Parent-Child Interaction Therapy

Study 1

Summary

Nixon et al. (2003, 2004) randomly assigned 63 families to three groups: 1) the standard program, 2) the abbreviated program, and 3) a waitlist control group. Nixon et al. (2003) reported assessments at baseline, posttest, and 6-month follow-up. Nixon et al. (2004) included additional follow-up assessments done at 1 year and 2 years after the intervention. However, long-term results compared changes for the standard treatment group and for the abbreviated treatment group (no comparisons to a control group were possible).

Nixon et al. (2003, 2004) found that the intervention group children relative to the control group children had significantly

• Fewer behavior problems
• Lower parent stress
• Improved discipline practices.

Evaluation Methodology

Design: The two papers used a randomized control design in which sampled children with behavior problems (called the clinical sample) were randomly assigned to treatment or control groups. The design included an additional sample of children without behavioral difficulties (called the non-problem sample). This sample obviously differed from the clinical sample at baseline but was not used to test the efficacy of the treatment. The goal was to see if treatment groups reached the levels of the non-problem sample at posttest and follow-up.

Recruitment of treatment subjects came from responses to announcements of a free treatment program for preschool-aged children exhibiting behavioral difficulties. To be included, the child needed to be diagnosed for oppositional defiant disorder, have had symptoms for 6 months, and score in the clinical range on the Eyberg Child Behavior Inventory. Past traumatic experiences, severe physical or mental deficits, or current use of medication disqualified potential subjects. Of 71 families volunteering to participate, 54 were used in the analysis.

The study assigned subject families to three groups: 1) the standard PCIT, 2) the abbreviated PCIT, and 3) a waitlist control group. The standard PCIT group began with 22 families and 5 (23%) dropped out. The abbreviated PCIT group began with 23 families and 3 (13%) dropped out. The waitlist control group began with 18 families and 1 (6%) dropped out. The three conditions thus ended up with final sample sizes of 17, 20, and 17, a total sample size of 54, and a total attrition rate of 14%. It is not clear if the investigators attempted to follow up the dropouts and if the study met the intent-to-treat criterion.

The other sample of non-problem families had 21 volunteers recruited from newspaper advertisements. They needed to meet the opposite inclusion criteria: no diagnosis of the child as having oppositional defiant disorder, no parent reports of having trouble managing their child's behavior, scores on the Eyberg Child Behavior Inventory in the normal range.

In the first paper (Nixon et al., 2003), assessments occurred at baseline, posttest, and 6-month follow-up. Normally, PCIT lasts until the parent has mastered the skills and the child no longer meets the criteria for oppositional defiant disorder, but this study used a set number of sessions to compare the standard and abbreviated forms of the program. The standard program treatment consisted of 12 1-2 hour weekly sessions; the abbreviated program treatment consisted of 5 face-to-face sessions with five 30-minute phone consultations and access to videotaped instruction. Both treatments included a 1-hour booster session that occurred one month after treatment.

In the second paper (Nixon et al., 2004), additional follow-up assessments were done at 1 year and 2 years after the intervention. At the 1-year follow-up, 53 of 54 families completed the assessment and at the 2-year follow-up, 52 of 54 families completed the assessment. The 1-year assessment included parent report and observational data, but the 2-year assessment included parent reports only - done by phone and mail-back questionnaire. Further, the 2-year assessment contained substantial missing data: Only 10 of 16 in the standard treatment group returned questionnaires.

None of the follow-up assessments allowed for a control-group comparison because waitlist subjects entered the therapy after posttest.

Sample Characteristics: For the clinical group, the children had an average age of 4 and were mostly male. Mothers and fathers of these children were on average age 34, nearly all were Caucasian, and most were married. The parents averaged 12 years or less of education. The children in the non-problem sample were mostly male and 4 years old on average. Their parents on average had some technical or college education.

Measures: The DSM-IV was used to diagnose baseline oppositional defiant disorder among the clinical sample children. Other measures came in four forms: 1) parent reports on children, 2) parent self-report, 3) investigator observations of parent-child interactions, and 4) classification of scores into clinical categories.

Parent Report of Child Behavior. Mothers' perceptions of children's behavior came from established instruments: the Eyberg Child Behavior Inventory, the externalizing subscale of the Child Behavior Checklist, and the Home Situations Questionnaire - Modified.

Parenting Attitudes and Discipline Methods.

• Parenting Stress Index, a measure of frustration and low fulfillment (reliability established in other studies).
• Parent Sense of Competence Index, a measure of self-esteem and efficacy in the parenting role (test-retest reliability ranging from .46 to .82).
• Parent Locus of Control, a measure of the degree to which parents feel they can influence their child's behavior (alpha = .92).
• Parenting Scale, a measure of dysfunctional discipline practices such as laxness, over-reactivity, and verbosity (alpha = .82)

Independent Assessment of Child and Parent Behavior. Investigators videotaped a set of structured interactions between the parent and child and coded behaviors using the Dyadic Parent-Interaction Coding Systems - II. The codes measured parental praise, criticisms, and commands and child compliance and deviant behavior. An undergraduate student performed the coding after going through 40 hours of training. The coder was blind to the group membership and ratings were checked by the study leader. Test-retest reliability was .91.

Clinical Classification. Measures from mother's reports of child behavior and observations were used to 1) classify children as normal or clinical, 2) assess movement from clinical to normal, and 3) identify significant improvement (such as a 30% change in an outcome).

Analysis: For comparison of the three groups in the clinical sample, the outcomes were examined with ANCOVAs and baseline measures as covariates. Planned contrasts involved standard treatment versus waitlist control, abbreviated treatment versus waitlist control, and standard treatment versus abbreviated treatment. The text first reported unadjusted significance tests but later reported results with Bonferroni adjustments for non-independence of the outcomes.

