Children respond differentially to parenting though with genetic and prenatal environmental factors contributing to an increased sensitivity to the environment called postnatal plasticity (15). While Belsky documented that this plasticity (measured as temperament) emerges from genetic factors (differential susceptibility) (16, 17), and Boyce and Ellis posited that this susceptibility would be rather environmentally induced, (16), both show that plasticity factors influence how individuals interact with the environment. Three patterns of environmental sensitivity have been described in the literature: diathesis-stress, differential susceptibility, and vantage sensitivity (17, 18). In diathesis-stress, a biological marker represents a disadvantage in unfavourable environments in that outcomes for carriers of that marker can only be approaching the outcome levels of noncarriers if the individual is exposed to average or advantageous environments. In vantage sensitivity, the opposite is true: the biological marker represents an advantage over noncarriers, such that in unfavorable environments carriers and noncarriers develop similarly, while carriers show increasingly better outcomes as the environment becomes more advantageous. Finally, in differential susceptibility, highly susceptible children are more responsive to both adverse and supportive environments than non-susceptible children for better or worse (15).

Surprisingly, few studies have examined the associations between genetic risk, prenatal adversity, and maternal care in predicting child psychological functioning, even though the quality of early parental care can be a crucial mitigating factor of the effect of prenatal environmental or genetic risk (19). In a three-way interaction model, we found that maternal looking away behaviour (negatively correlated with maternal sensitivity) moderates the risk associated with prenatal depression and the 5-HTTLPR genotype to predict depressive symptoms at 18 months, but not at 24 months (20). Similarly, we have reported that maternal looking away behaviour moderates the developmental risk from low birthweight and DRD4 to predict disorganized attachment (21). These findings suggest that, in children with genetic and prenatal risk, the risk for psychopathology was attenuated when the mother looked away less frequently. These findings are in line with previous studies reporting that the frequency of self-reported maternal stroking during early infancy moderated the effect of pregnancy anxiety on internalizing problems when the children were 3.5 years of age (22). For mothers who experienced high levels of pregnancy-specific anxiety, high levels of postnatal stroking were related to lower internalizing scores in their children. Similar results of a moderating role of maternal stroking on child internalizing problems have also been reported for prenatal maternal depression (23) and general anxiety (24). Finally, in our recent study in the Maternal Adversity, Vulnerability and Neurodevelopment (MAVAN) sample, we found evidence for the presence of two-way interaction effects on toddler attention function, namely that positive maternal behaviors observed during mother-child interactions at 6 months postpartum mitigated the effects of both prenatal adversity and dopaminergic polygenic risk on toddler attention function (25). However, our sample was limited to find a significant three-way interaction between prenatal adversity, dopaminergic risk, and parenting behavior, and the above two-way interaction effects need to be replicated in independent samples.

Fresh perspectives in understanding the complexities of the parent-child dynamics are also offered by behavioral epigenetic studies, which posit that the quality of maternal care sets epigenetic processes (e.g., DNA methylation) in motion that may ultimately affect offspring psychological development through modifying expression of genes involved in behavioral and stress regulation (e.g., NR3C1, BDNF, OXTR) (26). For example, harsh parenting contributes to similar epigenetic modifications in the child as early adversity, potentially affecting cognitive and socioemotional development in childhood (27, 28) and attachment style in adulthood (29). Importantly, epigenetic modifications are also affected by positive parent-child interactions (30, 31), translating into “positive” epigenetic mechanisms, which may act as a protective mechanism in the face of adversity-related increased DNA methylation of genes involved in behavioral and stress regulation (32, 33).

The replication crisis in science (34) is exemplified in psychological research, where the replication rate of key experiments is just above 10% (35). One reason for the lack of replication is that cohorts do not always assess the same developmental constructs, and, even when they do, they often use different measures to assess them. In addition, without the initial registration of research hypotheses and analytic plans (e.g., in the Open Science Framework) of observational studies, akin to that found with randomized controlled trials (e.g., www.clinicaltrials.gov), it is not often clear which findings (a priori vs. post hoc) are most likely to be replicated. For clinical trials, there are broadly endorsed initiatives of prospective harmonization of outcome measures, such as the COMET initiative (https://www.comet-initiative.org/) and the CROWN initiative (http://www.crown-initiative.org/). For already existing observational studies, the retrospective harmonization of relevant predictor and outcome variables across cohorts with similar measures may prove essential in producing replicable and generalizable research findings. As well, pre-registering planned correlational analyses in intentional initiatives to replicate findings will make for more convincing results.

Gathering detailed observational data on parent-child interaction in large epidemiological cohorts is costly and unfeasible. However, complex developmental models that account for the interplay of genetic and pre- and postnatal environmental influences are incomplete without including precise measures on the quality of parental care. Measurement error can reduce statistical power for detecting true interaction effects in complex developmental models, as it inflates the variance of the estimate of the interaction term, similarly to multi-collinearity and non-normal distribution of the interaction terms (36–38). Observational measures of mother-child interactions are thus strongly preferred to self-report measures of parenting in studying complex developmental models in longitudinal cohorts. Observational measures, due to their complexity and cost, restrict potential sizes of epidemiological developmental cohorts. Thus, the need for larger sample sizes and valid cross-study comparisons, has led to increased interest in co-analyzing already existing data across studies. However, heterogeneity in study design and measures collected limit our capacity to easily compare or integrate data across studies (maelstrom-research.org) (39).

