Using latent class analysis in a cohort linking GP records and research data we identified a substantial group of mothers with a high likelihood of potentially persistent CMD with no record of treatment in primary care. Compared to children of women who were unlikely to have CMD, the 3-year-old children of women with untreated CMD were more likely to be rated as having socio-emotional and behavioural difficulties. Little difference was observed between these difficulties in children whose mothers were treated and those whose mothers were not.
Strengths and limitations
Women enrolled in BiB1000 were broadly representative of the maternal population of Bradford at the time of recruitment, and although there were some demographic differences between our analysed sample and those excluded due to missing data, there appeared to be little variation in mental health between analysed and excluded women, or indeed in their child’s pattern of behaviour. Therefore, we do not think that missing self-reported data have significantly affected our results or distorted our conclusions. The quantity of missing data from the primary care dataset, however, is unknown; for example CMD treatment noted in a free-text field or a referral letter (only) was not picked up. The tight geographic focus and relatively short (9 months) recruitment period are strengths in that they minimise potential regional and temporal variation in GP coding practice. However, while Bradford may be representative of other ethnically and socio-economically diverse UK cities, our findings may have limited generalisability to other settings.
Our approach sought to minimise the effect of cultural variation in the self-reporting of distress/difficulties by the mother, both for herself and her child. Although numbers were too small in this study to explore differences in latent classification by ethnicity, the association between latent classification and SDQ outcomes appeared robust to variation by ethnic group. We use the term ‘untreated’ as a label for the group of women classified as having an increased likelihood of self-reported distress scores over the 75th centile across three time points without visible note of treatment in the electronic primary care record, and not as an established clinical diagnosis. Our findings may be distorted if we misclassified women who were actually treated for CMD as ‘untreated’, for example if their treatment was noted in a free-text field or letter to which we did not have access, or if the methods we used tended to incorrectly classify women with transient non-pathological distress, or treatment in one particular period, as having untreated CMD (false-positives). Due to small numbers we did not distinguish between types of treatment, and we were unable to distinguish between offer and uptake of treatment, or between failed and successful treatment, which may have introduced heterogeneity in regards to the children’s outcomes. Compromised precision in our analysis for disorder and treatment classification might be a factor in the observed lack of difference between children’s outcomes by maternal treatment status.
Common mental disorders are chronic, relapsing conditions, including over the maternal period [
37‐
39]. This was one of the reasons we chose to model chronicity of mother’s mental health rather than permitting classification on discrete risks in each period. We did not, therefore, capture any dynamic or transitional variation over time, for example trying to characterise women who had either a prenatal or postnatal CMD episode, but not both. That we classified nearly five times more women as ‘untreated’ than ‘treated’ indicates our results may have been in favour of classifying women as ‘untreated’ rather than as ‘treated’. A feature of latent class analysis is that it allows for sample-specific flexibility in modelling approach; some studies using latent class methods to model both trajectory and severity of self-reported symptoms over the maternal period have found that their data were best characterised by periods of discrete risk [
40], but others have reported that chronicity models best fit their data [
5,
7,
41]. The main driver for predefining a chronicity model was to provide a platform against which we could distinguish group membership based on treatment of CMD in primary care. This approach has merit as a method for estimating risks given that CMDs tend to be chronic and relapse rates are high, the outcomes of treatment in any maternal period are not well studied and effects are modest at best [
42‐
44]. In using a chronicity model and a centile-based threshold approach for the self-reported measures, however, we did not account for potential variation in severity of disorder, which may have implications for treatment offered, or take up [
45], or effect on child outcomes. Variation in categorised distress could have also been introduced by alternation of ascertainment by GHQ-28 and Kessler-6 at follow-up, which was beyond our control, although both assess symptoms of psychological distress and our use of a centile threshold may have helped to minimise any effect. Due to small numbers of treated women and uncertainty about timing of incident diagnosis and continuity of treatment events we were unable to unpick whether treatment was sufficient to be effective. Future studies with data linkage for greater temporal ranges and in larger populations are required to answer specific questions on the effectiveness of specific treatment regimes.