The 1-year follow-up performed an intent-to-treat analysis by estimating missing data with the method of carrying the last observation forward. It also examined a subsample of those who completed all sessions.

Outcomes

Implementation Fidelity: Sessions were videotaped and 20% were coded for treatment adherence. Average therapist accuracy was 99%. Outside clinical psychologists rated the proficiency of the therapists as, on average, 7.24 and 7.71 on a scale ranging from 0 (poor) to 10 (excellent).

Baseline Equivalence: A series of ANOVAs on parent-report and observation data for the three clinical conditions did not reveal any difference on the baseline variables.

The clinical sample was equivalent to the non-problem sample on all demographic variables except maternal education, with the clinical sample having lower levels than the non-problem sample. However, mother's education had no relationship with any of the outcome measures. The clinical sample appropriately differed from the non-problem sample on the outcomes.

Differential Attrition: The authors say only that "Completers and noncompleters did not differ on any demographic or pretreatment variables."

Nixon et al., 2003: Posttest and 6-Month Follow-Up

Posttest: Results for the clinical sample for the continuous and classification measures were supplemented by results for the non-problem sample.

Continuous Measures. Tests of the program effects included five parent-report measures, four parent behavior measures, and five independent observations - a total of 14 outcome measures. The standard treatment did significantly better than the waitlist control group on 10 of the 14 outcomes, and the abbreviated treatment did better than the waitlist control group on 9 of the 14 outcomes. The significant effects occurred for each of the different types of measures (parent reports, parent behaviors, and observations). However, additional tests qualified these positive findings. The authors reported results that adjusted for the non-independence of the 14 outcomes. Using a Bonferroni standard of p = .05/14, they found significant benefits of the standard treatment for only 3 of 14 outcomes and significant benefits of the abbreviated treatment for only 3 of 14 outcomes.

Comparing the standard to the abbreviated treatment revealed no significant differences. Additional tests examined whether or not the outcome means at follow-up for the two groups differed by more than one standard deviation. These tests largely confirmed the equivalence of the two types of treatment.

Classification Measures. Tests for nine clinical classification measures largely confirmed the benefits of the standard treatment but less so for the abbreviated treatment. The standard treatment did significantly better than the waitlist control on seven of nine measures, and the abbreviated treatment did significantly better than the waitlist control group on three of nine measures. The standard treatment did significantly better than the abbreviated treatment on one outcome.

Comparisons to Non-Problem Group. The standard treatment and abbreviated treatment groups, although improved relative to the control group, continued to report significantly more behavior problems on most outcome measures than the non-problem group.

6-Month Follow-Up: Treatment of the waitlist group after the posttest prevented comparison of the control and treatment groups during the follow-up period. Comparisons of the standard and abbreviated treatment on measures of clinical classification showed, much like at posttest, no differences. The lack of main effects for time indicated that both groups maintained gains from posttest to follow-up.

Nixon et al., 2004: 1-year and 2-Year Follow-Up

1-Year Follow-Up: Results compared changes from baseline to the 1-year follow-up for the standard treatment group and for the abbreviated treatment group (no comparisons to a control group were possible). The tables presented results for both an intent-to-treat sample with imputed missing data and a completer sample of subjects who attended all sessions. However, the results differed little for the two samples.

For the continuous measures, the standard treatment group showed significant improvement on eight of nine measures, while the abbreviated treatment group showed significant improvement on six of nine measures. The trend over time did not differ significantly for the two groups. The measures failing to change significantly for one or both of the treatment groups included observational counts of mother criticisms, child compliance, and child deviance.

For the classification measures, the two groups showed similar percentages of families making significant change for the better, but the standard treatment tended to show somewhat more change than the abbreviated treatment.

2-Year Follow-Up: Because of new measures used in the phone and mail-back questionnaires at the 2-year follow-up, the tables do not examine changes over time. Also, response rates for the questionnaire were low. That said, the results indicated no differences between the standard and abbreviated treatment groups after 2 years.

Study 2

Summary

Matos et al. (2009) used a randomized control design involving a sample of 32 Puerto Rican families with children 4-6 years old with an ADHD diagnosis of the combined or hyperactive-impulsive type. The pretreatment assessment was followed by the intervention and then a posttest at about 3.5 months after baseline. A follow-up occurred 3.5 months after the posttest.

Matos et al. (2009) found that the intervention group children relative to the control group children showed significantly greater improvement in

• Hyperactivity
• Aggression
• Disruptive behavior
• Positive parental practices.

Evaluation Methodology

Design : The study used a randomized control design. A sample of 32 families was selected from children referred to the program by preschool centers and from parents responding to newspaper, TV, or radio ads. The sample was specialized. According to the inclusion criteria, the children must have had the following characteristics:

• 4-6 years old and attending a preschool program,
• hyperactivity and behavior problems, as reported by parents,
• an ADHD diagnosis of the combined or hyperactive-impulsive type,
• an IQ equal to or greater than 80,
• no significant developmental difficulty,
• a Puerto Rican mother who lived with them,
• not receiving stimulant or psychotropic medication,
• parental agreement of no other form of therapy.

Other inclusion criteria for parents included the absence of domestic violence, mental retardation, severe major depression, substance abuse, and psychopathology. Further, the sample excluded children with ADHD of the inattentive type because it typically begins at later ages and presents lower risk for problem behavior.

Of the 128 children referred to the study, 89 were excluded, leaving 39 who fit the criteria. Of these, 7 did not show up or agree to participate and 32 were randomized to the PCIT (n = 20) or waitlist (n = 12) conditions.

For the intervention, a therapist met individually with each family in a weekly 90-minute session. Since the therapy ended when parents demonstrated mastery of the program skills, some families attended more sessions than others. Control group members were contacted monthly by phone but received no treatment.

