ACEs and adult outcomes: association or causality?

In the studies by Felitti et al. (Reference Felitti, Anda, Nordenberg, Williamson, Spitz, Valerie, Koss and Marks1998) and Bellis et al. (Reference Bellis, Hughes, Leckenby, Perkins and Lowey2014a, Reference Bellis, Lowey, Leckenby, Hughes and Harrison2014b, Reference Bellis, Hughes, Leckenby, Hardcastle, Perkins and Lowey2014c), notions of association, causality and impact were used interchangeably. The relationship between ‘childhood exposures’ and ‘risk factors’ in adulthood was referred to as an ‘association’ and ‘cause’. Felitti and colleagues accepted that if ACEs contribute to the development of risk factors in adulthood ‘then these childhood exposures should be recognized as the basic causes of morbidity and mortality in adult life’ (Reference Felitti, Anda, Nordenberg, Williamson, Spitz, Valerie, Koss and Marks1998: 246). They considered health behaviours and lifestyle factors to be the leading causes of morbidity and mortality in the United States and extrapolated that if ACEs contribute to these factors then ACEs should be recognised as the ‘basic causes’ of morbidity in adult life. At the same time, the authors acknowledged that ACEs data were based on retrospective self-reports that could only demonstrate associations between ACEs and health risk behaviours in adulthood. As such, it cannot be deduced that ‘the impact of these adverse childhood experiences on adult health status is strong and cumulative’ (Reference Felitti, Anda, Nordenberg, Williamson, Spitz, Valerie, Koss and Marks1998: 251), considering that other risk (e.g. poverty) and protective (extended family support, community ties) factors may explain this association.

Similarly, the findings from the studies by Bellis and colleagues cannot support claims about ‘links between such behaviours and childhood circumstances are likely to operate through the impact of ACEs on the developing brain. Thus, early life trauma can lead to structural and functional changes in the brain and its stress regulatory systems, which affect factors such as emotional regulation and fear response’ (Reference Bellis, Hughes, Leckenby, Hardcastle, Perkins and Lowey2014c: 9). We cannot infer causality and state that ACEs ‘can lead to structural and functional changes in the brain’ considering that the research design was not experimental and child brain functioning was not examined. Also, identifying a control group (adults with 0 ACEs) retrospectively rather than at the time of experiencing risk and abuse as children is problematic when making causal inferences. Finally, to establish causality, the order of the occurrence of events matters. Thus, retrospective recall data have substantial limitations due to possible inaccuracies in the recall of time sequences (Hardt and Rutter, Reference Hardt and Rutter2004).

Cross-sectional designs and retrospective recall

The foundational studies by Felitti and Bellis and their colleagues are cross‐sectional relying on between‐group comparison rather than within‐individual change to test retrospectively recalled ACEs and adult health outcomes. Cross-section designs with retrospective recall are vulnerable to possible biases associated with attrition as well as recall biases that might be relevant to the causal hypothesis being tested (Hardt and Rutter, Reference Hardt and Rutter2004). Longitudinal designs are more likely to be valid because they examine both within‐individual change and case‐control comparisons through the collection of multiple data over time to deal effectively with missing data. They also separate person effects on the environment from environmental effects on the person and assess attrition bias (Hardt and Rutter, Reference Hardt and Rutter2004).

The reliability and validity of retrospective recall of a traumatic experience could also be threatened by inconsistencies in early memories; awareness of the experience at the time of its occurrence; capacity to recall time sequences (e.g. whether an event occurred before or after a traumatic experience); infantile amnesia during the first two or three years or so of life (Lewis, Reference Lewis1995); the degree to which an experience can be objectively verifiable; the frequency of recall and memory consolidation of an event; and individual’s mood / personality style or presence of disease at the time of retrospective reporting (Lewinsohn and Rosenbaum, Reference Lewinsohn and Rosenbaum1987; McFarland and Buehler, Reference McFarland and Buehler1998). Ferguson et al. (Reference Ferguson, Horwood and Woodward2000) examined the temporal stability of retrospective reports of physical and sexual abuse by asking 1,000 young adults at the ages of eighteen and twenty-one years. The consistency of recalling having been abused was low (correlation coefficient about .50) whereas the recall consistency of not having been abused was high (correlation coefficient about .95). In other words, agreement was higher for events that did not happen than for events that did.

Retrospective reports of adverse experiences are sufficiently valid when applied to experiences and events objectively verifiable such as the death of parent, divorce or homelessness. The recall of subtle aspects of child–parent interactions (e.g. emotional closeness to parent) or experiences that are culture specific (e.g. corporal punishment, displays of parental affection) have not been found to have satisfactory validity (Hardt and Rutter, Reference Hardt and Rutter2004; Offer et al., Reference Offer, Kaiz, Howard and Bennett2000). In a study by Offer and colleagues (Reference Offer, Kaiz, Howard and Bennett2000), accounts of parent-child interactions and family life (e.g. questions such as ‘Was the discipline you received unfair?’ ‘Was love withheld as a punishment?’) at age fourteen showed little agreement with those retrospectively recalled at age forty-eight.

