METHODS

A postal survey was administered to the parents of all the children aged 2–8 years inclusive, registered at three general practices in Oxford (n = 1788).¹⁹ This survey comprised the Eyberg Child Behaviour Inventory (ECBI)²⁰ and demographic information. A total of 1105 questionnaires were returned, giving information on children in 800 families (response rate 70%). The mean score on the Eyberg Intensity Scale in this population was 102.7 and the median value was 100. A total of 193 children (17.9%) scored above the clinical cut off point of 127.

The parents of all children scoring above the median value (487 children in 391 families) were invited to participate in a randomised controlled trial of the Family Nurturing Network’s parenting programme (Webster-Stratton’s Parents and Children Series). The median rather than the mean was used because of the skewed distribution of behaviour scores. Parents of either sex and any ethnic group were eligible. Children already receiving treatment for behaviour problems (27 children, 2.5%) and those with learning difficulties (78 children 7.1%) were excluded. These children were felt to be special cases warranting subgroup analysis for which the trial was not powered. A sample size calculation gave a total sample size of 100 to allow detection of an effect size of 0.6 (a moderate effect)²¹ on the Eyberg Child Behaviour Inventory Intensity Score with 80% power at the p < 0.05 level of significance, or 120 to detect an effect size of 0.5, and 116 parents were actually recruited into the trial. (See web supplement for full description of sample size calculation.)

Parents were invited to join the trial by a letter addressed to the parent who had returned the survey questionnaire. Parents who consented to join the trial were compared with eligible parents who did not consent, with regard to the mean intensity scores of their children, whether they were defined as a “case” on the intensity score or not, and by social class based on the respondent’s occupation.

We adopted a block randomisation strategy. The parents who consented to join the study (n = 116; uptake = 30%) were allocated to one of two groups with the aim of achieving groups balanced in terms of child “caseness” on the Eyberg intensity score, age, sex, social class, and ethnicity,²² as well as preferences for attendance. These two groups were randomised, by tossing a coin in the presence of an independent witness, to treatment or control. It is particularly important to take parents’ preferences into account in the implementation of health promotion interventions, and this method of randomisation allowed us to allocate parents according to the time of day and day of the week they were available to attend, as well as their need for a créche, at the same time as producing six viable groups of 10 parents for delivery of the programme. Each consenting parent had an equal chance of being in the control or intervention group. The allocation to group was undertaken blind to general practitioner practice and area of residence, so neither of these factors influenced allocation to control or treatment group.

The intervention comprised sessions lasting about two hours each, once a week for 10 weeks. The activities included video vignettes of parent-child interactions, group discussion, role play, rehearsal of parenting techniques, and home practice. The techniques covered included play and positive interaction with the child, clear commands, limit setting, ignoring undesirable behaviour, praising and rewarding desirable behaviour, and following through on discipline. Four of the programmes were run at a health centre in the evenings between 7 30 and 9 30 pm. The other two programmes were run at a local community centre in the daytime, one between 9 30 and 11 30 am (with a créche available) and the other between 12 noon and 2 00 pm. Each group was run by at least one health visitor with either a second health visitor or a nursery nurse as co-leader, all of whom had received three days of training from the Family Nurturing Network in Oxford, specifically to deliver this programme. The group leaders (seven in all) received weekly supervision meetings as a group during the delivery of the programmes, to provide support and ongoing training, and ensure equivalence of the six intervention programmes and integrity of programme delivery compared with the instruction manual.

The 56 control group parents received no extra intervention.

Self completion questionnaires were administered to all consenting parents prior to the intervention, immediately post-intervention, and at six months post-intervention. The primary outcome measure was the Eyberg Child Behaviour Inventory (ECBI). The ECBI was developed in the USA from parent reports of children’s problems behaviour and has two scales, one which measures the intensity or frequency of the problem behaviour, and the other which measures the extent to which this behaviour is a problem for the parents. The number of items and range of responses ensure that this scale is sensitive to change, but it includes “maturational” items such as “wets the bed” which may change as the child matures. Secondary outcome measures included the Strengths and Difficulties Questionnaire (SDQ).²³ This was developed in the UK and includes prosocial as well as antisocial behaviours. This scale does not have the “maturational” items, but focuses on temper tantrums, fighting, and stealing. It also contains an “impact” measure of the effect of the child’s behaviour on their friendships, learning, leisure, and family life. It is generally recommended for use in children aged 4 years and above. The General Health Questionnaire (GHQ)²⁴ was used to capture the parents’ level of anxiety and depression; the Parenting Stress Index (PSI)²⁵ was used to measure changes in the specific stresses associated with the parenting role; and the Self Esteem Scale²⁶ was used to measure the parents’ self esteem.