The pretreatment assessment was followed by the intervention and then a posttest done about 3.5 months after baseline. A follow-up occurred 3.5 months after the posttest. Given the use of a waitlist control group that received the opportunity to participate in the therapy after the posttest, the study could not make follow-up comparisons of the treatment and control groups.

All subjects completed the posttest and were included in the analysis. As appropriate for intent-to-treat analysis, the one PCIT subject who did not complete the treatment was included in the analysis. However, the data contained missing values for 3% of the sample at posttest and 9% at follow-up.

Measures: The numerous measures have accepted reliability for Puerto Rican samples or demonstrated reliability from other studies.

• The Disruptive Behavior Scale for Children included numerous ADHD items rated by a clinician.
• The Hyperactivity and Aggression Subscales of the Behavioral Assessment System for Children-Parent Rating Scale provided parental ratings of children's behavior.
• The Peabody Picture Vocabulary Test was used as an indicator of intelligence.
• The NIMH Diagnostic Interview Schedule for Children IV - Parent Version focused on the presence among children of ADHD, anxiety disorders, and depression as rated by clinicians.
• The Children's Global Assessment Scale yielded clinician ratings of impairment in adaptive functioning.
• The Eyberg Child Behavior Inventory came from a 36-item parent report on the frequency and number of conduct problem behaviors.
• The Family Experiences Inventory measured stressful experiences reported by the mother.
• The Parent Practices Inventory measured parental monitoring, supervision, involvement, and discipline.
• The Beck Depression Inventory measured depression among parents.
• The Treatment Evaluation Scale measured parent satisfaction with treatment.
• The Therapy Attitude Inventory measured the impact of the training on dealing with children.

Analysis: The analysis compared the intervention and control groups with ANCOVA and baseline scores treated as covariates. An alpha level of .0125 was used, with the .05 level "divided by the number of secondary outcomes (4)."

Values were imputed for the 3% missing data at posttest and 9% missing data at follow-up. The imputation was justified by a statement that "There was no identifiable pattern for the small percentage of missing data" and "no significant differences in participation of evaluation sessions as a function of treatment condition."

Outcomes

Implementation Fidelity: The investigators evaluated the content of each parent-child interaction therapy session with a checklist of therapist's actions and met weekly with therapists to review past sessions and plan future sessions. All treatment sessions were videotaped and analysis of a 20% random sample showed an integrity rate of 98%.

Baseline Equivalence: At baseline, children in the treatment and waitlist group did not differ significantly on gender, IQ, parent's sociodemographic characteristics, family structure, or the numerous outcome measures.

Differential Attrition: All subjects were included in the analysis. The small percentages missing data at posttest and follow-up were included with imputed values.

Posttest: The treatment group demonstrated significantly better outcomes on all six of the primary measures (relating to hyperactivity, aggression, and problems of the child) and on three of the four secondary measures (relating to parent and family outcomes).

To further illustrate the clinical significance of the effects, the percentage of children who moved out of the dysfunctional range on the six primary measures ranged between 50-75% in the intervention group and between 0-25% in the control group.

Long-Term: Comparisons could not be made with the waitlist control group, which had started the treatment by the time of the follow-up assessment. Alternatively, the analyses showed no significant differences between the intervention group means at the posttest and follow-up for the primary and secondary outcomes. Thus, the improvements appeared to be sustained in absolute terms.

Study 3

Summary

Boggs et al. (2004) conducted a quasi-experimental study of 61 families who had been referred to a psychology clinic for treatment of child disruptive behavior problems.

Boggs et al. (2004) found that the intervention group relative to the control group had significantly

• Fewer children classified as having oppositional defiant disorder, ADHD, or conduct disorder.

Evaluation Methodology

Design: The data came from a quasi-experimental study of 61 families who, in the initial study, had been referred to a psychology clinic for treatment of child disruptive behavior problems. The children were diagnosed as having oppositional defiant disorder and often comorbid conditions of conduct disorder and ADHD. The sample included children taking medication, as long as the medication dose would not change during the study. The sample excluded children with mental or physical disabilities or with parents having an IQ below 70. Families were randomly assigned to an intervention or a 4-month waitlist control group.

Dropouts from the original study are crucial for this follow-up study. Of the 33 families in the intervention group, 14 dropped out before completing the training. Of the 28 families in the control group, 17 dropped out while on the waitlist or while in treatment. For this study, all 61 families, both dropouts and completers, were contacted and 46 (75%) agreed to participate and returned questionnaires - 23 completers and 23 dropouts. The 75% willing to participate came equally from those completing and not completing the original treatment, allowing sufficient cases for comparisons of those receiving full treatment to those receiving partial treatment.

The follow-up assessment occurred on average 19.59 months after the pretest of the original study. Since some families started the program earlier than others, the time between the pretest and follow-up assessments varied from 10 to 30 months. The original study appeared to last 4 months, so the follow-up period ranged from 6 months to 26 months. Most but not all subjects were tested at least a year after the end of the intervention.

Sample Characteristics: The 46 children in the follow-up sample were mostly boys (78%), Caucasian (74%), and African American (15%). The mean age at original enrollment was 4 years, 11 months, but at the time of the follow-up they had reached an average of 6 years, 7 months.

Measures: The measures all came from the responses of parents, either over the phone or by mail-back questionnaire. The measures replicated those used in the original study.

The DSM-III-R Structured Interview for Disruptive Behavior Disorders was administered to the child's mother by telephone, but the researchers diagnosed the presence of the disorder. Studies have reported high levels of inter-rater agreement on this instrument. It is not clear if those who diagnosed the disorder were blind to group membership.

The Eyberg Child Behavior Inventory included 1) an Intensity Scale for the frequency of disruptive behavior and 2) a Problem Scale for parental tolerance of the child's behavior. These scales have been shown to have high reliability.

