Optimizing parenting and child outcomes following parent–child interaction therapy – toddler: a randomized controlled trial

Abstract

Background

Parent–Child Interaction Therapy—Toddler (PCIT-T) is an attachment-informed intervention model designed to meet the specific developmental needs of toddlers aged 12–24 months presenting with challenging behaviors.

Methods

This study used a randomized controlled design to evaluate outcomes of PCIT-T for children aged 14–24 months with disruptive behaviors. Ninety toddlers with parent-reported disruptive behavior were randomly allocated to PCIT-T (intervention), an active control condition (Circle of Security– Parenting™; COS-P), or a non-treatment control condition (wait-list; WL). Outcomes were assessed at baseline (Time 1), post treatment/post waitlist (Time 2) and 4-month follow-up (Time 3).

Results

At follow-up, the PCIT-T group displayed the highest levels of parenting sensitivity and positive parental verbalizations, and the lowest levels of negative child-directed verbalizations and non-attuned mind-minded statements. Of the three groups, the PCIT-T group showed the greatest degree of change on these variables, followed by the COS-P group and then the non-treated controls. The PCIT-T group were also the only group to show significant within-group improvements in sensitivity, self-reported parental reflectiveness, empathy and emotional understanding, parent-reported child social competence, child internalizing problems, and general behavior issues. Significant reductions in parental stress, child externalizing behaviors and parenting behaviors were seen for both the PCIT-T and COS-P groups.

Conclusions

Delivered in the early intervention period of toddlerhood, Parent–Child Interaction Therapy—Toddler has the potential to bring about significant changes for children presenting with early onset behavioral issues.

Trial registration

Australian New Zealand Clinical Trials Registry (ANZCTR), 12,618,001,554,257. Registered 24 September 2018 – retrospectively registered, https://anzctr.org.au/Trial/Registration/TrialReview.aspx?ACTRN=12618001554257.

Background

Toddlerhood, defined as the period from around 12 to 30 months of age, is a unique developmental period characterized by increased physical mobility, language, autonomy, and individuation [12, 21]. These key developmental processes take place at a time when the capacities for emotional and behavioral self-regulation are still developing, and when there is ongoing need for physical and emotional nurture and protection from key attachment figures [25, 56, 82]. As a normative response to the challenges of this developmental phase, many toddlers will, at times, display heightened levels of emotionality, sometimes manifested as ‘tantrums’ and other externalizing behaviors (e.g., aggression and non-compliance) [21, 35, 42, 70]. While such difficulties usually resolve naturally across the transition into the pre-school years [86], for a small proportion of children, difficulties persist and represent the start of a pathway to mental health challenges [4, 11, 15, 16, 24, 44, 71].

While various genetic, social, and temperamental pathways to behavioral difficulties in early childhood have been proposed [33, 49, 57], one of the strongest known predictors of persistent behavioral issues in early childhood is the quality of the early caregiving environment [76]. In particular, evidence suggests that receiving sensitive caregiving from a primary attachment figure, defined as the capacity to notice and correctly interpret the child’s signals and to respond contingently, is key [22, 26]. According to attachment theory, sensitive caregiving in the earliest months and years of life leads to the establishment of a ‘secure’ caregiver-child attachment relationship in which the child develops an internal working model or ‘representation’ of the caregiver as a reliable ‘secure base’ from which they can confidently explore the environment and new relationships, and a ‘safe haven’ to which they can return for comfort when required [8]. It is in the context of this relationship that the capacities for behavioral and emotional control develop [81]. In contrast, children who receive insensitive, unresponsive, or inconsistent caregiving in the early years are thought to develop internal working models of the caregiver as unreliable and/or unavailable. As a result, these children develop compensatory strategies to manage distress while maintaining the relationship, referred to as insecure avoidant or ambivalent attachment. In extreme cases, such as when children have received abusive, extremely insensitive or hostile caregiving, or when they have received neglectful or frightened/frightening caregiving, there can be a breakdown or incoherence in organisational strategy within the relationship (referred to as ‘disorganised’ attachment; [37, 59]). The theoretical assertions of attachment theory are backed by empirical evidence showing that insecure and disorganized attachment patterns in infants and young toddlers can be reliably observed experimentally with the Strange Situations Procedure (SSP; [3]), and that attachment security can be assessed using in-home assessment protocols including the Attachment Q-set (AQS; [91]). There is also a large body of evidence documenting longitudinal links between insecure and disorganised attachment in infancy/toddlerhood and a range of social-emotional, mental health and functional outcomes in middle childhood and adolescence [32, 39, 58].

Providing early intervention is known to be more effective and cost-efficient than waiting until adulthood when problems are entrenched [28, 84]. Specifically, it has been suggested that the first two years of life offer the best opportunities for prevention and effective early intervention because they are characterized by high levels of neuroplasticity and sensitivity to environmental influences [63]. Over the past two decades, a number of attachment-based parenting intervention approaches targeting high-risk children and caregivers have been developed (Steele & Steele, [83]). Examples include Video-feedback Intervention to promote Positive Parenting and Sensitive Discipline (VIPP-SD; [47, 88]), the Circle of Security Intensive (COS-I) and Circle of Security Parenting (COS-P) interventions [23, 43, 94], and ABC Biobehavioral Feedback [29, 40]. These interventions target parenting sensitivity or factors known to contribute to sensitivity such as parental reflective functioning and representations (e.g., hostility, caregiving helplessness). While many of these interventions have been shown to be effective, most of the available randomized controlled trials (RCTs) have utilized waitlist-controlled designs (rather than comparisons with active treatment conditions), and with the exception of VIPP-SD, none have been developed to specifically target toddler-aged children.

Parent–Child Interaction Therapy – Toddler (PCIT-T; [36, 52]) is a new attachment-based parenting intervention, designed specifically for parents and toddlers. PCIT-T is an adapted version of the standard PCIT intervention, which was designed for pre-school and early primary school aged children with conduct problems [30, 67]. Like standard PCIT, PCIT-T draws from social learning theory and play therapy principles. Given the salience of the parent–child attachment relationship in toddler development, enhancing parenting sensitivity to the child’s emotional needs was added as an explicit intervention target for PCIT-T. At its core, PCIT-T assumes that: disruptive behaviors in toddlers typically signal emotional dysregulation; capacities for emotional and behavior regulation emerge and are consolidated in the context of a secure parent–child attachment relationship; and toddlers have the capacity to learn new skills (e.g., cooperating with parental instructions). The primary aims of PCIT-T, therefore, are to: 1) increase parents’ use of positive caregiving skills (e.g., increasing praise and reducing negative language directed at the child), 2) increase caregivers’ capacities for reflectiveness/mentalization and caregiving sensitivity, 3) enhance caregivers’ capacities for emotional self-regulation, and 4) increase caregivers’ understanding and skills relating/speaking to their toddlers in a way that is developmentally appropriate and helpful for promoting learning and listening. It is expected that by increasing these parenting capacities, there will be flow-on effects for the child such as increased attachment security and social-emotional functioning (e.g., emotion regulation, decreased externalizing and internalizing behaviors).

Like standard PCIT, PCIT-T uses in-clinic parent–child play sessions with direct live coaching from the therapist, usually from behind a one-way mirror using a Bluetooth in-ear microphone. The program comprises two sequential treatment phases, ‘Child-Directed Interaction-Toddler’ (CDI-T) and ‘Parent-Directed Interaction-Toddler’ (PDI-T). Both phases commence with a parent-only ‘teaching’ session in which the rationale and skills to be used in that phase are discussed followed by a series of ‘coaching’ sessions in which the therapist provides coaching and support during live parent–child play sessions. All sessions also seek to enhance parental reflectiveness; this is done through structured questions/discussion at the start of each session, and in the coaching statements made by the therapist.

In CDI-T, the focus is on fostering a positive parent-toddler attachment relationship and parents are coached to engage in child led-play and to use positive parenting skills including Labeled Praise, Reflections of child verbalizations, Imitating child play, Behavioral Descriptions, and Enjoyment (referred to as the ‘P-R-I-D-E’ skills), while also reducing use of questions, commands and criticisms. In the CDI-T phase, parents are also encouraged to notice and label emotions in their toddler, and to support their toddler’s emotional regulation through application of a series of practical techniques referred to as the ‘C-A-R-E-S’ skills. This includes increasing physical proximity (‘Coming in’), supporting the child’s problem solving around emotional triggers (‘Assisting’), providing verbal reassurance that the child is not alone with the emotion (‘Reassuring’), verbalizing and validating the child’s emotional experience and expressions (‘Emotionally validating’), and providing the child comfort, verbally and physically (‘Soothing’). Throughout the CDI-T phase, parents are also supported to develop skills in identifying/understanding the underlying reasons for their child’s disruptive behaviors. In situations when disruptive behaviors are thought to be driven by emotional dysregulation, the CARES techniques are applied; when behavior appears to be driven by a desire for attention (e.g., throwing toys) or unhelpful exploration (e.g., tearing pages in a book, putting puzzle pieces in mouth, reaching for television remote controls), parents are coached to under-react and re-direct the child towards positive play. In the case of aggressive toddler behavior (e.g., hitting or biting the parent, pulling the parent’s hair), parents are coached to use a brief, developmentally appropriate “no hurting” limit-setting sequence. The CDI-T phase also focuses on encouraging parents to foster their own emotion regulation through engagement with the ‘adult CARES’ model—a series of techniques that parents can utilize at the start or during parent–child play sessions, or at other times of the day, to manage their emotions [Check your cognitions, Assist yourself to manage the feelings (e.g., deep breathing, progressive muscle relaxation), Reassure yourself (cognitive challenging, positive self-talk), Label and validate the Emotions you are experiencing, Self Soothe (e.g., self-care)].

