Strong evidence for socio-economic differences in cancer survival and mortality has been shown for many cancers and across populations 1 . In 2004, cancer was among the leading four causes of death in high income countries 2 3 , and its increase is typically associated with the third stage of epidemiological transition 4 . Cancer prevalence and mortality is likely to become increasingly predominant in high-income countries and by consequence so is their contribution to health inequalities.

The causes of cancer are complex and multifactorial occurring over a lengthy time. Explanations for mechanisms linking adverse events early in life and cancer have been provided from animal models 5 6 . In humans, early life exposure to Adverse Childhood Experiences (ACE), like trauma, abuse or maltreatment in childhood has been linked both ‘indirectly’, through tobacco and alcohol use or via ‘direct’ associations 7 to alterations of the brain structure and neurobiological stress-response systems which in turn have consequences for health and emotional well-being 8 . Studies have described associations between retrospectively collected ACE and health outcomes such as liver disease 9 , ischaemic heart disease 10 , obesity 11 12 , perceived health 13 and psychopathology 14 as well as premature mortality 15 16 . Regarding cancer, data are sparser and inconsistent 7 12 17 . One of the reasons for this inconsistency is that the ACE measure used is usually self-reported by adults asked about trauma and adversity they may have experienced during childhood. Such questions are inevitably vulnerable to various forms of recall bias 18 one example being that individuals’ memories of the past may be influenced by their later health 19 .

In the present study we aim to determine if exposure to ACE, determined using prospectively collected data, is linked with the onset of cancer in adulthood. The causes of cancers are indeed diverse, however our aim here is to understand the development of susceptibility to this disease at its earliest phase, at the common roots of all cancers 20 . To do this we take a lifecourse approach, considering the whole life span. We hypothesise that the physiological embedding of stress induced by adversity in childhood is linked with biological susceptibility as well as with the development of stress-reducing health behaviours favouring the development of cancers in adult life 21 22 . The objective of this study is to examine the relationship between psychosocial adversity in childhood and cancer after controlling for the effects of material disadvantage, health behaviours and education level using a large prospective cohort study.

Data are from the 1958 National Child Development Study (NCDS) which included all live births during one week in 1958 (n = 18558) in Great Britain. Subsequent data collections (sweeps) were carried out on cohort members aged 7, 11, 16, 23, 33, 42, 46 and 50. At the last sweep, carried out in 2008, 9790 individuals participated in the self-reported questionnaire and face to face interview, representing 53% of the original sample. The NCDS has been described in detail elsewhere 23 .

Descriptive statistics are presented in Table 1 for the subsample (n = 6138) who provided information on whether or not they had had a cancer in the last four data sweeps (ages 33, 43, 46 and 50). Among respondents who answered questions on cancer, 3.3% of men and 10.6% of women reported having had cancer by the age of 50. In total 444 self-reported cancers were identified in the cohort reported between the ages of 33 and 50, 80% being reported by women (n = 351). Based on the available information on cancer types reported by the cohort members (see Additional file 1), breast and cervical cancers were most often reported (19% and 15% respectively), followed by skin cancers (11%). Nevertheless, most cancers remained ‘undefined ‘(39%) The majority of men (74%) and women (75%) had had no adverse childhood experiences, based on the prospectively collected information at ages 7, 11 and 16 y. The distribution of ACEs was similar for both sexes, with 5.6% of men and 5.4% of women having two or more ACEs.

*Subsample of individuals who provided information used to create adversity and cancer variables.

Table 2 shows bivariate analyses between ACE, the covariates and cancer for men and women. In men, no relationship is apparent between ACE and cancer. The multivariate models for men (Table 3) show no association between early life socioeconomic variables or perinatal variables and cancer. ACE was not significantly associated with cancer, and neither were social variables at age 23 y, or smoking and alcohol age 23 y.

*p-values comparing ‘yes’ v ‘no’ cancers for each covariate.

In women (Table 2), a bivariate relationship between ACE and cancer was shown. The proportion of women reporting a cancer before 50 increased from 9.1% for those with no ACEs to 13.0% for those with one ACE (p = 0.004), and 23.0% for those with 2 + ACEs (p < 0.001). Both manual parental social class and overcrowding in the household at birth were significantly related to reporting a cancer before 50 y, and mother’s education was of borderline significance. Women who had experienced childhood pathologies were also more likely to have had a cancer than those without childhood pathologies (13.3% vs 9.9%, p = 0.016). Many of the female respondent’s adult variables were also related to having had a cancer before 50 y in the bivariate analyses. Low educational attainment, being inactive in terms of occupation, being a heavy alcohol drinker, being a smoker and having symptoms of depression/ anxiety, and having a first pregnancy after the age of 33 y, were also related to having had a cancer before 50.