The Parenting Stress Index had a parent domain with items focusing on stress in the parent role and a child domain with items focusing on child behaviors that make it hard to develop a relationship. Other studies have reported high reliability for the index and domain subscales.

The Parent Locus of Control Scale focused on the parenting role and parent-child interactions. An alpha reliability of .81 has been reported for the scale.

The Therapy Attitude Inventory was designed to assess parent satisfaction with participation in parent training, family therapy, or parent-child therapy. The scale has been shown to have high reliability and correlate with changes in outcome measures.

Analysis: Analysis was intent-to-treat. Estimates of group-by-time interactions came from repeated-measures ANOVAs with controls for baseline values of the outcomes.

Outcomes

Implementation Fidelity: Since this follow-up study did not perform the intervention, implementation fidelity was not an issue.

Baseline Equivalence and Differential Attrition: For this study, the effectiveness of the randomization in the original study has less relevance than the difference between the completers and dropouts in this follow-up study. The quasi-experimental design made it crucial to demonstrate no differences at baseline between the completers and dropouts who agreed to participate in the follow-up.

Chi-square and t-test comparisons of the pretreatment characteristics of the completers and dropouts showed no significant differences on sociodemographics of parents and children, presence of conduct disorder, presence of ADHD, use of medication, child IQ, and child symptoms. Among the five outcome variables, one differed significantly - the Parenting Stress Index (child domain) was higher in the dropout group than the completer group.

Appropriately, the only significant difference was in the number of treatment sessions attended, with completers attending nearly four times as many as dropouts.

However, the dropout sample appears to have higher scores on the problem behaviors. With the small sample size of 46, some fairly large differences did not reach significance. For example, 30% of the completers were from single parent families compared to 57% of the dropouts. And 13% of the completers had comorbid conduct disorder compared to 30% of the dropouts. Thus, the dropout group had greater risks than the completer group.

Long-term: F tests for the group-by-time interactions were significant for four of five outcome measures. The completers improved significantly more from baseline than the dropouts on all four measures of child behavior but not on the Parent Locus of Control Scale.

The results also showed significantly different changes in the percentage of children meeting diagnostic criteria for oppositional defiant disorder, ADHD, and conduct disorder. For example, at baseline, 21 of 23 in the intervention group and 23 of 23 in the control group were classified as having oppositional defiant disorder. At follow-up, 5 of 23 (22%) in the intervention group and 13 of 23 (57%) in the control group were so classified.

Dose-Response: The design of the study in which dropouts participated in fewer sessions than completers is similar to a dose-response model, but the study did not specifically examine the influence of sessions attended on the outcomes. Rather it compared all completers and all dropouts as separate groups.

Study 4

Summary

Leung et al. (2008) used a quasi-experimental design involving 130 families with children ages 2-8. Subjects were referred to the program by hospitals, social service agencies, schools, and parents themselves because of concerns about the behavior of children. Assessments occurred at baseline and post intervention.

Leung et al. (2008) found that the intervention group relative to the control group demonstrated significantly

• Fewer child behavior problems

Evaluation Methodology

Design: The study used a quasi-experimental design. Subjects were referred to the program by hospitals, social service agencies, schools, and parents themselves because of concerns about the behavior of children. The children, limited to ages 2 to 8, had to meet certain unspecified inclusion criteria, probably those applied in other studies. All these subjects were assigned to the intervention. The control group was recruited separately from the intervention group and during a period from up to 12 months later. They came from referrals for parent training from preschools and primary schools and therefore appear to differ from the intervention group.

The intervention group had 53 dyads and the control group had 77 dyads. The intervention group underwent 1-hour sessions each week and ended when parents had mastered the skills and child behavior fell within normal limits. Assessments occurred at baseline, post intervention, and 6-month follow-up. The comparisons group completed baseline and posttest assessments but, having been designated as waiting for intervention, did not complete a follow-up assessment.

The sample experienced modest attrition by posttest. Of the 130 participations, 110 had complete data at the posttest. In the intervention group, 48 of 53 (89%) subjects completed the posttest, while in the control group, 62 of 77 (81%) subjects completed the posttest. Incomplete data in the intervention group resulted from program dropouts, who cited personal, health, or family problems as reasons for attrition. Incomplete data for the control group came from inability of investigators to find or contact the subjects. A note that seven of the intervention group subjects dropped out before completing the treatment but had nonetheless participated in the posttest indicates that the design satisfied the intent-to-treat criterion.

Attrition reached higher levels by the follow-up assessment. Data was obtained on 34 of the 54 (63%) subjects in the intervention group.

Measures: Along with sociodemographic information about parents and children, the study obtained data on several outcome measures. Most of the measures came from parent self-reports rather than unbiased observers and might be subject to misreporting. Observational measures coded by the investigators provided a check on the self-report but were done only for the intervention group and again might be biased toward finding a program effect.

Parent Report Measures. The Eyberg Child Behavior Inventory included 1) an Intensity Scale for the frequency of disruptive behavior and 2) a Problem Scale for parental tolerance of the child's behavior. The Chinese versions of these scales had reliabilities above .7 for the intervention and comparison groups. Cutoff scores of 131 for the Intensity Scale and 15 for the Problem Scale indicated serious behavioral problems.

The Parenting Stress Index has 36 items that measure three factors: parental distress, parent-child dysfunctional interaction, and difficult child. A total score from the three scales can also be calculated. With one exception, reliabilities exceeded .7.

Observational Measures. The Dyadic Parent-Child Interaction Coding System - II was used to code observed parent-child interactions at a clinical setting. The codes measured six behaviors: questions, criticisms, commands, descriptions, reflections, and praise. These measures were available only for the intervention group. At the time of the observations, investigators also asked parents about the frequency of corporal punishment during the previous month.

