Association between parent-infant interactions in infancy and disruptive behaviour disorders at age seven: a nested, case–control ALSPAC study

The sample comprised participants from the Avon Longitudinal Study of Parents and Children (ALSPAC). ALSPAC is an ongoing population-based study investigating a wide range of environmental and other influences on the health and development of children. Pregnant women resident in the former Avon Health Authority in south-west England, having an estimated date of delivery between 1 April 1991 and 31 December 1992 were invited to take part, resulting in a ‘core’ cohort of 13,988 singletons/twins alive at 12 months of age [16]. The study website contains details of all the data that are available through a fully searchable data dictionary (http://​www.​bris.​ac.​uk/​alspac/​researchers/​data-access/​data-dictionary/​).

Ethical approval for the study was obtained from the ALSPAC Law and Ethics Committee and the Local Research Ethics Committees. All adult participants gave their informed consent prior to their inclusion in the study. A 10% sample of the ALSPAC cohort, known as the Children in Focus (CiF) group, attended clinics at the University of Bristol at various time intervals between 4 to 61 months of age. For the current study a sample was drawn from this sub sample of the core ALSPAC cohort of 1240 families (usually mother/infant dyads) who attended the ‘Children in Focus’ clinics when children were 12 months old. A range of measures was collected at the clinic including anthropometry, cognitive function, vision, speech and hearing. At the age of 12 months one of the sessions involved the Thorpe Interaction Measure (TIM) [17]. The TIM involves a carer (usually the mother) and her child looking at a picture book. Adults were asked to engage their child in this activity in the same way they would at home, stopping when the child lost interest. All interactions took place in the same ‘living room’ style environment in the clinics and were videotaped. The video recording was terminated if the child became distressed.

Sixty of these infants were later diagnosed with probable autism, conduct disorder, ADHD, anxiety or depression using the Development and Wellbeing Assessment (DAWBA) [18] which was included in a questionnaire sent to the parents of all children remaining in the cohort at 91 months (7.6 years) of age. In parallel, and completely independently the child’s teacher was asked to complete a questionnaire which included similar information. The DAWBA is a structured diagnostic assessment which relies on parental report as well as teacher reports, but final diagnoses are assigned by a child psychiatrist. More than one psychiatric diagnosis can be assigned, although pervasive development disorder (autism) precludes additional diagnoses. Numerous studies have shown the reliability of DAWBA expert diagnoses to be very satisfactory (i.e. [19, 20],). From the remaining non-case videos, 120 controls, group matched on gender, were randomly selected by the ALSPAC team. For this study we included 160 of these videos where the mother was identified as the lead caregiver; 54 cases and 106 controls. Including just the mothers removed the potentially confounding issue of which parent was present in the videos. Given that fathers were such a small number, conducting separate analysis for each parent would be statistically uninformative.

Wolke et al. [21] investigated whether attrition from the ALSPAC study was systematic or random. They found systematic participant drop-out according to the family variables (having a mother who was single; had no educational qualifications; financial difficulty experiences; being raised in large family where the mother smoked; had a poor relationship with the partner; lived in poor housing; had been involved in crime and been convicted or suffered psychopathology during pregnancy). Attrition did not however alter the association between family factors obtained in pregnancy and disruptive behaviour disorder at 7 years of age and we believe it unlikely that the direction of the associations we have investigated would be modified by the attrition.

The Mellow Parenting Observational System (MPOS) [22] was used to assess parent-infant interactions at one year (see Additional file 1). The observers were blind to case or control status. Measurements of the rate of positive interactions were moderately reliable with an inter-class correlation of 53%. Measurements of the rate of negative interactions had a correlation of 0.60 using Kendall’s τ [23]. Inter-rater reliability for the rate of total positive interactions was assessed using the interclass correlation coefficient. Due to non-normal distribution of the rate of negative interactions, we used Kendall’s τ to assess the agreement between raters. Kendall’s τ gauges concordance among the ranks, not the measures themselves, but we justify its use on the grounds that non-parametric measures of true reliability (i.e. concordance) are not available. Measures with τ > 0.6 were considered reliable. MPOS has previously been shown to have good inter-rater reliability [24]. Within the MPOS, the rater counts the number of occurrences of positive and negative interactions between parent and child. There are six Mellow Parenting dimensions, each rated for positive and negative interactions; Anticipation of Child’s Need; Autonomy; Control and Conflict; Cooperation; Distress; Responsiveness. The positive and negative poles of each dimension were summed to provide total positive and total negative interaction scores.