The data were analysed using SPSS version 9. All analyses were carried out on an intention to treat basis, including in the intervention group both parents who had attended the parenting programme and those who had not. The statistical significance of changes in scores from baseline to post-intervention and six month follow up, within the control and intervention groups separately, was calculated using paired t tests for normally distributed outcome measures (the hyperactivity subscale of the SDQ and the SDQ total score, the PSI parent and difficult child domains, and the PSI total score). The Wilcoxon matched pairs signed rank sum test was used for outcome measures which were not normally distributed (ECBI intensity and problem scores, SDQ conduct, emotional, peer problems, prosocial and impact scores, PSI parent child interaction domain, GHQ somatic symptoms, anxiety, social dysfunction, depression and total scores, and the SES). Grouped t tests were used to compare the mean change in scores in the control and intervention groups where the differences were normally distributed (ECBI intensity score, SDQ total score, PSI parent child interaction, and parent domains), and Mann-Whitney U tests for the mean change in scores in the two groups where the differences were not normally distributed (ECBI problem score, SDQ conduct, hyperactivity, emotional, peer and prosocial scales, GHQ somatic anxiety, social, depression and total scores, PSI difficult child domain and total score, and SES).

RESULTS

The parents who consented to join the trial were representative of all the eligible families, in terms of the behaviour scores, social class, single parents, and ethnicity, as shown in table 1⇓. However owing to over representation of non-manual families in the survey, manual families were under represented compared with population norms. Consenting parents had slightly more children with behaviour in the clinical range, but this difference was not significant.

Thirty four of the 60 parents randomised to the intervention group attended 50% or more of the sessions (fig 1⇓). Reasons for dropping out of the programme included increased work commitments, moved away from the area, depression, other life stress, holiday fell at start of group sessions, missed too much to begin. Only one potential participant stated that the group was not helpful as a reason for dropping out. Forty six control group and 46 intervention group participants (79% of all initial participants) returned questionnaires at the six month follow up. There was no difference in the proportion of attenders or non-attenders returning questionnaires at follow up.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Attrition.

DISCUSSION

Our initial survey had a good response rate of 70%. While all social groups were well represented in the trial, the sample from which the study was drawn was not entirely representative. Manual class families were under represented and ethnic minority families over represented. The families who consented to enter the trial were representative of eligible families from a socioeconomic point of view and they were also more likely to have a child with clinical level behaviour problems than were those who refused. This single study cannot provide evidence about what would happen were such a programme to be offered in other areas, and the under representation of manual groups detracts from the generalisability of the results. However, the fact that parents from all social groups, including unskilled manual workers, minority ethnic groups, and single parents did attend and that the programme preferentially attracted parents of children with behaviour problems must be viewed as encouraging.

After excluding children with learning disabilities and previous treatment for behaviour problems, 30% of families consented to enter the trial, to be randomised to receive a 10 week parenting intervention or not, and to complete several questionnaires. Given the number of other commitments that families of children in the 2–8 year age range are likely to have, this represented a significant level of interest in the intervention on offer. If the intervention were to be rolled out as a universal service, uptake might be expected to be higher than the 30% achieved for a trial, since parents would not be making a commitment to filling in questionnaires or to being randomised. In addition, if parenting programmes were regularly available within general practices, uptake would be cumulative with additional chances for parents to participate later if they could not attend the first group to which they were invited. This could also reduce the dropout rate because interested parents who found one group inconvenient to attend could participate in another. Although the prevalence of unhelpful parenting practices is high,²⁷ some parents will not need this intervention because they have already developed the skills and attitudes required of helpful parenting. Conversely it is likely that some parents, particularly those living in the most stressful social circumstances, will never be reached by this type of intervention; and for this group, alternative, more intensive, tailored programmes may be required. Thus a goal of 100% attendance is likely to be both unrealistic and unnecessarily high.

Allocation to group was achieved by block rather than individual randomisation. This method, although possibly inferior from a statistical point of view, was considered important from the point of view of this trial. Health promoting interventions such as this are different from clinical interventions. In the latter, parents who have consulted a doctor about their child’s behaviour can be expected to accept whatever treatment is offered and to make time to attend a group. Our trial invited parents who did not necessarily recognise that they had a problem to attend a group. In this situation it is more important to ensure that invitations take their preferences for attendance (for example, time of day and day of week) into account. In addition the randomisation process needed to deliver groups of 10 –12 parents who could attend on the same day and time. The stratification process ensured an even distribution of potential confounding variables between the control and intervention groups.