The PDI-T phase aims to promote toddler listening skills. In this phase, parents are taught how to give instructions to toddlers that are developmentally appropriate and effective, and they are coached to use a guided compliance teaching sequence. Designed to be fun and game-like, the guided compliance sequence is applied in parent–child play sessions when the child is feeling calm and relaxed. It includes a series of steps designed to support the toddler’s learning: ‘Tell’, ‘Show’, ‘Try Again’, ‘Guide’.

Preliminary evidence to support the efficacy of PCIT-T has been positive. In an initial study, Kohlhoff and Morgan [51] retrospectively reviewed outcomes of 29 clinic-referred parent-toddler dyads (Child Mage 19.8 months, SD 2.60) treated with an early version of the PCIT-T (CDI-T phase only). Results showed statistically and clinically significant improvements in parental verbalizations (e.g., increased praise, decreased criticism) and child behavior (e.g., decreased externalizing behaviors, increased compliance) following the intervention, reductions in parental depressive symptom severity, and high levels of consumer satisfaction. In a subsequent study, 66 clinic-referred mother-toddler dyads (child Mage 19.01 months, SD 2.36) participated in a randomized controlled trial (RCT) study comparing outcomes of PCIT-T (CDI-T phase only) to a waitlist controlled condition [54]. Results showed PCIT-T to be superior to waitlist in a range of domains including increases in parental verbalizations and emotional availability, and reductions in the severity of child externalizing behaviors. When the sample was followed up 4-months after the completion of PCIT-T treatment, the gains in these key parent and child domains were maintained [50, 53]. While the sample size at follow-up was small for categorical analyses (n = 16), there was a pattern (non-significant) of positive change from disorganized to organized infant attachment.

Study Aims

While preliminary evidence points to the effectiveness of PCIT-T, there is limited understanding of the intervention’s capacity to bring changes in key intervention target areas including emotion regulation and parent–child attachment quality. In addition, apart from one published case study [19], there have been no evaluations of the full PCIT-T program (i.e., CDI-T + PDI-T) [36]. The current study thus aimed to examine whether the PCIT-T intervention leads to positive changes in parent and child outcomes. To test these aims, three study conditions were used: PCIT-T; a waitlist control group who were not treated; and an ‘active’ control group who received Circle of Security – Parenting™ (COS-P), a group-based attachment-informed preventative parenting education program. COS-P was chosen as the active control condition because it is a well-respected psychoeducational program for parents, including those with children in the toddler age range, and is a parenting treatment of choice in many government and community-based settings across Australia. Furthermore, like PCIT-T, COS-P addresses the parent–child attachment relationship, it has a similar ‘dose’ to PCIT-T (16 h across 8 weeks), and it was being provided on a regular basis at the clinical site at which the study was conducted. It was acknowledged from the outset that COS-P was not a perfect comparator, particularly due to differences in intended client group and treatment targets (PCIT-T is an intensive therapeutic intervention model for children with behavioral issues whereas COS-P is an education-based program, and it does not specifically target child behavior issues) and a different mode of delivery (PCIT-T is delivered as an individual intervention and COS-P is delivered to groups of 6–8 participants). However, given that the primary objective of the study was to test outcomes of PCIT-T through comparison with both non-treated and ‘active’ control condition (rather than to test outcomes of COS-P), COS-P was considered an appropriate inclusion.

Overall, we expected that compared to the active control condition (COS-P) and the non-active control condition (non-treated waitlist), PCIT-T would be associated with the greatest treatment improvements. The rationale for this prediction related to the fact that: 1) PCIT-T is the more intensive treatment model (PCIT-T is delivered on an individual basis; COS-P is a group program); 2) PCIT-T is targeted at a narrower age range and is therefore more developmentally tailored (PCIT-T is for children aged 12–24 months; COS-P does not specify a child age range); 3) PCIT-T is designed specifically for children with clinically significant disruptive behaviors and COS-P is not; and 4) compared to COS-P, PCIT-T has an explicit behavioral component, which aligns with meta-analytic findings indicating that interventions that are brief, have a behavioral component, and which enhance parental sensitivity are most likely to bring about lasting improvements in children’s attachment security and thus capacity for behavioral and emotional regulation [6].

The primary study hypothesis was that parent–child dyads who received PCIT-T would show positive outcomes across both parent and child domains (listed below), both immediately following treatment (‘Time 2’) and at 4-month follow-up (‘Time 3’), and that these outcomes would be superior to those who received COS-P (active control) or a non-treated waitlist control condition (waitlist control). The secondary hypothesis was that on all the domains examined, there would be a gradient effect from baseline (‘Time 1’) to post-treatment/waitlist (‘Time 2’) whereby parent–child dyads who received PCIT-T would show the greatest improvement, and waitlist controls would show the least improvement. We expected that for PCIT-T and COS-P, this gradient effect would be maintained 4-months post-treatment, with parent–child dyads in the PCIT-T intervention group showing greater improvements from Time 1 to Time 3 on all variables compared to those who received COS-P. The key outcome domains (and expected direction of effects) were as follows:

• 1. Improved caregiving capacity indexed through a) increased i) positive child-directed verbalizations, ii) parenting sensitivity, iii) parental mentalization, iv) empathy towards the child and understanding of the child’s emotions, and v) parental emotion regulation; and b) decreased i) negative child-directed verbalizations, ii) caregiving helplessness, iii) hostility, and iv) parental stress.

• 2. Improved child outcomes including a) increased social-emotional functioning (initiative, attachment/relationship functioning, self-regulation, social competence, emotion regulation maturity) and attachment security; and b) decreased problem behaviors (externalizing, internalizing, and general problem behaviors).

Methods

Design

A RCT design was utilised to compare outcomes for participants who received PCIT-T (‘PCIT-T intervention’), Circle of Security-Parenting (COS-P; ‘active control’) and those receiving no treatment (‘waitlist control’). The study was registered with the Australian New Zealand Clinical Trials Registry (ANZCTR registration number: 12618001554257) and was delivered in accord with a published protocol [50, 53].

Participants and Procedure

Participants

Participants were 90 toddlers aged 14–24 months and their mothers referred by a health professional (e.g., General Practitioner, Paediatrician, Early Childhood Nurse, Psychologist) to a specialized outpatient child treatment clinic located in South Western Sydney, Australia for treatment of behavioral difficulties. Parent-toddler dyads were eligible for the study if: 1) the child was aged 14–24 months; 2) the parent responded positively to one or both of two screening questions, ‘do you have concerns about your child’s behavior?’ and/or ‘do you have difficulties managing your child’s behavior?’; 3) there was no evidence of severe parental depression with suicidality or other serious mental health conditions causing significant impairment in cognition or behaviors (e.g., psychosis); and 4) the parent was proficient in English. Participating parents provided written, informed consent during the first face-to-face assessment session at the clinic.

One hundred percent of the participating parents were female. The mean parent age was 33.07 years (SD = 5.60, range = 19-45) and the mean child age was 19.48 months (SD = 3.15, range = 14–25 (despite the 12–24 month age criteria, due to an administrative error, one of the children in the study was aged 25 months at the time of recruitment). Close to half of the children in the sample were female and 26.9% of the mothers were single. The sample was ethnically diverse with 52.9% identifying as non-Caucasian, most commonly Asian (17.4%), Middle Eastern (11.6%) and non-Caucasian European (8.6%); almost half (48.7%) spoke a language other than English at home. Almost half (48%) of the sample had a university education; mothers worked in a range of professions including managers and professional occupations (32.4%), community/personal service workers (14.9%), clerical/administrative workers (6.8%), technical/trade workers (4.1%), and students (5.4%). At the time of the study, almost a third were not working in paid employment due to full-time caretaking responsibilities in the home. Demographic characteristics of the sample are shown in Table 1.

Table 1 Demographic characteristics of the sample, and results of independent samples t-tests and Chi-square tests comparing the PCIT-T, COS-P and waitlist groups

Recruitment and Randomisation

Participants who met the study inclusion criteria were randomly assigned to receive one of the three conditions using restricted block randomization (block size n = 6). Within each block, 2 participants were allocated to the PCIT-T condition, 2 were allocated to the COS-P condition and 2 were allocated to the Waitlist condition (order randomized within each block). A research assistant not involved with the trial generated the random allocation sequence. Randomization took place at the start of the first face-to-face assessment session, with allocations pulled out of pre-prepared sealed envelopes; the research team and family were masked to condition allocation until after completion of the entire baseline (‘Time 1’) assessment. Recruitment continued until the required sample size had been recruited.

T1, T2 and T3 assessments

Participants completed assessments at baseline (Time 1; T1), immediately post waitlist/post intervention (Time 2; T2) and at 4-month follow-up (Time 3; T3; PCIT-T and COS-P groups only). For a detailed description of the T1, T2 and T3 assessments, see Kohlhoff, Cibralic, et al. (2020a, b).

Participant flow through the study is shown in Fig. 1. Of the N = 90 parent child dyads who were recruited into the study, 63 (70.0%) completed the T1 and T2 assessments (PCIT-T group: n = 24; COS-P group: n = 13; WL: n = 26), and 27 (30%) completed the T1 assessment and dropped out prior to completing the T2 assessment (PCIT-T group: n = 6; COS-P group: n = 17; WL: n = 4). Participants allocated to the WL condition ended their participation in the study following the T2 assessment. Of the 60 participants who were allocated to the PCIT-T or COS-P groups, 24 (40%) completed the T3 assessment (PCIT-T group: n = 19; COS-P: n = 5). There were no significant demographic differences between participants who did, and participants who did not, complete the T2 assessment in terms of child age, mother age, number living in the household, child sex, or family income (ps > 0.05). Participants who dropped out prior to T2 were, however, less likely to be university educated, χ² (1) = 5.07, p = 0.024. Of the PCIT-T and COS-P participants, those who dropped out between T2 and T3 were more likely to have younger mother age compared to those who completed the T3 assessment, t (52) = 2.20, p = 0.032 but no other significant differences were identified.