Analysis: ANCOVAs included controls for gender and baseline outcomes, many of which differed significantly across groups.

Outcomes

Fidelity Implementation: Social workers with 56 hours of training and 200 hours of supervision served as therapists for the parent-child interaction.

Baseline Equivalence: Likely due to the non-random assignment of subjects, the study failed to establish baseline equivalence. The treatment and control groups differed significantly at baseline on gender, the Eyberg Intensity Scale, the Eyberg Problem Scale, the Parenting Stress Index, the Parenting Stress index parent-child dysfunctional interaction measure, and the Parenting Stress Index difficult child measure.

Differential Attrition: Participants with complete and incomplete data did not differ significantly on baseline child behavior, parenting stress, or sociodemographic characteristics. In comparing dropouts across conditions, only one significant difference emerged: Those with incomplete data in the intervention group had higher baseline scores on the Parenting Stress Index than those with incomplete data in the control group.

A check on attrition imputed values for participants with incomplete data by assuming no change (i.e., substituting the baseline score for the missing posttest score). The results replicated those reported below for the sub-sample with complete data.

Posttest: The results showed significantly lower posttest scores for the intervention on six key outcomes: the Eyberg Intensity Scale, the Eyberg Problem Scale, the Parenting Stress Index total score, and all three Parenting Stress Index subscales. Effect sizes ranged from .97 to 1.59. To illustrate, among the intervention families with Eyberg scores above the cutoff for serious behavioral problems, 71% had scores below the cutoffs at posttest. Among the control families with scores above the cutoffs, 16% had scores below the cutoffs at posttest.

Long-term: Because of the waitlist control group, the follow-up examined maintenance of change among the intervention group rather than maintenance of differences between the intervention and control groups. Consistent with but not proof of sustained effects, the results showed a significant drop in the posttest and follow-up scores from baseline and no significant difference between the posttest and follow-up scores.

Study 5

Summary

Chaffin et al. (2004) used a randomized control design with 110 parents referred to the study upon entering the child welfare system for a new confirmed physical child abuse report. Assessments included self-report questionnaires, observational coding of parent-child interactions, and administrative data from the state child welfare database, and data were collected at baseline and at posttest (after the 6-month treatment).

Chaffin et al. (2004) found that the intervention group relative to the control group had significantly

• Fewer reports of child violence.

Evaluation Methodology

Design: The study used a randomized control design. Abusive parents and their abused child were referred to the study upon entering the child welfare system for a new confirmed physical abuse report. Referred parents were accepted into the study if they met the following conditions: 1) the parent (including step-parent or other caregivers) and child could participate (no legal termination of parental rights had been initiated), 2) the abusive parent had an IQ of at least 70, 3) the child was between 4 and 12 years old, 4) the parent did not have a designation as a sexual abuse perpetrator, and 5) the parent provided voluntary consent to participate.

Of 300 dyads referred, 112 (37%) met the inclusion criteria and were enrolled as participants. Most of those not included declined to participate or could not be located. Data from the child welfare agency showed no difference between those enrolled and not enrolled on gender, age, race/ethnicity, family structure, or kinship ties between parent and child. However, two subjects were dropped when it became apparent that the parents could not understand or answer the assessment questions. The final sample size thus equaled 110.

Subjects were assigned to one of three conditions:

1. PCIT. The treatment included two components: a six-session orientation group to increase parent motivation for active participation (needed because many participants are coerced into attending) plus the 12-session parent-child interaction therapy. The intervention occurred in a clinic setting.
2. Enhanced PCIT. In addition to the standard treatment, subjects received services targeting depression, substance abuse, and domestic violence; the services were commonly provided by staff and designed to meet the particular needs of the parent.
3. Standard Community Group. A community-based non-profit agency offered a parent training program that provided information on improving skills such as listening, discipline, stress management, and anger control. The sessions covered the same topics and lasted the same length of time as PCIT but did not involve interaction with children.

Assessment occurred at baseline and at posttest, after the 6-month treatment. The final outcome of repeated abuse incidents came from a database search after the end of treatment.

Following the intent-to-treat principle, investigators attempted to obtain 6-month posttest data on dropouts. About 25% of early dropouts and nearly all completers provided the psychometric posttest data, but an unreported number of subjects were missing posttest data. Although it is not possible to compute the attrition rate, it appears to be non-trivial. For posttest data, the researchers used multiple imputation methods. For follow-up, figures on re-reports of child abuse from a central database could be measured independent of the involvement of the subjects and contained no missing data (n = 110).

Sample Characteristics: The parents were largely female (65%) and had a mean age of 32 years. About 34% were married at baseline, 52% were white, 40% were African American, and 27% had some college or a college degree. Over 62% were characterized as living below the poverty line, and 64% received some form of public assistance.

Subjects on average were referred for two prior child welfare physical abuse reports and two child welfare neglect reports. IQ scores ranged from 70 to 114 with a mean of 95. About one-third had symptoms of a lifetime alcohol or drug disorder. Antisocial personality disorder (16%) and depression (22%) were common.

Measures: Data came from three sources: 1) self-report questionnaires, 2) observational coding of parent-child interactions, and 3) administrative data from the state child welfare database. Questionnaire data was obtained at baseline, after the six-session orientation program, and after the 12-session parenting program. Observation coding occurred at the same three times. Database matching occurred for a median follow-up time of 850 days.

The Child Abuse Potential Inventory used a 160-item questionnaire and four subscales to estimate risk for committing child abuse (alpha reliability ranged from .79 to .94 for the four subscales).

The Behavioral Assessment for Children used parent and, if possible, teacher reports to measure adaptive as well as problematic behaviors of children. It had temporal stability and internal consistency scores ranging from the mid .70s to the low .90s.

