDIES
Footnotes
BHC: Building Healthy Children; DD-NOS: depressive disorder not otherwise specified; EPDS: Edinburgh Postnatal Depression Scale; ERA: Early Relational Assessment; HRSD: Hamilton Rating Scale for Depression; IPT: interpersonal psychotherapy; RCT: randomised control trial; MDD: major depressive disorder; PPD: postpartum depression; SCID-IV: Structured Clinical Interview for DSM Disorders.
12 Summary of findings tables
12.1 1 PARENT-INFANT PSYCHOTHERAPY VERSUS CONTROL FOR IMPROVING PARENTAL AND INFANT MENTAL HEALTH: PARENT MENTAL HEALTH AND INFANT ATTACHMENT
Footnotes
1 Risk of bias due to study designs, randomisation and allocation concealment; and not possible to blind participants or personnel.
2 Relatively few participants and wide confidence intervals.
3 Moderate to high levels of statistical heterogeneity.
12.2 2 PARENT-INFANT PSYCHOTHERAPY VERSUS OTHER (PSYCHOEDUCATIONAL) PARENTING INTERVENTION: PARENT MENTAL HEALTH AND INFANT ATTACHMENT
Footnotes
1 Risk of bias due to study designs, including randomisation and allocation concealment; and not possible to blind participants or personnel.
2 Relatively few participants and wide confidence intervals.
3 Moderate to high levels of statistical heterogeneity.
Footnotes
*
For the comparison of PIP versus control, we used the primary outcomes of parental depression at post-intervention and infant attachment at post-intervention and follow-up to complete Summary of findings table 1. For the comparison of PIP versus alternate intervention in Summary of findings table 2, we used the primary outcomes of parental depression at post-intervention and follow-up, and for infant attachment, we used outcomes from post-intervention and follow-up.