We employed a robust psychometric approach to analysing SDQ data, generating scores for each child relative to other children in the sample that control for measurement error through use of an appropriate latent variable model approach. Contrary to other findings in pre-school community samples [
20,
46], we were unable to establish an acceptable baseline psychometric model for the theorised structure of the SDQ in this multi-ethnic sample and suggest further validation studies to establish population-based concepts of problem behaviour in such young children. We replicated the factor-score regression findings using the single-item Perceived Difficulties question, but this item has been shown to discriminate between clinical and community samples using the ‘severe’ and ‘serious’ categories [
24], not as we used it, within a community sample including ‘minor difficulties’. As with all studies that ask parents to assess their child’s behaviour, our results may be distorted by the effect of more distressed mothers being potentially more likely to rate their child’s behaviour as problematic [
47‐
49]. Future linkage of this sample with routinely collected school attainment data and teacher-rated behaviour may help overcome some of these measurement limitations and establish whether our findings related to untreated CMD are confirmed by more objectively measured outcomes.
Research findings in context
To the best of our knowledge, this study is the first to attempt to quantify the association between unrecognised maternal mental health problems and child behaviour outcomes, and we found a small, but significant association. Although previous research has found increased levels of child behavioural problems associated with maternal mental health difficulties [
1‐
5,
7,
50], these studies have not accounted for treatment of disorder, which makes direct comparison with our study difficult. We can, however, use data from these studies to help interpret our findings. Goodman et al. [
51] estimated a weighted
r of 0.24 (95 % CI; 0.22, 0.26) in a meta-analysis of 39 observational studies reporting the association between maternal depression and children’s general psychopathology (grand mean age 7). An older review estimated a weighted
r of 0.26 and weighted
d of 0.53 for the association between depressed mothers (excluding studies of postnatal depression) and behaviour problems in pre-school children from 11 studies [
52]. These average effects are ‘moderate’ in size [
53], and are larger than the difference in SDQ scores in our study between children of mothers without CMD and those in either the treated (
d = 0.27,
r = 0.07) or untreated (
d = 0.32,
r = 0.14) groups. This is particularly noteworthy, as effect sizes have been reported to be larger in studies of low income families and where the children were assessed by their mothers, as in BiB [
51]. Aside from differences in setting and population, there are several potential explanations for the smaller observed effects in our study.
First, we categorised women in the ‘no CMD’ group as unlikely to have persistent CMD; however, this group may have contained some women with a disorder (treated or untreated) in a single, potentially sensitive, period. If this were the case, affected children may have skewed the SDQ scores of the low-risk group upwards, reducing the observed effect size relative to studies that screened or diagnosed to classify maternal disorder. Second, children of the treated mothers in BiB may have less psychopathology due to successful maternal treatment of CMD. A meta-analysis of mental health outcomes in infants and very young children after maternal psychological treatment for depression in five small trials found a small to moderate pooled effect size (
g, broadly equivalent to
d) of 0.40 (0.21, 0.59) between the treated and control groups [
11], indicating that, although long-term effects are unclear, maternal treatments have the potential to have positive effects for children in the short term. If the SDQ scores of BiB children have been moderated by maternal treatment, i.e. their SDQ scores would have been higher had the mothers not been treated, then this raises another question; why are not the SDQ scores in the untreated group higher than the treated group? A plausible explanation is that untreated women had less severe symptoms with less impact on their children. While this might explain the difference in magnitude of our results compared to other studies, even low level or subclinical maternal symptoms can have observable relationships with children’s outcomes [
5,
7]. A third explanation for the smaller effects observed in our study is that women who were distressed (in the treated and/or untreated groups) rated their children as having fewer problems on the SDQ for actual level of child problems, relative to mothers without CMD. We think this unlikely as maternal distress has generally been associated with higher child problematic behaviour ratings [
47,
48]; however, differential effects may have distorted our findings [
49]. Fourth, we examined treatment for anxiety, depression, mixed disorders and symptomology, not just depression, which was the subject of the meta-analyses. As with depression, anxiety disorders have detrimental effects on children [
54]; however, differential effects have been noted which may have affected our results if women with one or other of these disorders were clustered in the treated or untreated groups [
55].