The Dyadic Parent-Child Interaction Coding System - II was used to code observed parent-child interactions during structured tasks. The coding scheme corresponded to the goals of the parent-child interaction therapy and therefore served as a measure of skills learned in the program. The correlation between the study scores with scores given by off-site coders was .94 for negative behaviors and .84 for positive behaviors.

The Abuse Dimensions Inventory rates the severity of sexual and/or physical abuse on the basis of written child welfare investigation information and/or an interview with the child welfare worker. The mean interrelated reliability was .76.

The Child Neglect Index was coded based on a review of written child welfare investigation information and/or an interview with the child welfare worker.

The Beck Depression Inventory is a 21-item self-report measure (alpha = .90).

The follow-up measure of re-reports came from checks of the central state child welfare database. The checks made certain to find matches on multiple criteria and avoid duplication.

Analysis: The study used survival analysis to model the timing to the first instance of abuse contained in the child welfare database. The study lacked details on the dates for which the subjects were followed but noted a median follow-up period of 850 days. The analysis did not control for exposure to child abuse, as a later study did (Study 6). Not measuring survival for the period in which the parent had access to the child might exaggerate the influence of the program (those most prone to abuse their children would be most likely to lose parental rights and show unusually long survival). However, the bias should work similarly across all groups.

Outcomes

Implementation Fidelity: Supervisors observed parent-child sessions and completed adherence checklists. Average protocol adherence reached 93%.

Measures of individual treatment services obtained by parents demonstrated expected differences between the PCIT and enhanced PCIT groups. Information on the number of referrals and the sessions attended (outside the parenting program) came from study organizers. Missing data were imputed on sessions attended for less than 10% of the cases. Group comparisons showed, as expected, that subjects in the enhanced PCIT condition received significantly more outside services and attended more outside sessions than the other groups. The mean number of additional service sessions was 9.3 for the enhanced condition and 1.9 for the other two groups. About 79% of the enhanced group received outside services compared to 32% in PCIT and 20% in the control. Investigators made sure that outside therapists did not counsel parents in parent skills that were obviously inconsistent with PCIT skills.

Baseline Equivalence: No significant baseline differences existed for measures of abuse, family characteristics, child characteristics, parent sociodemographic characteristics, lifetime drug use, alcohol symptoms, depression, or antisocial personality symptoms.

Differential Attrition: The PCIT and enhanced PCIT conditions had better retention than the community control condition. A three-category measure identified subjects who dropped out during the orientation, subjects who completed the orientation but less than half the core parenting program, and subjects who completed the orientation and most of the core parenting program. Controls for the categorical attrition pattern and the attrition pattern by condition did not significantly influence survival. This finding indicates that differential attrition did not bias the results.

Long-term: Re-reports of child abuse occurred for 19% of the PCIT condition, 36% of the enhanced PCIT condition, and 49% of the community group. Consistent with the percentages, the PCIT condition had significantly better survival than the control condition (p = .02). The enhanced PCIT group did worse than the PCIT group but not significantly so (p = .13). The enhanced PCIT did better than the control group in timing to recidivism but again not significantly so.

The results held across gender and race/ethnic groups and across younger as well as older children.\

Study 6

Summary

Chaffin et al. (2011) conducted a randomized control design with 192 parents of children (2.5 to 12 years of age) that had been referred by a child welfare agency for neglect and/or physical abuse. Many were mandated to participate in the treatment by child welfare services. Outcome measures included self-report questionnaires, observation coding of parent-child interactions, and administrative data collected from the state child welfare database. Database matching occurred from baseline up to 3.5 years later.

Chaffin et al. (2011) found that the intervention group children relative to the control group children showed significantly

• Reduced child welfare reports.

Evaluation Methodology

Design: The study used a randomized control design. Subject parents were referred to the program by a child welfare agency for neglect and/or physical abuse. The agency was located in the inner city and contracted with the state system to operate a parenting program. The parents had to have a child ages 2.5 to 12 who would be permitted to participate in the parent-child interaction therapy. Many were mandated to participate in the treatment by child welfare services. Parents were excluded if they had an IQ below 65 or previously received the treatment. Enrollment occurred between January 2004 and August 2006.

The analysis sample included 153 parents. Recruitment began with 291 parents approached to participate in the study, but 42 declined, 38 were not eligible, and 19 failed to complete the pretest. The remaining 192 (66%) were randomized, but 21 lost parental rights and were removed, 11 refused services, and 7 dropped out. The remaining 153 biological parents, step-parents, or primary caregivers comprised the analysis sample and an attrition rate of 20% of those randomized. No attrition occurred among the 153 parents over the remainder of the program. Efforts to follow all subjects meet the intent-to-treat criterion.

Randomization occurred in two steps, first for the orientation conditions and then for the therapy conditions. The first randomization assigned 192 parents to either a self-motivation orientation condition or a usual orientation condition. The self-motivation orientation involved meetings without the children and included activities to highlight the benefits of the program; the usual orientation was primarily informational and educational.

The second randomization assigned the 153 parents remaining after the orientation to a parent-child interaction condition or a service-as-usual condition. The parent-child interaction condition followed program guidelines to promote positive interaction between both parent and child; the service-as-usual sessions provided information to the parent on child development, reasonable expectations, compassionate parenting, and the value of empathy.

The sequential randomization thus defined four conditions: 1) self-motivation orientation and parent-child interaction therapy; 2) usual orientation and parent-child interaction therapy; 3) self-motivation orientation and service-as-usual; and 4) usual orientation and service-as-usual (i.e., control). The design allowed for the analysis to disentangle the benefits of the orientation from the benefits of the therapy.