The multivariate models for women (Table 4) show that. ACE was significantly associated with reporting cancer before the age of 50 after control for potential confounding variables (model 2). The graded association showed that women with one ACE had a 40% increase in the odds of reporting cancer (p = 0.016), and those with two or more ACEs had a 2.5 increase in the odds of reporting a cancer versus women with no ACEs (p < 0.001). When information on adult behaviours, social characteristics and psychological malaise was added to the model this association weakened, so that women with two or more ACEs had a 2.1 increase in the odds of reporting a cancer versus those with no ACEs (p < 0.001). Women who had their first pregnancy after the age of 33 y had a significantly increased risk of having cancer before 50 y (p < 0.001). Furthermore, women who were smokers at age 23 y had a 70% increase in the odds of having cancer before 50 versus non-smokers (p = 0.001), and women who were heavy drinkers at age 23 y also had an increased risk of having cancer before 50, however this result was of borderline statistical significance (p = 0.055). Having symptoms of depression/anxiety increased the odds of having a cancer before 50 by 35%, however this result was not significant at the 5% level (p = 0.1).

The main finding from this study was that psychosocial adversity in childhood was related to cancer incidence before 50 y among women, after adjusting for prior confounding factors and potential mediators, in a large prospective cohort. An accumulation of ACE remained a strong predictor of cancer in women, after taking important potential mediating factors at age 23 y into account, including smoking and drinking. Women who experienced two or more ACE doubled their risk of having a cancer before 50 relative to women who had had no childhood adversities. There was a tendency towards a graded association between childhood adversity and adult cancer across the ACE categories. These results make a significant contribution to demonstrating and understanding links between ACE and cancer because they use an a priori definition of adversity and identify ACE with prospectively collected data. The strength of the relationship between adversity and cancer was of the same magnitude as that observed between age at first pregnancy and cancer, a well-known risk factor for breast cancer 30 .

The two main weaknesses of this study are in the self-reported nature of cancer incidence, and the amount of missing data caused by attrition in the cohort study. These weaknesses are partly addressed by the conservative nature of our cancer variable, and by using estimates obtained from multiple imputations to account for the missing data. We conducted sensitivity analyses by running the model using imputed datasets, using a case–control sub-sample, and using the full cohort dataset without imputation, and found that the relationship between ACE and cancer incidence was stable. Comparisons were made between complete-case analyses and those run on estimates obtained by imputation. The models yielded similar results until the inclusion of variables at age 23 (model 3). The differences observed in the results for model 3 indicate selection bias in the complete case sample, where individuals who had experienced ACEs in childhood were more likely to have missing data at age 23 regarding health behaviours. The multiple imputation model therefore enables adjustment for this bias. We are confident, therefore, that our analyses show a ‘real’ association, the nature of which needs to be established in further more complex modelling of mediating factors.

The nature of the questions collecting information on cancer varied by data sweep, some being retrospective (have you ever had cancer) while other questions asked about suffering from an illness at the time of the survey. This means that there are information gaps, notably between the age of 33 and 42 where the cohort member could have had cancer and not report that they were suffering from it subsequently. However, this means that our cancer variable is most likely conservative. The validity of self-reported cancer is likely to vary based on the cancer type diagnosed, and tends towards an underreporting of cancers, given the high levels of specificity reported in studies comparing self-reports to registry data 31 . Indeed in studies where cancer registry data have been compared to self-reported information individuals tend to underreport rather than overreport cancer history, and variation in inaccurate cancer reporting varied considerably by cancer type 32 . It is also important to consider the biases linked with overreporting cancers. Individuals whose self-reported cancer was unconfirmed by the Finnish cancer registry were more likely to have accumulated psychosocial strain across their lifecourse 33 . The probability of misreporting cancer, both over-reporting and under-reporting, has been associated with socio-demographic characteristics: gender, age, BMI, size of household, place of birth, smoking, social participation, educational level, type of employment, alcohol consumption and poor well-being 34 . The sensitivity of self-reported cancer is also likely to be higher among respondents with a high level of education 31 . Stavrou et al. have demonstrated a good level of sensitivity and specificity for self-reported cancer diagnoses in a cohort of older women in Australia, comparing self-reports to registry data 35 . Gupta et al. also describe good concordance between self-reported cancer and medical chart information 36 . The multivariate models also show associations between well known risk factors and cancer before 50, such as age at first pregnancy, smoking and drinking, which contributes to enhancing the validity of the cancer variable.

Women were four times more likely to report having had a cancer than men which is understandable given the age of cohort participants at the time of study. Based on 2007–2009 estimates of cancer incidence in the UK, 44% of cancers diagnosed among women aged 25–49 were breast cancers 37 . Rather than indicating a sex, or gender difference, the lack of association between virtually any of the variables and cancer among men is likely be mainly due to the far lower incidence in men up to age 50 (93 cancers). The distribution of cancer types observed in the cohort will continue to evolve over time, and begin to represent a greater proportion of men due to the increased occurrence of prostate and lung cancer among men >50 years. Linkage will hopefully be made between the cohort and the cancer registries which would significantly increased the validity and reliability of the information, and allow analyses by cancer type in the future, and survival analyses, which are currently not possible due to a lack of information on the timing of events. For these analyses we limited the health behavioural variables to those collected at age 23 rather than using variables available afterwards. This enabled to limit any further problems of missing data, but may lead to an underestimate of the contribution of health behaviours in the associations.