Only 57% of those randomised to the intervention attended, but the trial had a good level of follow up, with responses from 79% of original participants at six months. The analyses were completed on an intention to treat basis so that responses from non-attenders were included in the intervention group, along with those of parents who attended the groups. This approach avoids bias attributable to unidentified differences between attenders and non-attenders and tends to an underestimate of the effect which would have been achieved had all those randomised into the intervention group actually attended.

The improvements we observed in control group children’s scores on the main outcome measure (ECBI) are potentially attributable to regression to the mean. We selected children for invitation to the trial on the basis of their above average scores on this measure. The ECBI, however, also includes items such as “wets the bed” which tend to resolve as the child gets older. Comparison with control group changes in SDQ scores, which do not include such items, suggests that some of the control group improvement we observed is likely to be attributable to a maturation effect. Unlike the ECBI improvement, the improvement in SDQ scores in control group children did not reach significance. On the other hand the SDQ questionnaire is intended for children over the age of 4 years, and some of the children in the trial were aged 3. We ran all the analyses on the SDQ excluding children under 4 years of age at recruitment, and the results remained as reported.

Our observation that the improvement in both the ECBI intensity score, a measure based primarily on problem behaviours, and the SDQ (conduct) scores was significantly greater in the intervention than the control group provides confidence that the intervention was effective, at least as far as these aspects of children’s mental health was concerned. The fact that the decline in scores on both measures in the intervention group was maintained over the six months following the intervention suggests that the intervention had an enduring effect on behaviour. It also suggests that the changes are unlikely to be caused by an immediate “feel good” factor attributable to attending a supportive group, or to the impact of lack of blinding of participants.

Although the changes shown in table 2⇑ are suggestive of a beneficial impact on other aspects of children’s mental health, the evidence here is not robust. When the small changes in the control group were taken into account there was no significant differences between groups. The Webster-Stratton Parents and Children Series is predominantly a behaviour management programme and it is therefore to some extent reassuring that it was this aspect of children’s mental health which was most effected by the intervention. It would appear that there is a need to develop alternative or additional strategies or programmes to improve anxiety, hyperactivity, and prosocial behaviour.

Improvements observed in the mental health scores of mothers in the intervention group were also, with the exception of social dysfunction at immediate follow up, not statistically robust. They are reassuring to the extent that mothers’ mental health did not suffer as a result of the intervention and, in terms of social dysfunction, the effect was beneficial. The latter finding could, however, be attributable, not to any specific effect of the programme, but to the “feel good” factor referred to above. The Webster-Stratton Parents and Children Series, in contrast to some other parenting programmes, does not focus on parents’ mental health and ways in which it could be improved. Any effect on parents’ mental health would therefore be secondary to improvements in the parent-child relationship or to the non-specific effect of group support.

The trial findings on parents’ mental health contrast with those obtained in a parallel qualitative study in which participants in this trial reported feeling more in control, with greater ability to cope with their children and a better relationship with them as a result of the intervention.²⁸ It is possible that such changes were not detected in the trial because the instruments used are not particularly sensitive to improvement in positive mental health.

Self report measures were selected owing to the resource intensive nature of independent observations, which were not practical for the purposes of this study. Although all measures of child problem behaviour were based on parental reports, these were all valid and reliable measures, and the specificity of the beneficial results to the predicted scales argues against the results being attributed to biased positive reporting by parents with elevated mood or improved social support as a result of the programme. In addition, a number of studies have shown adequate correlation between the results from parent reports and independent observations.¹⁶

Parents were invited to attend this trial if their children’s scores on the ECBI were above average. Although parents of children whose scores fell in the clinical range were slightly more likely to accept the invitation than parents whose children’s scores were normal, both groups were well represented in the trial, and our results suggest that both groups benefited. These results lend support to the feasibility of a population approach to the promotion of mental health through parenting programmes. They are consistent with results from an observational study,¹⁹ suggesting that the parents of children most in need are more likely to accept an invitation to a group than parents of other children, and that these two groups can derive benefit from attending the same group together. There remains a need, which this trial cannot fulfil, of showing improvements in positive mental health as a result of parenting programmes among children whose mental health is already above average. In order to establish the benefits of such a truly population approach, a much larger trial would be needed, powered to detect small changes. The conduct of such a trial would be greatly facilitated by the development of outcome measures which captured the full range of the components of positive mental health.