Fig. 1

Participant flow through the study

Sample size

An a-priori power calculation based on Child Behavior Checklist (CBCL/1.5–5; [2]) externalizing subscale scores obtained in a similar study at the same site [54] suggested that n = 30 in each group would give 80% power to detect a mean difference between groups of at least 8.0, with SD of 9. For more details about the sample size calculation, see [50, 53].

Interventions

Parent–Child Interaction Therapy – Toddler (PCIT-T)

PCIT-T was delivered according to the protocol outlined by Girard and colleagues [36]. This included in-clinic live coaching from behind a one-way mirror using a ‘bug-in-the-ear’ during clinic-based parent–child play sessions (CDI-T and PDI-T phases). Families attended a CDI-T teaching session followed by approximately 6–8 CDI-T coaching sessions and then a PDI-T teaching session followed by approximately 2–4 PDI-T coaching sessions (breakdown of CDI-T vs PDI-T and total number of sessions were determined based on individual need). In total, of the 30 participants allocated to receive the PCIT-T intervention, 4 did not commence the treatment and of the 26 participants who commenced PCIT-T, 24 (92.31%) completed and 2 (7.69%) did not complete the treatment (one attended 4 sessions in total, the other attended 2 sessions in total). While the aim was for parents in this condition to receive a total of 16 h of therapy (two 45 min sessions per week, for 8 weeks), for the 24 parents who completed treatment and the T2 assessment, the mean total hours of therapy was 13.95 h (SD = 1.92, range = 9.08–16.42) over an average of 11.42 weeks (SD = 2.60, range = 7–18). The PCIT-T treatment was provided by three therapists, all of whom were co-developers of the PCIT-T program. Therapist 1, a clinical psychologist and PCIT International within agency trainer, treated 19% of the cases; therapist 2, a psychologist and PCIT International within agency trainer, treated 69% of the cases, therapist 3, a clinical nurse consultant and PCIT International Regional Trainer, treated 12% of the cases). Treatment fidelity was maximized by the use of: 1) post-session fidelity checklists completed by PCIT-T clinicians at the end of every treatment session [36]; and 2) clinical reflective supervision sessions provided to each of the PCIT-T clinicians on a fortnightly basis throughout the course of the research trial by one of the PCIT-T program developers.

Circle of Security – Parenting.™ (COS-P)

COS-P was delivered according to the protocol outlined by Cooper, Hoffman, and Powell [23]. COS-P is a manualized, facilitator-led, eight-session parent-education group program that uses archived clinical video footage of problematic parent–child interaction and healthy alternatives to illustrate attachment patterns and parenting styles and promote group discussion. COS-P seeks to increase caregiver sensitivity and responsiveness to child cues, empathy for the child by supporting parental reflective functioning, recognition and understanding of child attachment cues, and awareness of the impact of the caregiver’s own attachment history on caregiving patterns. In total, of the 30 participants allocated to receive the COS-P intervention, 4 were prevented from taking part in COS-P due to the Covid-19 pandemic (groups ceased to be run) and 11 did not commence the treatment because they could no longer attend the COS-P sessions on the day they were scheduled (e.g., to due to changes in the parent’s work schedule) or because they no longer wished to engage in the COS-P intervention. Of the remaining 15 participants allocated to receive the COS-P intervention, 13 completed the T2 assessment and attendance data was missing for 2 participants. While the aim was for parents in this condition to receive a total of 16 h of therapy (one 90 min group sessions per week, for 8 weeks), of the 11 participants who attended the COS-P groups, completed the T2 assessment, and for whom attendance data was available, the mean number of sessions attended was 6 (SD = 1.27, range = 4–8) and the mean total amount of group time was 11.72 h (SD = 2.86, range = 8—16). The COS-P groups were provided on-site at the clinic, co-facilitated by two clinicians who had been trained and accredited in the intervention, at least one of whom had experience in running previous COS-P groups and one who had no training or experience delivering PCIT-T. Across the trial, 7 COS-P groups were run and there were 7 different COS-P facilitators (professional backgrounds: Clinical Psychology, n = 4; Psychiatry, n = 1; Nursing, n = 2). COS-P facilitators were chosen based on clinical availability at the time of the trial; all were clinicians employed and working in perinatal, infant and early childhood mental health at the clinical service. To ensure that the COS-P groups had sufficient numbers of participants, they were attended by a mix of trial participants and non-trial participants with children in the same age range (who were referred from within the same clinic). Childcare during the groups was provided on-site. Treatment fidelity was maximized using: 1) fidelity checklists, completed individually by each of the group facilitators at the end of each session; and 2) 1-h clinical reflective supervision sessions, provided to the group facilitators twice during the course of each group program by an experienced COS-P facilitator and clinical supervisor.

Waitlist Controls

Participants allocated to the waitlist control condition completed the T1 assessment and then no treatment for 8-weeks. They then completed the T2 assessment, representing the completion of their participation in the study. Participants allocated to the waitlist condition were offered the opportunity to receive PCIT-T or COS-P (their choice) at the completion of the T2 assessment (not as part of the trial).

Measures

Parenting Outcomes

Positive and negative child-directed verbalizations

Positive and negative child-directed verbalizations were coded from 20-min parent–child interaction sessions using the Dyadic Parent–Child Interaction Coding System, Fourth Edition (DPICS-IV; [31]). Coders transcribed the video-recorded sessions verbatim, creating moment-by-moment transcriptions, and then coded parent and child verbalizations (using the transcription and video) according to the DPICS-IV coding scheme. A ‘positive parenting’ score (total number of labelled praises, i.e., specific praise of a child's behaviors, products (e.g., creation), or attributes) and a ‘negative parenting’ score (total number of negative statements directed towards the child) was calculated based on the number of times the verbalizations were observed. To create standardized scores, frequency counts were transformed into ratios as follows: 1) total number of labelled praises divided by total number of verbalizations, and 2) total number of negative statements divided by total number of verbalizations. The DPICS-IV coding was conducted by graduate level research assistants trained by a PCIT International Global Trainer. All coders were masked to participant group allocation and assessment time point. Of the 178 coded videos, 25% (n = 44) were double coded (n = 24 at T1, n = 9 at T2, and n = 11 at T3). Interrater reliability was assessed using the Cohen’s Kappa statistic, which accounts for chance agreement. The agreements between coders for the total number of codes, labelled praise and negative talk codes were κ = 0.894 (95% CI, 0.887 to 0.901), κ = 0.961 (95% CI, 0.946 to 0.976), and κ = 0.846 (95% CI, 0.831 to 0.860), respectively.

Parenting sensitivity

Parenting sensitivity was coded from the 20-min parent–child interaction sessions using the National Institute of Child Health and Human Development Study of Early Child Care and Youth Development: Sensitivity Scales (NICHD-SECCYD; [66]), Composite parental sensitivity scores were calculated by summing scores on the six NICHD scales [sensitivity, positive regard for the child, negative regard for the child (reversed), intrusiveness (reversed), respect for autonomy, and hostility (reversed)], each scored on a 4-point scale with higher scores indicating that the variable was more characteristic of the parent. For the current study, the NICHD-SECCYD coding was conducted by three coders, each masked to participant group allocation and assessment time point. Of the 159 coded videos (n = 84 at T1, n = 54 at T2, n = 21 at T3), 23% (n = 36) were triple coded (intra-class coefficient, ICC = 0.95, 95% CI, 0.91–0.97).

Parental mentalization

Parental mentalization was assessed in two ways: observed (operationalized as mind-mindedness, defined as the propensity to view the child as an ‘agent’ with independent thoughts and feelings [61]) and parent-report (reflective functioning). Mind-mindedness was coded from the 20-min parent-toddler interaction. Parent and child verbalizations during the interaction were transcribed verbatim and then divided into individual comments based on temporal (1-s gap) or semantic discontinuities. Comments were then coded according to Meins and Fernyhough’s mind-mindedness coding guidelines [61]. This involved applying one of five categories (‘Mental’, ‘Behavioral’, ‘Physical’, ‘General’, ‘Self-referential’) to every comment. Comments coded as ‘Mental’ attributes were descriptors related to the child’s mental life, intentions, interests, likes and emotions. Behavioral attributes referred to the child’s behavior (e.g., games or activities that the child does). Physical attributes were comments about the child’s appearance, position in the family or age; and general attributes include comments that did not fit into the other three categories. The self-referential code was given when the parent spoke about themselves, rather than the child. Comments coded as ‘mental’ comments were then coded as either “appropriate” (i.e., reflected a correct interpretation of the child’s internal state, linked current activity with similar events in the past or future, or served to clarify how to proceed after a lull in the interaction) or “non-attuned” (incorrect reading of the child’s current internal state; referred to a past or future event unrelated to the infant’s current activity; questions or suggestions of what the child should do when the child was already actively engaged in something else; attributions of internal states not implied by the infant’s behavior). Frequency counts were then transformed into ratios as follows: 1) total number of appropriate mental-state comments divided by total number of verbalizations, and 2) total number of non-attuned mental-state comments divided by total number of verbalizations. Mind-mindedness coding was conducted by two coders masked to participant group allocation and assessment time point. Of the 166 coded videos (n = 88 at T1, n = 61 at T2, n = 17 at T3), 23% (n = 38) were double coded (ICC = 0.99, 95% CI, 0.98-0.96).