The program included six orientation sessions and 12 parenting sessions. Assessments occurred at baseline, after the six-session orientation, and after the 12 sessions of the parenting program. The length of the intervention varied across subjects, making it difficult to pinpoint a specific end, but the median time from baseline to collection of post-treatment data was 247 days. The period of follow-up differed as well, but the median was 2.5 years.

Sample Characteristics: Participants were 75% women with a mean age of 29. By race/ethnicity, 60% were white, 19% African American, 9% Native American, 7% Hispanic, and 6% other. The sample was disadvantaged: 75% had income below the official poverty level, and 65% were not married.

Measures: Data came from three sources: 1) self-report questionnaires, 2) observation coding of parent-child interactions, and 3) administrative data from the state child welfare database. Questionnaire data was obtained at baseline, after the six-session orientation program, and after the 12-session parenting program. Observation coding occurred at the same three times. Database matching occurred from baseline up to 3.5 years later.

The Readiness for Parenting Change Scale measured motivation for participating in a child welfare parenting program and attitudes toward being mandated to attend the program (alpha reliability = .84). This scale served as a measure of responsiveness to the orientation.

The Dyadic Parent-Child Interaction Coding System - II was used to code observed parent-child interactions during structured tasks. The coding scheme corresponded to the goals of the parent-child interaction therapy and therefore served as a measure of skills learning in the program. Coder training required at least 90% correct with standard video stimulus sets.

The Child Abuse Potential Inventory used a 160-item questionnaire to estimate risk for committing child abuse (alpha reliability = .92).

The Child Welfare Database was used for the key measure of recidivism. Investigators checked the central state database of all child welfare reports, allegations, dispositions, and placements for incidents involving the subjects. The checks made certain to find matches on multiple criteria and avoid duplication. The placement information was also used to determine periods of time when children were removed from the parent.

Analysis: Event-history or hazard models were used to estimate the time from baseline to an incident of child abuse recidivism. However, comparisons across conditions in the models would be biased by differences in the time parents spent living with their child. Those parents who have high potential for abuse might have low recidivism because the child was not allowed to live at home with them. The analysis used two approaches to deal with this potential bias, both of which gave much the same results. First, the models examined the hazard for abuse relative to the interval during which the parent was at risk of abuse (i.e., living or spending unsupervised time with the child). Second, the models used imputed data for the period of not living or spending time with the child. The authors devoted an extended discussion to these relatively uncommon procedures.

The hazard or event history models included covariates (age, prior referrals, and number of children in the home). Because all subjects entered baseline with an abuse history, a control for the baseline outcome was unnecessary.

Outcomes

Implementation Fidelity: The self-motivation orientation sessions and the parent-child interaction sessions were monitored with checklists. To quote, "All therapists were able to maintain reasonable fidelity." Mean attendance at the six-session orientation was 5.2, and mean attendance at the 12-session parenting treatment was 10.1.

Baseline Equivalence: Comparisons of the self-motivation orientation group and the usual orientation group showed no significant baseline differences on sociodemographic characteristics, prior welfare referrals, number of children in out-of-home placement, the Child Abuse Potential Inventory, the Readiness for Parental Change scale, or the Dyadic Parent-Child Interaction codes. Comparisons of the parent-child interaction and service-as-usual groups also did not show any significant differences on the same set of variables.

Differential Attrition: Of 192 parents starting, 153 remained for the analysis. The main source of attrition was loss of access to children, which prevented assignment to the parent-child interaction condition. The investigators stated that attrition rates and reasons for attrition did not differ significantly across the conditions. However, they did not formally examine differences in baseline characteristics among completers and dropouts.

Long-term: The long-term results focused on newly reported incidents of child abuse available from the administrative data. The raw percentages show clear evidence of the program benefits:

• The combined self-motivation orientation and parent-child interaction condition had a recidivism rate of 29%.
• The parent-child interaction only condition had a recidivism rate of 47%
• The self-motivation orientation condition had a recidivism rate of 34%.
• The control condition a rate of 41%.

However, these figures do not control for exposure to the risk of recidivism (i.e., having children at home). Results from the event history models that are based on the observed risk intervals and the imputed data correct for this bias. With the corrections, significantly longer periods without abuse were found for the combined self-motivation orientation and parent-child interaction condition relative to all three of the other conditions. The results indicated that the combined treatments were needed for effective reduction of recidivism and that neither the self-motivation orientation nor the parent-child interaction alone had much benefit.

Further, the combined condition was significantly more effective in reducing the risk of recidivism when children were present in the home for long periods. It most helped when children were at home at the start of the study or returned soon after the start.

Study 7

These articles compare standard PCIT and a version adapted specifically for Mexican American children (called Guiando a Ninos Activos or GANA) to a treatment as usual control condition. GANA was designed to maintain the core elements of PCIT while adapting them to be more culturally relevant for Mexican American families. Effects of both treatment conditions compared to the control condition are reported below.

Summary

McCabe & Yeh (2009) and McCabe et al. (2012) tested the standard program and a version of the program adapted for Mexican American children using a randomized controlled trial. Families with Mexican American children ages 3-7 were randomly assigned to receive the standard program (n = 19), the version modified for Mexican American children (called Guiando a Ninos Activos or GANA, n = 21), or treatment as usual (n = 18). Several measures of child behavior problems and parenting were assessed at baseline, posttest, and long-term follow up (average of 15.9 months post-treatment).