To be truly informative, birth cohorts, particularly those with genetic data, require large samples to test complex computational models of developmental trajectories (40, 41). While harmonization of key predictor and outcome variables across multiple birth cohort studies will greatly assist in overcoming the replication problem, a priori or post-hoc harmonization of parenting measures will also help overcome the problem of small sample sizes, that are typical of focus cohorts of large epidemiological samples, where observational measures of parenting are available (42). Combining focus cohorts from multiple large birth cohorts with harmonized parenting data can increase sample size to levels sufficient to conduct tests of complex models.

The present paper discusses the relevance of a key methodological concept (i.e., retrospective data harmonization), which can help mitigate some of the challenges inherent in replicating study findings involving observational parent-child interaction measures. More specifically, this paper offers two valuable approaches to researchers who are interested in retrospective harmonization and integration of parent-child interaction data across independent samples.

The first approach comes from our ongoing initiative, the DREAM BIG consortium, which performs cross-cohort retrospective data harmonization of key constructs relevant to probe complex models of child development (e.g., the prenatal environment, genetic susceptibility, child psychopathology, and early parental care). The second strategy, as used in the CATS-project, focuses on the initial stages of retrospective data harmonization of observed maternal sensitivity. This method first evaluates the theoretical constructs underlying the measures and then the measures themselves prior to the recoding of original values. The DREAM BIG and CATS approaches offer helpful analytical solutions for restructuring parent-child interaction data collected with different instruments across multiple studies to indicate a comparable construct.

This is a harmonized prenatal cumulative risk index derived from four major areas: stressful life events (i.e., death in family, accident, illness), contextual risks (i.e., poor housing conditions, financial problems), parental risks (i.e., alcohol and substance abuse, criminal involvement), and interpersonal risks (i.e., family conflict, domestic violence) (51, 52) using confirmatory factor analysis with a second-order hierarchical model.

This is a set of harmonized prenatal maternal psychological symptoms constructed using confirmatory factor analyses that identified a general prenatal affective symptoms factor and three specific factors: anxiety/depression; somatic symptoms; and pregnancy-specific anxieties across cohorts (50). These prenatal maternal affective symptoms factors predicted offspring psychopathology at age 4–8 years in a meta-analysis of three cohorts (50).

Both measures of adversity (i.e., prenatal social- environmental and maternal affective) build on previous successful harmonization initiatives between ALSPAC and Generation R (51, 52). DREAM BIG innovated by separating these two measures.

This includes a harmonized, age-adjusted, general psychopathology factor (P-factor), and specific uncorrelated internalizing and externalizing factors consistent with the Hierarchical Taxonomy of Psychopathology (HiTOP) model (53), constructed using psychopathology measures rated by different informants (parent, child, teacher) at multiple time points between the ages of 4 and 8 years (54). A HiTOP approach to harmonizing psychopathology addresses important methodological concerns about diagnostic co-morbidity, homotypic and heterotypic discontinuities, and rater differences (53, 55).

Polygenic scores (PGS) for internalizing (anxiety, depression), neurodevelopmental (ADHD, ASD), psychotic (schizophrenia, bipolar), and compulsive problems (anorexia nervosa, obsessive-compulsive, Tourette syndrome) will be computed in each cohort based on results of a Genomic Structural Equation Model (GenomicSEM) of 11 common psychiatric disorders using publicly available GWAS summary statistics (56). This approach models the structure of psychopathology at the genomic level and exploits genetic correlations between multiple psychiatric disorders modeled simultaneously. Further details about the GenomicSEM approach are available elsewhere (56).

Four of the six DREAM BIG cohorts have observational measures available on maternal sensitivity and parent-child interactions. In MAVAN, maternal care and mother-child interactions were observed during free play using the Parent-Child Early Relational Assessment (PCERA) (57) at 6, 18, 36 and 60 months, the Ainsworth Maternal Sensitivity Scales (6) at 6 and 18 months, and the Behavioral Evaluation Strategies and Taxonomies (BEST) (Educational Consulting, Inc. Florida, US; S & K NorPark Computer Design, Toronto) at 6 months. In Generation R, maternal sensitivity was observed during free play using two subscales of the Ainsworth Maternal Sensitivity Scales at 14 months (sensitivity and cooperation), and during two structured mother-child interaction tasks using the revised Erickson 7-point rating scales for supportive presence and intrusiveness (58) at 36 and 48 months in a subsample (n = 1,079). In GUSTO, maternal sensitivity was observed during free play using the Revised Mini-A short form of the Maternal Behavioral Q-Sort-V (Mini-MBQS-V) (59) at 6 months, and during a structured mother-child interaction using the Erickson 7-point rating scale at 54 months. Finally, in a subsample (N = 1,240) of the ALSPAC cohort, the Mellow Parenting Observational System (60) was used to code mother-child interactions during the Thorpe Interaction Measure (61) at 12 months.

Our current work in DREAM BIG will also harmonize the measures of observed maternal sensitivity across cohorts to test for the presence of replicable two-way and three-way interactions between prenatal adversity, child genetic susceptibility, and early maternal sensitivity on the development of child mental health problems. Integrating both harmonized parenting and child measures allows for the inclusion of complex questions assessing a wide range of well-defined observable parental care measures.