Parent-reported mentalization was assessed using Diamond’s reflective functioning scale (DRFS; [27]) administered as part of the Composite Caregiving Questionnaire (CCQ; [60]). The DRFS comprises 18 parent-report items, each answered on a 4-point scale. Scores load onto 3 scales (Cue recognition, Mentalization of Infant, Own childhood experience) and a total score can also be computed (score range = 0–54; higher scores reflecting better parental reflective functioning). In the current study, the total DRFS score was used. The Cronbach’s alpha for the total DRFS score at T1 was 0.76.

Parental self-efficacy regarding empathy and emotional understanding

Parental self-efficacy regarding empathy and emotional understanding was assessed using the empathy and emotion subscales of the Tool to Measure Parenting Self-Efficacy (TOPSE; [48]), administered as part of the CCQ [60]. The TOPSE-empathy and the TOPSE-emotion scales each comprise 7 items answered on a five-point scale. Item scores for each scale are totalled (score range = 0–60; higher scores indicate greater empathy/emotion). In the current study, the Cronbach’s alphas for the TOPSE-empathy and the TOPSE-emotion scales at T1 were 0.81 and 0.77, respectively.

Parental emotion regulation

Parental emotion regulation was assessed using the Difficulties in Emotion Regulation Scale (DERS; [38]). The DERS comprises 36 self-report items, answered on a 5-point scale. It yields six subscale scores: 1) Nonacceptance of emotional responses, 2) Difficulties engaging in goal directed behavior, 3) Impulse control difficulties, 4) Lack of emotional awareness, 5) Limited access to emotion regulation strategies, and 6) Lack of emotional clarity, and a total score (range = 36–80, higher scores indicate greater emotion dysregulation). In the current study, the total DERS score was used. The Cronbach’s alpha for the total DERS at T1 was 0.85.

Caregiving helplessness

Caregiving helplessness was assessed using the ‘mother helpless’ subscale of the Caregiving Helplessness Questionnaire (CHQ; [79]), administered as part of the CCQ [60]. The CHQ-mother helpless scale comprises 7 items, each answered on a five-point scale. Item scores are totalled (score range = 7–35; higher scores indicate greater caregiving helplessness). In the current study, the Cronbach’s alpha for the CHQ mother helpless subscale at T1 was 0.84.

Parenting hostility

Parenting hostility was assessed using the Longitudinal Study of Australian Children (LSAC) Parenting Hostility scale [73], administered as part of the CCQ [60]. The LSAC Parenting hostility scale comprises 5 items, each answered on a five-point scale; item scores are totalled (score range = 0–15; higher scores indicate greater hostility). In the current study, the Cronbach’s alpha for the LSAC-Parental hostility scale at T1 was 0.88.

Child abuse potential

Child abuse potential was assessed using the Brief Child Abuse Potential Inventory (BCAP; [69]), an abbreviated version of the widely-used 160-item Child Abuse Potential Inventory [62]. The BCAP comprises 24 self-report items, each rated as ‘agree’ or ‘disagree’. Scores are totalled to yield a total Risk and a Lie score. In the current study, the total risk score was used, with the Lie score used as a covariate. The Cronbach’s alphas for the BCAP total risk score at T1 was 0.89.

Parenting stress

Parenting stress was assessed using the Parenting Stress Index–Short Form (PSI-SF; [1]). The PSI-SF comprises 36 self-report items each answered on a five-point scale. Items load onto three subscales: ‘Parental Distress’, ‘Parent–child Dysfunctional Interaction’, and ‘Difficult Child’, and a total score (score range = 36–180; higher scores indicate greater parenting stress). In the current study, the Cronbach’s alphas for the PSI-SF at T1 was 0.91.

Child outcomes

Child social-emotional functioning

Child social-emotional functioning was assessed with the Devereux Early Childhood Assessment (DECA; [55, 72]), and the Brief Infant–Toddler Social Emotional Assessment (BITSEA; [9, 10]). The DECA is a strengths-based parent-report scale designed to assess emotional, social, and behavioral capacities during early childhood. It has two aged-based versions, an infant version (DECA-I; designed for ages 1–18 months; 33 items) and a toddler version (DECA-T; designed for ages 18–36 months; 36 items). The DECA-I comprises two scales (initiative, attachment/relationships) and the DECA-T comprises three subscales (initiative, attachment/relationships, self-regulation). In this study, parents completed the DECA-I or DECA-T depending on the age of their child; t-scores were calculated and used to enable inclusion of participants who completed the two different forms of the measure in the same analyses. In the current study, the Cronbach’s alphas for the DECA-I initiative and attachment/relationships subscales at T1 were both 0.93. The Cronbach’s alphas for the DECA-T initiative, attachment/relationships, and self-regulation subscales at T1 were 0.87, 0.92 and 0.68, respectively. The BITSEA is a parent-report screener for social-emotional issues in infants and toddlers. It comprises 42 items, each answered on a 3-point scale. Items load onto two subscales: ‘General problems’ (31 items; higher scores indicate greater behavioral problems) and ‘Competence’ (11 items; higher scores indicate better social competence). In this study, the Competence scale was used to assess the child’s social-emotional abilities. The Cronbach’s alpha for the BITSEA Competence scale at T1 was 0.70.

Child attachment security

Child attachment security was assessed in two ways allowing assessment of attachment dimensionally and categorically. First, to examine attachment security dimensionally, the video-taped home visit was coded using the Attachment Q-set (AQS) [90, 92], a validated coding procedure designed to assess attachment security in a naturalistic environment in children aged 12–48 months. In the AQS, various descriptors of child behavior, including attachment behavior, are sorted into nine piles from “most characteristic” to “least characteristic”. An overall security score is derived by correlating the sort for each child with a criterion sort of the prototypical behavior of a secure child. All home visits were video recorded and coded by 5 trained and reliable AQS coders who were masked to intervention assignment and time point; 29.7% (n = 41) of the home visits were double coded (ICC = 0.621, 95% CI, 0.4–0.86). Second, to examine attachment security categorically, including disorganization, child behavior observed during the SSP was coded according to the coding systems of Ainsworth [3] and Main and Solomon [59]. The SSP has demonstrated validity [7, 93] and is widely considered to be the gold standard measure of infant attachment. The SSP comprises eight three-minute episodes involving separations and reunions between the mother, the child and a friendly stranger. Infants are subsequently assigned to one of three ‘organized’ classifications based on their observed behavior, namely secure (B), anxious-avoidant (A), or anxious-resistant (C), following the Ainsworth coding system [3]. In addition, a primary ‘disorganized’ (‘D’) classification can also be applied with the ABC classifications as secondary, using procedures described by Main and Solomon [59]. All SSP assessments were video recorded and coded by experienced SSP coders and trainers from the University of Minnesota; 28% (n = 30) of the SSP tapes were double coded, inter-rater agreement was 86% (k = 0.78, p < 0.001).

Toddler emotion-regulation maturity

Toddler emotion-regulation maturity was rated from videotapes of the 5-min ‘toddler frustration task’ using the coding protocol developed by Johnson [46]. First, displays of distress during the five minutes were coded in 30 × 10 s intervals on a four-point scale ranging from 0 (no distress) to 4 (extreme distress, e.g., screaming, gasping for air, having a tantrum, crying persistently with consistent or increasing intensity, hitting the box or mother while crying). If a child demonstrated both mild and moderate distress in one interval, only the highest level of distress (e.g., moderate) was coded. Overall distress scores were calculated by adding the scores, with higher scores indicating greater distress. Next, maturity of ER strategies was coded for the intervals in which participants did not display distress (i.e., distress score of zero). Specifically, behaviors observed during these time intervals were coded on one of three scales reflecting least mature (attentional deployment) to most mature (cognitive change) regulatory strategies. A code of Attentional Deployment was given for behaviors such as diverting attention away from the stressful situation without trying to change the situation in any way (e.g., looking at the door). Situation Modification was coded for active behaviors that changed the situation to decrease/increase the likelihood of negative/positive reactions (e.g., walking away from the box, playing with something else in the room). Cognitive Change was coded for behaviors that indicated that the child had changed the way they perceived the task or reassessed their capacity to manage the demands of the task (e.g., trying different strategies to open the box such as flipping the box). Scores on the Attentional Deployment, Situation Modification and Cognitive Change scales for each of the 10 s intervals were added, then weighted to reflect increasing maturity, and combined to form a composite ER maturity score (ER Maturity score = 1 × Attentional Deployment + 2 × Situation Modification + 3 × Cognitive Change) with higher scores indicating greater maturity. Coding was conducted by two coders; 5% of the tapes were initially coded by the two coders together, with codes allocated via consensus. A further 30% of the tapes were independently coded by the two coders to determine coding reliability. The ICCs for toddler distress, attentional deployment, situation modification, and cognitive change were 0.93, 0.80, 0.88, and 0.98, respectively.

Problematic child behaviors

Problematic child behaviors were assessed with two measures: the Child Behavior Checklist for ages 1.5–5 years (CBCL/1.5–5; [2]), and the Brief Infant–Toddler Social Emotional Assessment (BITSEA; [9, 10]). The CBCL/1.5–5 is a validated parent report scale designed to measure behavioral, emotional and social functioning in children aged 1.5–5 years, and which has been used reliably with children as young as 12 months [54, 89]. The CBCL/1.5– 5 comprises 99 items, each rated in terms of the frequency with which the child displays given problem behaviors on a scale of 0–2 (higher scores indicating presence of the behavior). CBCL scores can be summed to yield externalizing and internalizing sub-scale scores. In the current study, the cronbach’s alphas for the externalizing and internalizing subscales were 0.91 and 0.87, respectively. The ‘General Problems’ scale of the BITSEA (described above) was used in the current study as a general measure of problematic child behavior. The Cronbach’s alpha for the BITSEA general problems scale at T1 was 0.83.