Using both prospective and retrospective measures of ACEs in a longitudinal study, Rueben et al. (Reference Rueben, Moffitt, Caspi, Belsky, Harrington, Schroeder, Hogan, Ramrakha, Poulton and Panese2016) found that, in comparison to prospective ACE measures, retrospective ACE measures may downplay the influence of negative childhood experiences on ‘objective’ adult outcomes and exaggerate it on ‘subjective’ outcomes. This is consistent with criticisms of the validity of retrospective recall of negative childhood experiences, in that mental health problems in adulthood have been found to associate with retrospective self‐reports than with other sources of information about childhood adversity. The foundational studies by Felitti and Bellis and their colleagues relied only on adults’ retrospective recall of ACEs without including any prospective measures (e.g. reports from siblings, official records from schools, hospitals) for comparison. Cross‐sectional studies with retrospective recall are limited in their examination of long-term impact of abuse and household dysfunction on adult outcomes (Hardt and Rutter, Reference Hardt and Rutter2004).

Regression analyses, comparable effects and interpretations

In the foundational studies, ACEs are assumed to have comparable effects although their origin and context may be distinctly different. As psychological constructs, they are multi-dimensional which leads to greater measurement error. Furthermore, certain ACEs – for example, physical, psychological or sexual abuse – are likely to strongly correlate with each other. In regression analyses, this refers to multicollinearity, which makes it difficult to separate the individual contribution of each ACE to adult outcomes.

Furthermore, a number of binary regressions were run examining associations between dichotomous predictors (i.e. Had/caused unintentional pregnancy < eighteen years; had sex < sixteen years with Yes/ No responses) and outcome variables (in Bellis et al., Reference Bellis, Lowey, Leckenby, Hughes and Harrison2014b). These two dichotomous variables could be thought of as dimensions of the construct ‘sexual behaviour’ so they could have been analysed within a multinomial logistic regression simultaneously instead of running separate binary logistic regressions: which are likely to increase Type 1 error, and falsely uncover statistically significant relationships between ACEs and outcome variables. Also, in the studies by Bellis et al. (Reference Bellis, Lowey, Leckenby, Hughes and Harrison2014b), logistic regressions were run with essentially one predictor variable: ACE with five categories (0,1,2,3,4+). It is not clear why a regression analysis was used if the intent was to examine bivariate associations – the idea behind regression is to examine multiple predictor variables (that do not correlate strongly with each other) to delineate their unique and cumulative contribution to the outcome variable. Finally, the analyses did not include different types of ACEs (only ACE counts) to examine their relative individual contribution, nor any protective factors, which affect not only recall but also how adverse the experience was understood to be in the first place.

In interpreting regression results, a distinction should be made between the odds for occurrence and actual occurrence of an event. Odd ratios show an increased or decreased likelihood of an event happening. Interpreting the likelihood of risk as actual behaviour is erroneous. Likewise, a slippage of language is observed when impact and odd ratios are discussed interchangeably. Impact should be examined within randomised experiments only whereby confounding variables are controlled through random assignment of persons to experimental and control groups.

ACEs, inequality and disadvantage

The foundational papers claim a unidirectional causal trajectory of ACEs and reduced social mobility in families. Bellis and colleagues argued that ACEs pose obstacles to social movement and contribute to intergenerational disadvantage, stating ‘those with 4+ ACEs were more likely to live in deprived areas, be unemployed/on long-term sickness and have no qualifications’. However, considering the study’s correlational design, we cannot deduce that ACEs cause poverty, disadvantage and reduced social mobility – in that a causal relationship between ACEs and adult deprivation and poverty was not examined. Rather, the findings showed that individuals who live in deprived areas were more likely to retrospectively recall four or more ACEs. Claims were also made about cycles of influence whereby individuals who reported ACEs were thought to ‘take sexual risks, become parents early and raise their children in environments where risks of ACEs are again high’ (Bellis et al., Reference Bellis, Lowey, Leckenby, Hughes and Harrison2014b: 5). Parents and families are held into account for ‘exposing their own children to ACEs’ – essentially being blamed for systemic inequality. The authors also stated ‘ACE counts correlate with worse health, criminal justice, employment and educational outcomes over the life course. The impacts of ACEs on criminality, violence, early unplanned pregnancy and retention in poverty means those with ACEs are more likely to propagate a cycle that exposes their own children to ACEs’ (Bellis et al., Reference Bellis, Lowey, Leckenby, Hughes and Harrison2014b: 5). The findings however cannot support that ACEs cause criminality, unplanned pregnancies, retention in poverty and intergenerational disadvantage.