Much of our health, and mental health, is socially patterned and influenced [
56,
57], and previous analyses of the larger BiB cohort and other research datasets confirm that women living in socio-economic disadvantage are more likely to be distressed, or potentially unidentified as being distressed [
10,
30,
41,
58]. Severity of child behaviour problems similarly appears to follow socio-economic gradients, whether rated by parents or teachers [
59]. We found that a measure of socio-economic disadvantage attenuated but did not remove the increased likelihood of higher SDQ scores in children of women with untreated CMD compared to scores of children whose mothers did not have CMD. This indicates that our findings are not confounded by socio-economic status, although residual confounding by ethnicity is a possibility that we could not explore due to small sample size. In practice, social and economic disadvantage are likely to exert rather complex and variable effects on whether a women with CMD is identified as such, and mitigate or exacerbate subsequent effects of distress on her family.
Implications for research and policy
Our study is unique in that we attempted to estimate the magnitude of behaviour problems displayed by children of women who are likely to be potentially unidentified as distressed and, therefore, untreated, by the health service. The size of the association did not appear to vary between treated and untreated women. As discussed above, this could be interpreted, broadly, as indicating that currently offered treatment in this cohort might not be successful in mitigating the effects of maternal anxiety and depression on children, or that currently offered treatment has mediated but not remedied the effect on children, or that untreated women have less severe symptoms which have less impact on their children.
There is thus far only a limited evidence base for treatment effectiveness in trials examining outcomes on mothers [
42‐
44], and even less for trials examining the effect of maternal, partnership or family treatment on children’s outcomes [
11,
60]. Robust trials that test interventions to improve mother and baby interaction and responsiveness as early markers of difficulties with longitudinal follow-up are needed with accurate assessment of the interaction problem with targeted intervention [
45,
61]. Studies that examine the specific effects of anxiety and/or depression on children’s functioning and behaviours are also needed. We suggest that the development of treatment innovation and improved access to psychological therapy should be complemented with more effort and resource applied to identify anxiety and depression in maternal women as a precursor to getting them into treatment, and the implementation of any successful treatment programme into routine health services should be subject to a rigorous inequality evaluation.
In previous analyses of the larger BiB cohort, we have noted ethnic disparities in both the identification and treatment of maternal CMD [
10,
62], of which the absolute and relative distal effects on children need to be explored in a larger dataset. A greater understanding of why some women who are experiencing persistent difficulties are not identified as such in primary care is needed.
It is possible, and likely, that in our study, women in the untreated group had less severe symptoms or transient distress that may not meet current criteria for treatment. If this is the case, our study confirms previously reported research about the effect of sub-threshold disorder and persistent low-level symptoms on children’s outcomes [
5,
7]. More research attention is needed in this area such as trials of treatment for women with less severe symptoms that include long-term follow-up of their children.
Implications for practice
It is now widely accepted that, in addition to depressive problems, anxiety disorders in the maternal period cause significant morbidity [
45,
63,
64]. Recently updated UK guidance for managing maternal mental health advises clinicians to consider screening for anxiety as well as depression at each contact, consider the needs of vulnerable women, and to be aware that some women may be unwilling to disclose their distress [
45]. Treatment staging advice and maximum time-to-treatment targets are clearly specified [
45]. There is, however, little focused advice on how to ensure that screening for any CMD is as successful in identifying cases among disadvantaged groups, or ethnic minority women, as more advantaged, or majority populations, or ensuring equitable treatment outcomes. Clinical reality in overstretched health services means that screening, referrals and treatment availability for mental health in maternal or primary care may be sub-optimal, highlighting the gap between evidence-based advice and practice [
65]. Primary and community care staff need help to identify patients who may not, on first appearances, appear to be vulnerable, to translate population or community-level socio-demographic information into individual risk assessments for poor outcomes, and to understand the impact of health inequalities on their patients [
66].