For a summary of outcome domains and measures, see Supplementary Table 1.

Statistical analysis

Demographic variables were examined with basic descriptive statistics. Frequencies and percentages for categorical variables and means and standard deviations (for normally-distributed variables) or medians and first and third quartiles (for skewed variables) were calculated for quantitative variables.

Continuous outcome variables were then analyzed across the three time points (T1, T2, T3) using a linear mixed models with random effect for individual and fixed effects of group, time and their interaction. For variables that were calculated as percentage scores (DPICS-positive verbalizations, DPICS-negative verbalizations, DPICS-compliance, MM-attuned, MM-non-attuned), generalized linear mixed models with binomial distribution and logit link were conducted with the total number of occurrences and the total number of opportunities defining the proportion (e.g., the proportion of all verbalizations [total number of opportunities] that were positive statements [total number of occurrences] directed towards the child). Planned comparisons addressing hypotheses 1 and 2 were specified, with family-wise adjustments to the raw p-values to account for multiple comparisons using Holm's (1979) stepdown Bonferroni procedure. Intention-to-treat analyses (ITT; [41]) were used, with participants included in the analyses within their randomly assigned treatment condition regardless of the amount of treatment received and group mean values estimated based on all observed data of those in the ITT sample.

To test the first hypothesis (i.e., that parent–child dyads who received PCIT-T would show positive outcomes at T2 and T3, and that outcomes for the PCIT-T group would be superior to those of the COS-P and the WL groups), planned comparisons were specified to examine: 1) within-group differences from T1 to T2 for the PCIT-T, COS-P and WL groups, and from T1 to T3 for the PCIT-T and COS-P groups, and 2) between-group differences between the PCIT-T, COS-P and WL groups at T1 and T2, and between the PCIT-T and COS-P groups at T3. To examine changes in categorical variables (SSP classifications), generalized linear mixed models with binomial distribution and logit link were conducted with insecure attachment (0/1) and disorganized attachment (0/1) as dependent variables. The clinical significance of within-group changes on continuous study variables were assessed using Cohen’s d with effect sizes evaluated using Cohen’s [20] guidelines,for categorical variables, the significance of within-group changes were assessed through examination of Odds Ratios (ORs). To test the second hypothesis (i.e., that there would be a gradient effect from T1 to T2 with the PCIT-T group showing the greatest and the WL group showing the least improvements, and that this gradient effect would be sustained from T1 to T3 with the PCIT-T group showing more improvements than the COS-P group), group means were examined to determine whether they were ordered in terms of the hypothesized gradient. In cases where the order was satisfied, significance was tested using planned comparisons.

Reliable change index (RCI) scores were also calculated to assess whether the magnitude of individual-level change on continuous study measures from T1 to T2 exceeded the margin of measurement error [45]. RCI calculations were not conducted on the T3 data due to small sample size.

To test the robustness of the data analysis strategy (given missing data associated with participant attrition), for variables that showed a significant change from T1 to T3 (for any group), a post-hoc sensitivity analysis involving multiple imputation by chained equations was performed using the mice package in R [85, 87]. The imputation model included the following variables as predictors: group; age and sex of child; age, education and ethnicity of mother; family income; number of people in household; marital status of parents). Twenty-five datasets were imputed, using normal Bayesian linear regression for normal continuous variables, and predictive mean matching for count variables. No values were imputed for the waitlist group at T3. The unadjusted mixed model was run on each imputed dataset, and all results, including the group difference in change (primary outcome) and effect size, were aggregated using (Rubin’s Rules [75]).

Changes to protocol

The following changes were made to the published study protocol [50, 53]: 1) Addition of an extra hypothesis (referred to above as the ‘primary study hypothesis’) to allow for comparison of within and between group changes on key outcome variables; 2) Coping with Toddler’s Negative Emotions Scale (CTNES) scores not reported [80] due to an administration error; 3) Mind-mindedness was coded from the 20-min parent-toddler play sessions rather than from 5-min parent speech samples, due to better data quality; 4) Addition of the Brief Infant–Toddler Social Emotional Assessment (BITSEA; [9, 10]), to allow for assessment of general child behavior problems and social competence in addition to externalizing and internalizing problems.

Results

Tables 2 and 3 show the mean (SD) scores for primary outcomes by group, statistical significance (p values) and clinical significance (Cohen’s d effect sizes) of score changes and between-group differences at T1, T2 and T3. Table 4 shows the number and percentage of treatment-completing participants with reliable change scores on continuous study variables.

Table 2 Mean (SD) scores for caregiver outcomes by group, clinical significance of change in scores between Time 1 and Time 2, and Time 1 and Time 3 assessments by group (Cohen’s d effect sizes) and between-group differences at Time 1, Time 2 and Time 3

Table 3 Mean (SD) scores for child outcomes by group, clinical significance of change in scores between Time 1 and Time 2, and Time 1 and Time 3 assessments by group (Cohen’s d effect sizes) and between-group differences at Time 1, Time 2 and Time 3

Table 4 Number and percentage of treatment-completing participants with reliable change on parenting and child variables in the PCIT-T, COS-P and WL groups at T2

Caregiver outcomes

Positive and negative child directed verbalizations

For positive parenting verbalizations assessed using the DPICS-IV, indexed as total proportion of verbalizations that were labelled praises, significant effects were identified for group (p < 0.001), time (p < 0.001), and group x time (p < 0.001). Planned comparisons revealed no significant between-group differences at T1 but showed that the PCIT-T group used significantly more positive verbalizations than the COS-P group at T2 and T3 (T2: OR 12.8, p < 0.001; T3: OR = 7.02, p < 0.001) and the WL group at T2 (OR 14.76, p < 0.001). There were no other significant between group differences. The PCIT-T group showed significant within-group increases in the number of positive verbalizations from T1 to T2 (OR 26.55, p < 0.001) and from T2 to T3 (OR 15.56, p < 0.001). There were no within-group changes for the COS-P or WL groups. The hypotheses that there would be a gradient effect in terms of the magnitude of effects from T1 to T2 [PCIT-T > COS-P > WL] and from T1 to T3 [PCIT-T > COS-P] were supported, T1 to T2: F (2, 163) = 34.36, p < 0.001; T1 to T3: F (1, 163) = 17.4, p < 0.05.

For negative parenting verbalizations assessed using the DPICS-IV, indexed as total proportion of verbalizations that were negative statements directed towards the child, significant effects were identified for time (p < 0.001), and group x time (p < 0.001). Planned comparisons revealed no significant between-group differences at T1 but showed that the PCIT-T group showed lowered odds of negative verbalizations than the COS-P group at T2 and T3 (T2: OR = 0.39, p < 0.001; T3: OR = 0.34, p < 0.001) and the WL group at T2 (OR = 0.48, p < 0.001). There were no other significant between group differences. All three groups showed significant within-group decreases in the odds of negative verbalizations from T1 to T2 (PCIT-T: OR = 0.25, p < 0.001; COS-P: OR = 0.70, p < 0.001; WL: OR = 0.47, p < 0.001) and the PCIT-T also showed a decrease from T1 to T3 (OR = 0.23, p < 0.001). The hypotheses that there would be a gradient effect in terms of the magnitude of effects from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were supported, T1 to T2: F (2,163) = 47.37, p < 0.001; T1 to T3: F (1, 163) = 48.06, p < 0.05.

Parenting sensitivity

For parenting sensitivity assessed using the NICHD-Sensitivity scale, significant effects were identified for group (p < 0.001), time (p < 0.001) and group x time (p < 0.05). There were no significant between group-differences at T1, but the PCIT-T group had significantly higher sensitivity scores than the COS-P and WL groups at T2 (ps < 0.05), and the COS-P group at T3 (p < 0.05). There were significant within-group changes in parenting sensitivity for the PCIT-T group from T1 to T2 (d = 1.03; large effect size; p < 0.001), and from T1 to T3 (d = 0.66, medium-large effect size; p < 0.05), but there were no significant within-group changes for the COS-P or WL groups. While the pattern of differences was consistent with the gradient hypothesis in change from T1 to T2 [PCIT-T > COS-P > WL], the global test of the hypothesis did not achieve significance, F (2,142) = 2.46, p = 0.09. The hypothesis that there would be gradient effects in terms of the magnitude of effects from T1 to T3 [PCIT-T > COS-P] was supported, F (1, 142) = 4.73, p < 0.05.

Parental mentalization

For parental mentalization assessed using the ‘appropriate’ mind-mindedness code in the MM coding system, no significant effects were identified for group (p = 0.785) or group x time (p = 0.505) but there was a significant effect for time (p < 0.05). Planned comparisons showed no significant between group-differences at T1, T2 or T3 (ps > 0.05), nor were there any significant within-group changes for any group (ps > 0.05). The hypotheses that there would be gradient effects in terms of the magnitude of gains from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2, 155) = 2.54, p = 0.08; T1 to T3: F (1, 155) = 0.59, p = 0.442.

For parental mentalization assessed using the ‘non-attuned’ mind-mindedness code in the MM coding system, there were no significant effects for group (p = 0.224) or time (p = 0.733) but there was a significant effect for group x time (p < 0.005). Planned comparisons showed the PCIT-T group to use significantly fewer non-attuned comments than the WL group at T2 (p = 0.005) and no other significant between-group differences were observed. There was a significant decrease in non-attuned comments for the PCIT-T group from T1 to T2 (p < 0.05, d = −0.69; medium to large effect size) and there were no other significant within-group differences. The hypothesis that there would be a gradient effect in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] was supported, F (2, 155) = 4.53, p < 0.05. The hypothesis of a gradient effect in improvements from T1 to T3 [PCIT-T > COS-P] was not supported, F (1, 155) = 0.00, p = 0.974.