One important strength of this study is in the prospective nature of the information on ACEs. Most studies examining the links between early adversity rely on retrospective questions, which are prone to recall bias, an important issue raised by Korpimaki et al. using data from a working age population and the cancer registry 19 . Using a retrospective questionnaire to identify ACE, and limiting their analyses to individuals whose cancer was diagnosed subsequent to answering questions on ACE to address the problem of recall bias, these authors found no association between working-age cancer and reporting ACE.

There is currently insufficient power to work on cancers at different sites using the cohort. Furthermore, we take a lifecourse approach to this study and are interested in the pre-cancerous phase where early psychosocial adversity may increase susceptibility to developing cancer earlier. The cause of cancers and their prognoses are indeed diverse, however our aim here is to understand the factors contributing to susceptibility at its earliest phase. The ACE literature indicates that there are early life factors which may increase inflammatory response 38 , decrease immune system efficiency 39 , increase exposure to viral infections 40 , and raise the probability of damaging health behaviours along the lifecourse 41 42 . All of these behavioural and biological processes are known to be involved in cancer development.

The association between stress and cancer development and progression has been shown in biological studies 5 . Stress-related immunological changes bring about declines in natural killer cell activity by depressing their ability to respond to tumour or virally infected cells, and causing a reduction in the body’s defences linked to the repair of damaged DNA 43 . Exposure to stressors is known to trigger responses via the central nervous system produced by the hypothalamic-pituitary-adrenal axis (HPA). This activity modifies neuroendocrine pathways which, over the long term, alter the critical physiological mechanisms involved in tumourogenesis 5 . When exposed to chronic stress, “the body remains in a constant state of overdrive” with adverse consequences on the regulation of systems implicated in cancer progression 6 . ACE has also been associated with risky health behaviours such as smoking, alcoholism, early sexual activity, and having multiple sexual partners 8 13 38 39 40 41 , all of which are ‘indirect’ risk factors for cancer.

Our hypothesis that adversity in childhood may be linked with biological susceptibility involved in the development of cancers in adult life is reinforced by our findings. It could operate via two main mechanisms: A direct biological effect and an indirect effect via health behaviours (or a combination of both). The fact that ACE is still associated with cancer even after adjusting for behavioural and social mediators favours argument for a direct “biological” role of ACE.

In epidemiological studies, evidence of a direct association between exposure to stress and cancer incidence is mixed and inconclusive. This is likely to be due to the different ways in which stress was conceptualised. A Danish cohort study on 8736 men and women found no direct association between cumulative stressful life events collected retrospectively, mostly during adulthood, and cancer incidence, though they did identify a relationship between stress and unhealthy lifestyles 44 . Ollonen (2005) et al. found support for an overall association between stressful life events across the lifecourse and breast cancer risk in their Finnish case–control study 45 . A meta-analysis of studies on the association between stress and breast cancer did not support an association between stressful life events in adulthood and breast cancer risk 46 . Using linkage to the cancer registry, Fang et al. found that bereaved parents were at increased risk for cancers with an infectious aetiology, especially those linked to infection by the Human Papilloma Virus (HPV) after controlling for confounders. The authors hypothesised that the stress induced by losing a child may accelerate the cancer genesis of an established infection 47 . These studies do not consider the timing of exposure to stressful events in their analyses, an important factor considering the differential effects of physiological stress responses on various areas of the brain depending on when exposure occurs along the lifecourse 48 .

Our findings establish an association between ACE and cancer in women using prospective data. The suggestion that cancer risk may be influenced by conditions in the first years of life is potentially important in furthering our understanding of factors contributing to cancer development, and consequently in redirecting scientific research and developing appropriate prevention policies. This nevertheless remains a first step in understanding the relationship between ACE and cancer in the 1958 birth cohort. A second step will require an in-depth exploration of both direct and indirect pathways along which biological, social and psychosocial mechanisms are likely to operate.

ACE: Adverse childhood experiences; NCDS: National child development study; HPA: Hypothalamic-pituitary-adrenal axis.

The authors declare that they have no competing interests.

MKI was involved in the conception and design of the study, analysing and interpreting the data, drafted the manuscript and made modifications. PG, CD, and TL were involved in the conception and design of the study, analysing and interpreting the data, and revising the manuscript. BL and DD analysed and interpreted the analyses, and revised the manuscript. RL, NC, MB, and DB interpreted the analyses and revised the manuscript. The first author had full access to all of the data in the study and takes full responsibility for the integrity of the data and the accuracy of the data analysis. All authors approved the current version.