Although Bellis and colleagues acknowledged that socioeconomic gradients are strong predictors of risk factors in adulthood, they consider childhood abuse and family dysfunction to be ‘additional factors’ to ‘explain the resilience and susceptibility of individuals’ (Reference Bellis, Lowey, Leckenby, Hughes and Harrison2014b: 82). Likewise, in the study by Felitti et al., the effects of race and educational outcomes were ‘accounted for’ statistically. However, measures of childhood abuse and household dysfunction are not simply ‘additional factors’ but mainly manifestations of poverty, deprivation and gender inequality; thus untangling their separate contributions to adult outcomes is nearly impossible (even if statistically these factors are ‘accounted for’). Although the foundational papers accept strong associations between socioeconomic and risk factors, they ‘adjusted for’ rather than examined their relative contribution (Hartas, Reference Hartas2017).

The assumed causal trajectory of ACEs and negative adult outcomes requires rethinking, especially in light of strong evidence on the ‘biological embedding’ of childhood poverty and the durability of socioeconomic differences in health and wellbeing across the life course (Elgar et al., Reference Elgar, Gariepy, Torsheim and Currie2016; RCPCH, 2017). The Royal College of Paediatrics and Child Health found that certain family and social indicators such as poverty impact upon children’s health (2017). Consistently, in the US, Biglan and colleagues (Reference Biglan, Van Ryzin and Hawkins2017) showed that societal conditions such as increasing financial inequality marked by a steep decline in the proportion of families on middle class income contribute to ACEs. Socioeconomic adversity in childhood has been found to associate with adverse health, developmental, cognitive, educational and long-term social and behavioural outcomes in adulthood (Dugravot et al., Reference Dugravot, Gueguen, Kivimaki, Vahtera, Shipley, Marmot and Singh-Manoux2009; Fors et al., Reference Fors, Lennartsson and Lundberg2009; Horvat et al., Reference Horvat, Richards, Malyutina, Pajak, Kubinova, Tamosiunas, Pikhart, Peasey, Marmot and Bobak2014; Lyu and Burr, Reference Lyu and Burr2016). These findings are particularly relevant considering that, currently, one in five children in the UK live in conditions of poverty and this figure is projected to rise (Hood and Waters, Reference Hood and Waters2017).

ACEs and policy implications

In public policy ACEs are invoked to indicate the possibility of risk in families and the need for intervention at an individual level. Increasingly, public health advocates accept childhood adversity to be a ‘hidden health crisis’ with ‘far‐reaching consequences’ (Center for Youth Wellness, 2014: 1). In the US, with 12.5 per cent of children thought to have experienced four or more ACEs, exposure to ACEs has become a major public health concern. Many states now monitor for childhood adversity among adults through an ACE module provided by the U.S. Centers for Disease Control in Behavioral Risk Factor Surveillance System surveys (Austin and Herrick, Reference Austin and Herrick2014).

With ACEs thought of as leading to poverty rather than being a manifestation or direct outcome of it, the foundational studies have contributed to shifting the policy focus on family intervention to support parents to ‘nurture’ their children as a means of countering the toxic effects of rising inequality. Bellis and colleagues advocated for better ways to ‘identify risky or maladaptive family conditions and intervene with families as needed’ and ‘helping families become less coercive and more nurturing’ (Reference Bellis, Hughes, Leckenby, Hardcastle, Perkins and Lowey2014c: 453) as if lack of nurturance is what underpins structural inequality. A disproportionate focus on ACEs without examining structural constraints in the form of poverty and disadvantage could mean that public policy is based on a narrow and potentially misleading evidence base.

ACE-driven policy trends highlight a disjuncture between the aims and scope of epidemiological research and public service provision. Epidemiological research focuses on understanding the mechanisms underlying chronic non-communicable diseases in adulthood at the level of population probabilities, whilst service provision works at the level of individuals. There is little doubt that poor adult health is intrinsically related to health inequality as well as to other forms of inequality, such as inequality of opportunity. If the aim of public policy is to prevent health inequality at a population level, the effects of childhood inequality should be examined and accounted for. Instead, policy priorities have shifted towards tackling the intergenerational transmission of ACEs via developing and testing family-level interventions that support resilience (seen as being affected by temperament, self-esteem, emotional regulation and trauma) and promote healthy childhood experiences.

On a positive note there is a growing acknowledgement of worse health and poorer educational outcomes in children who grow up in low-income families. The introduction to a recent special issue of Academic Pediatrics – a number of whose studies (Biglan et al., Reference Biglan, Van Ryzin and Hawkins2017; Bruner, Reference Bruner2017; Davis et al., Reference Davis, Costigan and Schubert2017) are cited in this article, concludes that 'to improve child health, there must be a reduction in health disparities by socioeconomic status, such as through community-building effort' (Davis et al., Reference Davis, Costigan and Schubert2017). Collectively these authors acknowledge the social context of childhood adversity shaped by poverty, racism and discrimination, arguing for community-level intervention to promote health and equalise opportunities. Davis and colleagues talked about the need to support children and families from the start to ‘build a culture of health: where everyone has an equal opportunity to be healthy, regardless of who they are, where they live, or how much money they have’ (Reference Davis, Costigan and Schubert2017: 4). This is a noble vision which, through public investment, could translate into steps towards reducing health inequalities seen as part of wider inequalities in children’s life that link to poor health outcomes and social, educational and economic gaps that only get intensified as children move into adulthood.