For parental reflectiveness assessed using the DRFQ, significant effects were identified for time (p < 0.05) but not for group (p = 0.099) or group x time (p = 0.174). Planned comparisons did not reveal any significant between-group differences (ps > 0.05) but there was a trend for the PCIT-T group to show higher reflectiveness than the COS-P group (p = 0.057). The PCIT-T group showed significant within-group improvements in reflectiveness from T1 to T2 (p < 0.05) and from T1 to T3 (p < 0.001). No other within-group differences were identified. The hypothesis that there would be a gradient effect in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] was not confirmed, and in fact, while the PCIT-T group showed the greatest degree of improvement, improvements shown by the COS-P and WL groups were not in the hypothesised order (i.e., PCIT-T > WL > COS-P), F (2, 153) = 2.478, p = 0.07. The hypothesis of a gradient effect in terms of improvements from T1 to T3 [PCIT-T > COS-P] was not supported, F (1, 153) = 2.22, p = 0.14.

Empathy and emotional understanding

For parenting self-efficacy regarding empathy assessed using the TOPSE-Empathy scale, no significant effects were identified for group (p = 0.576), time (p = 0.101) or group x time (p = 0.250). Planned comparisons revealed no significant between group-differences at T1, T2 or T3 (ps > 0.05). There was a significant within-group change for the PCIT-T group from T1 to T2 (p < 0.05, d = 0.60; medium-large effect size) but there were no other significant within-group changes. The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2, 155) = 0.96, p = 0.386; T1 to T3: F (1, 155) = 0.93, p = 0.338.

For parenting self-efficacy regarding emotion assessed using the TOPSE-Emotion scale, no significant effects were identified for group (p = 0.593), time (p = 0.432) or group x time (p = 0.072). Planned comparisons revealed no significant between group-differences at T1, T2 or T3 (ps > 0.05). There was a significant within-group change for the PCIT-T group from T1 to T2 (p = 0.03, d = 0.46; medium effect size) but there were no other significant within-group changes. The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and from T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2,155) = 2.52, p = 0.08; T1 to T3: F (1, 155) = 0.27, p = 0.604.

Parental emotion regulation

For parental emotional regulation difficulties assessed using the DERS, no significant effects were identified for group (p = 0.33) or time (p = 0.26), or group x time (p = 0.18). Planned comparisons revealed no significant between group-differences at T1, T2 or T3 (ps > 0.05) and no significant within-group changes for any group (ps > 0.05). In terms of the hypothesis that there would be a gradient effect in terms the degree of change from T1 to T2 [PCIT-T > COS-P > WL], while the global test was significant, F (2,155) = 3.58, p < 0.05, the hypothesis was not supported because the order of magnitude was not met [order of change = PCIT-T > WL > COS-P]. The hypothesis of a gradient effect in terms of improvements from T1 to T3 [PCIT-T > COS-P] was not supported, F (1, 155) = 0.45, p = 0.502.

Caregiving helplessness

For caregiving helplessness assessed using the CHQ-MH, a significant effect was found for time (p < 0.05), but not for group (p = 0.068) or group x time (p = 0.887). Planned comparisons revealed no significant between group-differences at T1, T2 or T3 (ps > 0.05). There were no significant within-group changes for the PCIT-T group (ps > 0.05). There was evidence of a trend for a decrease in caregiving helplessness for the COS-P group from T1 to T2 (p = 0.06; d = −0.46; medium effect size). There were no other significant within-group changes. The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2,153) = 0.16, p = 0.851; T1 to T3: F (1, 153) = 0.48, p = 0.490.

Hostile parenting

For parenting hostility assessed using the LSAC hostility scale, no significant effects were identified for group (p = 0.112) or group x time (p = 0.828) but there was a significant effect for Time (p < 0.05). There were no significant between group-differences at T1, T2 or T3 (ps > 0.05), nor were there any significant within-group changes for any group (ps > 0.05). The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2, 155) = 0.01, p = 0.987; T1 to T3: F (1, 155) = 0.56, p = 0.456.

Child abuse potential

For child abuse potential assessed using the BCAP (total score, controlling for total Lie scale score), significant effects were identified for time (p < 0.05) but not for group (p = 0.195) or group x time (p = 0.252). Planned comparisons showed no significant between group-differences at T1, T2 or T3 (ps > 0.05), or significant within-group changes for any group (ps > 0.05). The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2, 140) = 0.81, p = 0.446; T1 to T3: F (1, 140) = 2.46, p = 0.119.

Parenting stress

For parenting stress assessed using the PSI-SF, significant effects were identified for time (p = 0.001) but not for group (p = 0.253) or group x time (p = 0.124). Planned comparisons revealed no between-group differences at any time point. The PCIT-T and COS-P groups each showed significant within-group decreases in parenting stress from T1 to T2 (PCIT-T: p = 0.007, d = 0.47, medium effect size; COS-P: p = 0.025, d = 0.55, medium effect size) and from T1 to T3 (PCIT-T: p = 0.028, d = 0.43, medium effect size; COS-P: p = 0.028, d = 0.79, large effect size) but the WL group showed no within-group change. The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P], were not supported (T1 to T2: F (2,130) = 2.19, p = 0.117; T1 to T3: F (1,130) = 1.19, p = 0.277).

Child outcomes

Child social-emotional functioning

For Attachment and Relationships assessed using the DECA-attachment scale, significant effects were identified for group (p = 0.046) but not for time (p = 0.102) or group x time (p = 0.203). Planned comparisons revealed a difference between the PCIT-T and WL groups at Time 1 (p = 0.019) and no other between-group, or within-group, differences. The hypothesis that there would be a gradient effect in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] was not supported because while the global test was significant, F (2, 160) = 3.50, p < 0.05, the pattern of change was not as predicted with the PCIT-T group showing an improvement in attachment and relationships from T1 to T2 and the COS-P group showing a deterioration (i.e., order of change = PCIT-T > WL > COS-P). The hypothesis of a gradient effect in terms of improvements from T1 to T3 [PCIT-T > COS-P] was not supported, F (1, 160) = 0.18, p = 0.669.

For Initiative assessed using the DECA-Initiative scale, no significant effects were identified for group (p = 0.889), time (p = 0.565) or group x time (p = 0.098). Planned comparisons revealed no between-group differences at any time point, or within-group difference for any group (ps > 0.05). The hypotheses that there would be a gradient effects from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2, 164) = 2.79, p = 0.064; T1 to T3: F (1, 164) = 0.56, p = 0.457.

For self-regulation assessed using the DECA-self regulation scale, no significant effects were identified for group (p = 0.329), time (p = 0.776) or group x time (p = 0.416). Planned comparisons revealed no between-group differences at any time point, or within-group difference for any group (ps > 0.05). The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T2 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2, 131) = 1.17, p = 0.312; T1 to T3: F (1, 131) = 0.04, p = 0.838.

For social competence assessed using the BITSEA-competence scale, a significant effect was identified for time (p < 0.05), but not for group (p = 0.749) or group x time (p = 0.130). Planned comparisons revealed no significant between group differences at any time point (ps > 0.05). The PCIT-T group showed significant within-group improvements from T1 to T2 (p < 0.05, d = 0.48, medium effect size) and from T1 to T3 (p < 0.05, d = 0.39, medium effect size) and no other within-group differences were identified (ps > 0.05). The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2, 153) = 0.2.67, p = 0.073; T1 to T3: F (1,153) = 0.10, p = 0.750.

Toddler emotion-regulation maturity

For toddler emotional regulation maturity there were no significant effects for time (p = 0.331), group (p = 0.453) or time x group (p = 0.125). Planned comparisons revealed no significant between group differences at any time point, and no within-group change for any group. The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] was not supported, F (2, 62) = 0.933, p = 0.399; T1 to T3. For the change from T1 to T3, the global test was significant, F (1, 62) = 4.24, p = 0.044, but the changes were not in hypothesised order, i.e., COS-P showed greater degree of improvement than the PCIT-T group. The sample size at T3 was also very small (ns = 7 and 3 for the PCIT-T and COS-P groups, respectively).

Child attachment security

For child attachment security assessed using the AQS, no significant effects were identified for time (p = 0.083), group (p = 0.218), or group x time (p = 0.325). Planned comparisons revealed no significant between group differences at any time point and no significant within-group differences were identified for any group. The hypothesis that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] or T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (1,131) = 1.11, p = 0.332; T1 to T3, F (1,131) = 0.27, p = 0.61.

For child attachment security assessed using the SSP, there were no significant between-group differences in the proportions of children who were classified as insecure at either T1 or T3. While neither group showed a significant reduction from T1 to T3 in terms of the odds of being classified as insecure, in both groups there was a reduction in the proportion of children classified as insecure (PCIT-T: 62% at T1 to 40% at T3, OR = 0.39; COS-P: 34.5% at T1 to 25% at T3, OR = 0.68). For child attachment disorganization, assessed using the SSP, there were no significant between-group differences in the proportions of children who were classified as insecure at either T1 or T3. While neither group showed a significant reduction from T1 to T3 in terms of the odds of being classified as disorganized, in both groups there was a reduction in the proportion of children classified as disorganized (PCIT-T: 65.5% at T1 to 40% at T3, OR = 0.31; COS-P: 44.8% at T1 to 25% at T3, OR = 0.48). The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T3 [PCIT-T > COS-P] for SSP-Insecurity and SSP-disorganization were not supported (ps = 0.708 and 0.780, respectively).