Compared to treatment as usual, families that received the intervention demonstrated significantly improved:

• Disruptive behaviors (at posttest and long-term follow-up)
• Parent-child interactions (at posttest)
• Child externalizing problems (at posttest and long-term follow-up)
• Symptoms of ADHD, oppositional defiant disorder, and conduct disorder (at posttest and long-term follow-up)

Risk and Protective Factors

• Positive parenting behaviors (at posttest)
• Parent stress (at posttest)
• Parent locus of control (at long-term follow-up)

Evaluation Methodology

Design:

Recruitment: Families were recruited from a community mental health clinic. Families were eligible if the parent identified their child as being Mexican American and between the ages of 3 and 7, if the child received a score above the clinical cut point on the Intensity Scale of disruptive behaviors from the Eyberg Child Behavior Inventory, and if neither the parent nor the child was participating in any other psychosocial treatment simultaneously. Possibly eligible families were screened by telephone. Of 103 families screened, 32 did not meet inclusion criteria, 11 declined, and 2 could not participate for logistical reasons. The remaining 58 families enrolled in the study.

Assignment: Eligible families were randomly assigned to either the GANA condition (PCIT adapted for Mexican American children, n = 21), the PCIT condition (n = 19), or a treatment as usual condition (TAU condition, n = 18).

Assessments/Attrition: Assessments occurred at baseline, posttest (immediately after treatment), and at long-term follow up. The long-term follow up occurred, on average, 15.9 months (SD = 4.25) post-treatment, with a range of 6-24 months. Of the 58 families that completed the baseline measures, 54 completed the posttest (93%), and 48 completed the long-term follow-up (83%).

Sample: The primary caregivers were mostly female (92%), although male caregivers participated in at least one session 40% of the time. Female caregivers were, on average, 32.2 years old; male caregivers were, on average, 35 years old. Overall, 51% of mothers and fathers reported high school education or less and yearly average income was $23,271. The children's average age was 52.8 months. The sample included more male children (71%) than female children.

Measures: The authors stated that assessors and families were blind to condition. Parents completed seven outcome measures at pretest, posttest, and long-term follow-up. There were also seven observational measures that were assessed at pretest and posttest. Note that many of these parent-reported measures overlap in child externalizing and disruptive behavior content.

Outcomes:

• Two subscales from the 36-item Eyberg Child Behavior Inventory (ECBI): the Intensity Scale measuring frequency of disruptive behaviors and the Problems Scale measuring total behavior problems. This measure was administered during the initial screening call, posttest, and long-term follow up.

• The Externalizing Problems scale from the Child Behavior Checklist (CBCL), which was administered at pretest, posttest, and long-term follow-up.
• The Early Childhood Inventory, which assesses symptoms of ADHD, oppositional defiant disorder, and conduct disorder. This measure was administered at pretest, posttest, and long-term follow-up.
• The Parent Child Dysfunctional Interactions Scale from the Parenting Stress Index, administered at pretest, posttest, and long-term follow-up.

Risk and Protective Factors:

• The Dyadic Parent Child Interaction Coding System (DPICS), which is an observational behavioral measure of the quality of parent and child interactions. This measure was administered at pretest and posttest.
• The Parenting Practices Scale, which assesses positive parenting behaviors and was administered at pretest and posttest.
• The Parenting Distress Scale and the Parent Child Dysfunctional Interactions Scale from the Parenting Stress Index, which was administered at pretest, posttest, and long-term follow-up.
• The Parental Locus of Control Scale (PLOC) of child behaviors, which was administered at pretest and long-term follow-up.

All measures demonstrated adequate reliability (alphas above .70, kappa coefficients above .59).

Analysis: Treatment effects at posttest were assessed using single-degree of freedom contrasts, comparing the treatment conditions with one another and the treatment as usual condition. Baseline outcome controls were used in analyses to determine pretest-posttest treatment gains.

Intent-to-Treat: Data were analyzed following an intent-to-treat approach.

Outcomes

Implementation Fidelity: Sessions were video-taped and coded for fidelity. Both treatment conditions scored well on the inclusion of items from the treatment manual (77% in GANA, 82% in PCIT). A random subset of 25% of videos were coded by both coders and their agreement was 85%. The authors reported that 76% of families (n=44) attended five or more sessions and 57% of families completed all sessions.

Baseline Equivalence: There were no significant differences at baseline on eight demographic measures. It is unclear whether the conditions differed on outcome measures at baseline.

Differential Attrition: Only four of 58 families did not complete posttest measures. However, formal tests for differential attrition were not conducted at posttest. The authors stated that the three conditions did not differ on rates of attrition, but it appears the reported rates are for those who dropped out of the intervention condition rather than completing measures at posttest. Attrition at long-term follow up did not significantly differ by condition.

Posttest: The GANA intervention condition produced significantly more improvement than TAU on all seven of the parent-reported outcomes: disruptive behaviors, total problem behaviors, symptoms of ADHD, oppositional defiant disorder, and conduct disorder, externalizing problems, and parent-child dysfunctional interactions. The standard PCIT treatment condition also produced significant improvements, relative to TAU, for one of seven behavioral outcomes, externalizing problems. In no instances did the GANA and PCIT conditions significantly differ from one another, on either parent-reported or observational measures of outcomes. Posttest effect sizes were reported within condition (pre-post) rather than between treatment and TAU conditions.

For risk and protective factors, the GANA intervention condition showed significantly greater improvement than TAU for parenting practices, parent distress, difficult child, and total parent stress. The standard PCIT also showed greater improvement in parent stress, compared to TAU.  According to Table 3, GANA and PCIT families also showed significant improvement, relative to TAU families, on several observational measures of parenting skills.  

Long-Term: Six to 24 months post-treatment, GANA, relative to TAU, produced significantly greater improvements on five of eight parent-reported outcomes: disruptive behaviors (d = .81), CBCL internalizing problems (d = .56), externalizing problems (d = .65), and total problems scales (d = .63), symptoms of ADHD (d = .52) and for the risk and protective factor of parent control of child behaviors (d = 1.24). GANA outperformed PCIT on the CBCL internalizing subscale (d = .31). PCIT and TAU did not significantly differ on any outcomes at long-term follow-up.