All 160 subjects had available interaction scores and were used in the analyses. Total positive and negative interaction scores were analysed as counts per minute of video material. We also examined whether video duration was associated with diagnostic outcome as some videos were stopped if the infant became distressed.

The interaction scores were used in predictive models of case and control status overall and within the following pre-defined sub-diagnostic groups; any ADHD; any emotional disorder; pervasive development disorder; disruptive behaviour disorder (DBD); any oppositional conduct disorder; conduct disorder alone; oppositional defiant and/or DBD-NOS; pure oppositional conduct disorder (Figure 1). Associations between interaction scores and psychiatric disorders were analysed using Firth’s penalized–likelihood logistic regression to correct for biases that may be induced by the low prevalence for some disorders [25]. To account for any potential confounding all models were adjusted for select parental and infant characteristics previously found to be associated with parent/infant interaction scores [24]. When positive interactions were the variable of interest the models were adjusted for the child’s gender, maternal age at birth, maternal educational attainment and pre-natal anxiety scores. Models with negative interaction scores were adjusted for maternal age at birth, maternal smoking status, the child’s gender and a social support score. Odds ratios (ORs), 95% confidence intervals and p-values are presented, with ORs reporting the effect associated with an increase of one interaction count per minute in the respective scores.

All analyses were carried out using R statistical package v2.15.

The camera recording the mother-infant interaction was placed in the upper corner of the room, so the faces of parents and children were often not visible. The video quality was relatively poor due to the age of the tapes, which may have contributed to the moderate reliability of the MPOS. Given the positive relationship between reliability and sensitivity, we might expect the use of more modern video equipment to substantially improve the sensitivity of the MPOS. Despite such limitations, we were able to confirm our hypothesis of an association between parenting behaviours and later development of conduct disorder. Our data have particular value because of their community base: previous studies have sampled high risk referred children or siblings of affected individuals and we have published a number of findings using the same dataset [32, 34‐36]. Also of note is that the predictor variables used in our analyses were based on videos recorded at one year of age, which were rated without knowledge of the future psychopathology of the child; the data can therefore be viewed as prospective, in contrast to previous studies which have been based on retrospective recall of predictive factors. There was no statistically significant relationship between negative parenting interactions and any of the diagnostic categories: a potential limitation of the present study is that negative interactions were rarely observed in this setting. We have performed numerous statistical tests, without adjustment for multiple comparisons, so there is the possibility of Type I error; however, overall case diagnoses was pre-defined as the primary outcome with the hierarchy of sub-diagnoses also defined in advance.. Given that significant associations were observed between total positive interactions and several sub-diagnoses, this adds to the robustness of our findings. The setting used for the TIM did not elicit many negative interactions: it was not devised specifically to study the type of negative interactions for which the MPOS is designed. Nevertheless, the MPOS did identify a large number of positive interactions, and substantial variability in these behaviours was demonstrated by the ALSPAC participants. Furthermore, if videos were available of settings specifically designed to elicit a range of positive and negative interactions, then this could enhance the reliability of the MPOS measure. Finally, some diagnoses, for example, pervasive developmental disorder, were represented by small numbers and the lack of associations may reflect Type II error. It is possible that lower positive parenting scores may represent an overall lower level of maternal activity: we have reported that less frequent maternal vocalisations [32] and lower levels of parental activity using a holistic measure [34] are associated with later psychopathology and we are currently exploring the inter-relationships between these predictors.

It was difficult to achieve high reliability on the MPOS in this setting, for which it was not uniquely designed. There are similar problems in applying other possible coding systems, for example the Care Index which has specific instructions based around three minutes of free play, rather than the constrained conditions of the TIM. Nevertheless MPOS has been shown to have utility and predictive or concurrent validity in other situations. For example, it distinguishes parent-child dyads with growth delay [37] and is sensitive to change, concurrently with maternal depressed mood [22]. Though not as well supported by reliability and validity studies as would be ideal, it was therefore chosen as a tool. It effectively predicts DBDs in this study, some six years after the original coding, supporting the assertion that is a useful system, albeit in need of further refinement for future use.

It is possible that observational assessment may have greater utility in more naturalistic social interactions than those studied here. Further community-based longitudinal studies of the predictive validity of the MPOS in different types of social interaction, for example feeding, nappy changing or free play are indicated. Situations where parenting skills are challenged by the task and thus negative interactions more likely to be observed may have additional value in predicting the onset of disruptive behaviour disorders.