Problematic child behavior

For externalizing behavior assessed using the CBCL externalizing scale, there was a significant effect for time (p < 0.001), but not for group (p = 0.071) or time x group (p = 0.591). Planned comparisons revealed no significant between group differences at any time point. The PCIT-T and COS-P groups displayed significant within-group decreases in externalizing behavior from T1 to T2 (PCIT-T: p < 0.001, d = −0.61, medium to large effect size; COS-P: p < 0.05, d = −0.56, medium to large effect size) and the PCIT-T group showed a significant decrease from T1 to T3 (p < 0.05, d = −0.51, medium effect size). The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2, 155) = 0.056, p = 0.572; T1 to T3: F (1,155) = 0.01, p = 0.908.

For internalizing behavior assessed using the CBCL internalizing scale, there was a significant effect for time (p < 0.001), but no significant differences for group (p = 0.248) or time x group (p = 0.334). Planned comparisons revealed no significant between group differences at any time point. The PCIT-T group showed a significant reduction in internalizing symptoms from T1 to T2 (p = 0.002, d = −0.48, medium effect size) but there were no other significant within-group decreases. The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P] were not supported, T1 to T2: F (2, 155) = 0.41, p = 0.664; T1 to T3: F (1, 155) = 0.25, p = 0.617.

For general problem behaviors assessed using the BITSEA-Problem scale, there was a significant effect for time (p < 0.005), time x group (p = 0.036) but no significant effects for group (p = 0.199). Planned comparisons revealed no significant between group differences at any time point but there was a significant within-group decrease in problem behaviors for the PCIT-T group from T1 to T2 (p < 0.001, d = −0.64, medium to large effect size) and from T1 to T3 (p < 0.001, d = −0.78, large effect size). The hypothesis that there would be a gradient effect in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] was supported, F (2, 153) = 4.04, p < 0.05. The hypothesis that there would be a gradient effect in terms of the magnitude of improvements from T1 to T3 [PCIT-T > COS-P] was not supported, F (1, 153) = 1.54, p = 0.216.

For compliance assessed using the DPICS-IV child compliance code (proportion of times the child complied with an effective parental command) there were no significant effects for time (p = 0.080), group (p = 0.859) or time x group (p = 0.250). Planned comparisons revealed no significant between group differences at any time point and no within-group decreases in child compliance (ps > 0.05). The hypotheses that there would be gradient effects in terms of the magnitude of improvements from T1 to T2 [PCIT-T > COS-P > WL] and T1 to T3 [PCIT-T > COS-P group] were not supported, T1 to T2: F (1, 155) = 0.23, p = 0.793; T1 to T3: F (1, 155) = 0.40, p = 0.525.

Reliable change

The rates of reliable change improvement (Table 3) were highest in the PCIT-T group for positive parenting verbalizations, negative parenting verbalizations, parenting sensitivity, non-attuned mind-mindedness, parent-reported parental reflective functioning, and empathy and understanding towards the child. The COS-P group showed the highest rates of reliable improvement on appropriate mind-mindedness, parental emotion regulation, caregiving helplessness, hostility and parenting stress. The WL group showed the highest rate of reliable improvement on child abuse potential. In terms of child outcomes, the PCIT-T group showed the highest rates of reliable improvement on all variables, except for child self-regulation, which was highest for the COS-P group. All participants (across all groups) showed reliable change in child compliance.

Post-hoc sensitivity analysis

Post-hoc sensitivity analysis results are shown in Supplementary Table 2. For positive parenting verbalizations (DPICS – LP), results remained largely unchanged as there were still significant effects for time, group, and time x group (ps < 0.05). The PCIT-T and COS-P groups showed significant increases in positive parenting verbalizations from T1 to T2 and the PCIT-T group showed significant improvement from T1 to T3 (ps < 0.05). At T1, the PCIT-T group used less positive parenting verbalizations than the COS-P and WL groups (p < 0.05). At T2, the PCIT-T group used significantly more positive parenting verbalizations than the COS-P and WL groups (ps < 0.05), and the COS-P group used more positive parenting verbalizations than the WL group (p < 0.05). At T3, the PCIT-T group used significantly more positive parenting verbalizations than the COS-P group (p < 0.05). For negative parenting verbalizations (DPICS – NTA), results were also like those of the main analysis as there was a significant effect for time (p < 0.001) but not group or time x group, and the PCIT-T group showed a significant decrease from T1 to T2 and from T1 to T3 (ps < 0.05). The WL group showed a significant decrease from T1 to T2 (p < 0.05). There were no significant between group differences at any time point. For parenting sensitivity (NICHD-Sensitivity), in alignment with the main analysis, the PCIT-T group showed a significant increase from T1 to T2, and there was a significant between group difference at T2, with the PCIT-T group showing greater sensitivity than the COS-P and WL groups (ps < 0.05). For all other variables (parental reflectiveness: CCQ-DRFQ; child social-emotional functioning: BITSEA-competence; problematic child behaviors: CBCL-externalizing; BITSEA-problem) the post-hoc sensitivity analysis results differed from the main analysis as there were no significant effects for time, group, or time x group, and planned comparisons revealed no significant between group differences at any time point and no significant within-group changes. The COS-P group showed significant decrease in child externalizing behavior problems from T1 to T2 (p < 0.05).

Discussion

This study represents the first RCT study to test outcomes of the most recent iteration of the PCIT-T model [36], and the first ever comparison of the PCIT-T model (in any form) with an active treatment condition. The study had a number of major findings. Compared to parents who participated in the active or non-treated control groups (COS-P and waitlist, respectively), parents who participated in PCIT-T were more sensitive, praised their children more, used fewer negative child-directed statements, and used fewer non-attuned mind-minded statements. Further, parents in the PCIT-T group showed statistically significant pre- to post-group changes in self-reported parental reflectiveness, and empathy and emotional understanding, and reported significant improvements in child social competence, internalizing problems, and general behavior issues. It is also of note that parents in the COS-P group showed increases in positive parenting verbalizations and decreases in negative parenting verbalizations while interacting with their child, and they also reported significant decreases in parental stress and child externalizing behaviors.

Given the well-document core role of caregiving sensitivity in the development of secure parent–child attachment relationships [22, 26], the results relating to improvements in parenting sensitivity following the PCIT-T intervention are noteworthy. Specifically, results showed that following treatment, the PCIT-T group showed more parenting sensitivity than the COS-P and WL groups, and at 4-month follow-up, they showed more sensitivity than the WL group. In terms of clinical significance, the PCIT-T intervention displayed a ‘large’ effect size for sensitivity immediately post-intervention (d = 1.03), a change that was sustained at the 4-month follow-up (‘medium-to-large’ range; d = 0.66). Further, the proportion of parents who showed reliable changes in sensitivity (74%) was larger among the PCIT-T than the COS-P and WL control groups (40% and 39%, respectively). These results confirm those of the one previous controlled study testing outcomes of PCIT-T [50, 53, 54] and suggest that PCIT-T is an intervention that is successfully able to support parents to respond to their toddler’s needs in a way that is sensitive and contingently responsive. The effect sizes for sensitivity obtained in this study are also striking and can also be considered alongside those reported for other attachment-based parenting interventions of similar intensity and duration. For example, ABC, which has had reported ds ranging from 0.08 to 0.7 [68], and VIPP-SD, which in Van Ijzendoorn et al. [88]’s recent systematic review had a combined effect size of r = 0.18, which the authors state was comparable to d = 0.37. Taken together, the impressive changes in maternal sensitivity obtained in this study suggest that PCIT-T is an effective way to enhance caregiving sensitivity, and that it may be as, or even more, effective than some of the other currently available and popular attachment-based intervention programs. The effectiveness of PCIT-T is likely to be attributable to its combination of live-coaching using the PRIDE and CARES models, and the focus on enhancing parental reflective functioning (both through coaching and conversation), however future studies are required to explore the specific ingredients of change.

The findings of this study regarding positive changes in parental reflective functioning following PCIT-T are also noteworthy. A parent’s ability to reflect on their child’s mental states, often referred to as ‘mentalising’ or ‘reflective functioning’, has been linked with the capacity to provide sensitive caregiving [14] and to the development of the child’s sense of self [77]. In this study, mentalization was assessed in two ways – observationally, operationalised a mind-mindedness [61], and via parent-reported, using the Diamond Reflective Functioning Scale [27]. Results showed that mothers in the PCIT-T group showed a reduction in observed non-attuned comments (with a medium effect size) and that the degree of improvement was greatest for the PCIT-T group. In terms of self-reported parental reflective functioning, the PCIT-T group were the only group to show significant improvement. When comparing reflective functioning outcomes between the PCIT-T and COS-P groups in this study, it is important to remember that PCIT-T is a program that is individualised to the parent–child dyad and that it is delivered in an intensive format (PCIT-T is twice weekly) for families with established problems, whereas COS-P is a group-based reflective parent psycho-educational program designed for a general parenting audience (i.e., not necessarily for families presenting with child behavioral issues). With these differences in mind, it is in some ways unsurprising that PCIT-T was associated with better reflective functioning outcomes than COS-P. Previous studies conducted with a more general parenting group (i.e., not those presenting with child behavioral issues/concerns) have shown changes in reflective functioning following COS-P [60] and so it is possible that the failure to find changes following COS-P in this sample related to the clinical nature of the sample. Further research is required to better understand the impacts that COS-P may have as an introduction to treatment for families with established problems. Overall, given results of the post-hoc sensitivity analysis—which did not find significant effects for any group—caution in interpreting the findings regarding reflective functioning is warranted, and further research in larger samples with less attrition is required.

In keeping with previous PCIT-T outcome studies [50, 51, 53, 54], PCIT-T also led to a range of positive changes in child functioning. Specifically, following PCIT-T, parents reported reductions in child externalizing and internalizing behaviors, reductions in general problematic behaviors, and improvements in social competence. It is of note, however, that reductions in child externalizing behaviors were also seen for the COS-P group, which suggests that participating in this educational group-based intervention may also be positive in terms of parent-reported reductions in child behavior problems. Caution in interpreting the findings regarding child behavior is warranted, however, given a) the lack of statistically significant differences between groups in externalizing, internalizing, general problems or competence at T2 and T3, and b) results of the post-hoc sensitivity analysis, which did not find significant effects for these variables.

The chief aim of most attachment-based early parenting interventions is to enhance the quality of the parent–child relationship, and to shift child attachment patterns from insecure and/or disorganized to secure and/or organized. It is of note that none of the changes on attachment variables in this study reached statistical significance. This may relate to the fact that the sample size was too small and thus underpowered to detect change, particularly for the SSP, which is a categorical measure (and so harder to detect change on) and which was only administered at T1 and T3 (where the sample size was small). It is, however, worth noting the effect sizes for the changes in AQS-security for the PCIT-T and COS-P groups. At both T2 and T3, the PCIT-T group showed medium effect size changes, placing it in a comparable range to other brief to medium-term attachment-based interventions, e.g., VIPP-SD (overall combined effect size r = 0.23, medium effect size; [13]). Interestingly, while the AQS-security effect size changes for the COS-P group were in the ‘small’ range at T2, they were in the ‘large’ range at T3. Again, while the sample size precludes firm conclusions, this is a promising result for the COS-P program and it is possible that attachment changes following COS-P occur but they take longer to develop. Given the small sample size at T3, any conclusions are speculative and there is a need for further examination in subsequent studies.

In this study, COS-P was used as an active control condition to which outcomes of PCIT-T could be compared, chosen largely because it is attachment-based and of a similar treatment dosage (approximately 16 h, usually over 8 weeks). The results obtained in this study for the COS-P group are, however, noteworthy in and of themselves. Like parents in the PCIT-T group, parents in the COS-P group showed statistically significant increases in positive parenting verbalizations and decreases in negative parenting verbalizations. They also reported significant reductions in parenting stress, and reductions in child externalizing behaviors. These are important findings because COS-P is a program that has been widely disseminated but not been extensively researched, and of the studies conducted to date, results have been mixed [18, 74, 95]. The findings in this study, with this specific sample of parent-toddler dyads (all who presented with toddler behavior issues), are impressive given that COS-P is not designed specifically for toddler-aged children with behavioral issues but is rather a more general parenting program for children of all ages and not focused specifically on addressing behavioral issues. It is of note, however, that parents who attended COS-P did not report changes in areas that the program specifically targets, e.g., parental reflectiveness, caregiving helplessness, and child attachment. One possible explanation for this finding is that the measures utilised in this study lacked validity. The validity of self-report measures to assess parenting representations such as parental reflective functioning and caregiving helplessness has been questioned [17, 64, 65], and a different result may have been obtained had an interview-based assessment been utilised, e.g., Parent Development Interview [78] or Caregiving Interview [34]. Regarding child attachment, it is possible that with a larger sample size, a different result may have been obtained. These measurement and sample size issues aside, it is also possible that the COS-P program did not bring changes in these domains because it is a discussion-based psycho-educational group program of only 8-weeks duration, and thus that a more intensive, individualised approach such as that of COS-Intensive may be required. It is also possible that the COS-P program may not be effective for this particular clinical sample (i.e., parents presenting with concerns about toddler behaviur issues). These potential issues notwithstanding, it is of note that changes on these variables were also not seen for the PCIT-T group, despite PCIT-T being a more intensive and individualised approach and designed specifically for this clinical group. Future research is required to better understand the impacts of COS-P for parents with different aged children, and with different presenting concerns. Given that neither COS-P or PCIT-T brought changes in parental reflectiveness, caregiving helplessness, and child attachment, future studies could investigate the impact of a combined treatment program for parents of toddlers with behavioral concerns, e.g., PCIT-T to teach practical play techniques and develop emotion regulation skills (in child and parent), followed by COS-P to focus on parental reflectiveness and caregiving helplessness. It is possible that such a combined approach may bring changes in child attachment.

Study strengths and limitations

This study had numerous strengths including use of a RCT design with a clinical sample, use of another active intervention, and a comprehensive mix of validated parent-report and observational measures, with a mix of in-clinic and home-based assessments. There are, however, limitations to acknowledge. First, the two active treatment conditions utilised different delivery modes (PCIT-T was delivered individually, COS-P was delivered in a group), which presents complexity and potential confounds. Significantly, while efforts were made to match the conditions in terms of treatment dose (total intervention time) and duration (total number of weeks from first to last session), in reality, there were clear dose differences between the groups because parents who attended PCIT-T received one-on-one attention and attended sessions twice weekly rather than once per week (as was the case for COS-P). In addition, while the majority of PCIT-T and COS-P participants completed the interventions, there were many more missed sessions in the COS-P group (PCIT-T and COS-P participants received, on average 13.95 and 11.03 h of treatment, respectively). While there was no evidence in this study that total treatment time moderated treatment outcomes, this is a significant study limitation, and it is possible that this may have played a role and so conclusions must be made with this in mind. Second, the study utilized many study measures, which increased the chance of Type I error. To address this, family-wise adjustments were made to the raw p-values to account for multiple comparisons however results should still be interpreted with this limitation in mind. Third, an inherent limitation of the study design was that the waitlist control group did not take part in the 4-month follow-up due to ethical reasons associated with denying patient care for this extended period. Fourth, the follow-up period (4-months post PCIT-T/COS-P completion) was brief. A longer-term follow-up (e.g., 9 or 12-months following treatment completion) would have provided more meaningful data about the longer-term impacts of the interventions, particularly changes in attachment, which can take longer to emerge. A fifth study limitation related to sample size. While a priori power analysis indicated that n = 90 would be sufficient to detect medium effect sizes, and the study used ITT analyses to account for missing data, it is of note that only 5 participants from the COS-P group completed the T3 follow-up. While results of the post-hoc sensitivity analysis suggested that the main study results relating to positive and negative parenting verbalizations and parenting sensitivity were not impacted by missing data, this is not the case for parental reflectiveness and child outcomes (social-emotional functioning and behavioral problems). Results thus need to be interpreted with caution, and future studies with larger samples (with less attrition) should be conducted. It is of note that participants who dropped out of the study were less educated and younger than those who completed it. This may have contributed to the differences between the main study and post-hoc sensitivity analysis results, highlighting an important direction for future research and clinical practice, i.e., implementation of deliberate strategies to engage and retain less educated and younger parents in clinical trials and clinical intervention programs such as PCIT-T and COS-P. Sixth, again with regard to education, overall, 48% of the mothers in the sample had a university education. This rate is slightly lower than the national average, with available data suggesting that 54.3% of people in Australia aged 25–34 years have a tertiary education, 60.8% who are women and 39.0% who are men [5]. While the educational profile of the parents in this study is likely a realistic representation of clinical populations, results and conclusions about transferability should nevertheless be interpreted with awareness that the tertiary education rates in this sample were slightly lower than the national average. Seventh, while one of the key parent-report measures of child externalizing problems, the CBCL, has been used successfully in previous studies with children aged as young as 12 months [89], this measure has not been validated for children aged less than 18 months and so there is a need to exercise caution in interpreting the results. To account for this, the study also used two additional measures of child behavioral issues (general behavioral issues were measured using the BITSEA, and child compliance was rated using the DPICS-IV coding system). The fact that the PCIT-T group showed positive changes across these measures suggests the likely validity of the CBCL results. An eighth limitation relates to the lack of information about parent-identified presenting issues at program entry. Families were deemed eligible for the study if the parent responded affirmatively to, ‘do you have concerns about your child’s behavior?’ and/or ‘do you have difficulties managing your child’s behavior?’ Unfortunately, without any other additional questions, little was known about the precise nature of the child’s behavioral concerns. Additional information regarding the child’s behavioral issues at program entry may have helped to characterise the sample and clarify the types of behaviors that parents were hoping to target when entering treatment. In this vein, detailed analysis of baseline questionnaire data, e.g., syndrome profiles on the CBCL, may be a potentially useful avenue for secondary data analysis of data from this trial. A final limitation was that neither participants nor research assistants were blind to treatment allocation once participants were allocated to treatment or waitlist. This may have biased results however given the nature of the trial it was not possible to blind participants. Furthermore, given that research assistants did not score measures, it is unlikely that being aware of participant treatment condition impacted outcomes.

Conclusions

The overarching aim of this study was to test whether PCIT-T outcomes were superior to a non-treated waitlist control condition, and to an active treatment comparison, COS-P. PCIT-T was developed as an adapted version of the standard PCIT protocol (a behaviorally focused program, designed for older children with conduct problems) and so in keeping with the standard PCIT literature and the skills-based nature of PCIT-T, the study examined behavioral outcomes (e.g., parenting skills and child behavior). However, it also examined variables that have not been well studied in PCIT-based programs but are common targets in attachment-based programs like COS-P (e.g., sensitivity, reflective functioning, attachment security). Taken together, results build on previous PCIT-T outcome studies [50,51,52,53,54] to highlight PCIT-T as an effective early parenting intervention program. The study shows that PCIT-T helps parents to develop and use positive parenting skills, increase parenting sensitivity, and enhance reflective functioning. Intervention effects also extend to toddlers, with evidence of reductions in problematic behaviors and increased social competence.

Future studies should test PCIT-T outcomes in larger samples, and with longer-term follow-up periods. Deeper exploration of associations between demographic factors and outcomes variables, as well as investigations of factors that moderate treatment outcomes, and of the mechanisms of change, would also be warranted; studies exploring treatment matching to determine what types of families and presenting problems map on best to different types of evidence-based interventions (e.g., PCIT-T, COS-P or COS-I, ABC, VIPP) would also be of benefit.