The current report describes the Childhood Trauma Study and examines retrospectively-assessed CSA and CPA in the sample of adult twins, their non-twin siblings, and their parents. Our results demonstrate the feasibility of obtaining a very detailed assessment of childhood maltreatment from individuals largely in their third and fourth decades of life and provide evidence that these data demonstrate strong psychometric properties.
Data from twins provided an opportunity to examine the agreement of screening item endorsement with a comprehensive reassessment administered after an extensive interval that averaged slightly more than 7 years. Although more severe abuse might be expected to display greater temporal stability, the lowest rate of CSA reported at time 2 was found for those who reported having been raped before age 18. Because the question assessing rape did not mention CSA (abuse status was instead determined from reported of age at first occurrence), lower time 2 CSA endorsement may have resulted from telescoping with respondents near the age cut-off reporting either an earlier or a later onset for the same experience at reinterview. Those who reported they had been physically abused in response to the screening question asking about CPA had the highest endorsement of CPA at reinterview. The relatively high specificity and lower sensitivity of this item is consistent with what other reports (Fricker et al., 2003
; Hardt & Rutter, 2004
; Thombs et al., 2006
) have found for screening questions that assess abuse status without incorporating any descriptors. Consistent with this, the cohort II screening question for sexual molestation which included a behavioral description had comparable specificity, but better sensitivity. For both forms of abuse, lower tetrachoric correlation values were found for screening items that were less frequently endorsed likely due in part to the comprehensive nature of the Christchurch Trauma Assessment; the substantial prevalence of CSA and CPA in control twins also supports this suggestion. Overall, a very acceptable degree of agreement was observed for both sets of abuse items particularly given the 7-year interval and extensive content differences. The kappa coefficient values for the composite CSA and CPA variables are very similar to estimates from other studies (Fergusson et al., 2000
; Durrett et al., 2004
; Bifulco et al., 2005
; Hardt et al., 2006
; Mills et al., 2007
) which often used the identical instruments for both assessments and had considerably shorter mean intervals between interviews. Fergusson and colleagues (2000)
reported very similar kappa statistics for CSA and CPA obtained from their administration of the Christchurch Trauma Assessment to their birth cohort at ages 18 and 21. They conducted latent class analyses with their data and concluded that the primary source of unreliable reports was from false negative responses occurring at either assessment rather than from false positive responses. They observed consistent risk for psychiatric outcomes conditional on exposure to abuse defined variously. One limitation of their analyses was that their model assumed that false positive reports did not occur (apparently, this assumption was necessary for their model to be fully identified).
Our data demonstrate significant relationships between offspring report of parental alcohol problems and self-reported CSA and CPA. These findings, consistent with prior community sample reports which have consistently observed significant risk associated with a history of parental alcoholism for CSA (Fleming et al., 1997
; Miller et al., 1999
; Dinwiddie et al., 2000
; Nelson et al., 2002
) and CPA (Mullen et al., 1996
; Fergusson & Lynskey, 1997
; Dube et al., 2001
), offer additional validation of retrospective data collected with the Christchurch Trauma Assessment. A similar dose–response relationship between the number of alcoholic parents and risk for emotional and physical neglect also has been reported (Dube et al., 2001
). Analyses using maximal parental educational attainment as a proxy for family-of-origin socioeconomic status observed higher parental educational status was protective against each form of abuse. As poverty-associated risk for child abuse is well-established (e.g., Sedlak & Broadhurst, 1996
), these findings provide additional evidence validating our abuse constructs. Other investigations (Fergusson et al. 2000
; Bifulco et al., 2005
) have used risk of distal outcomes as an alternative approach to construct validation. Although we plan to report outcome data in a separate publication, we provide exemplary data on alcohol dependence risk in twins and their non-twin siblings here. Including CSA and CPA in a single logistic model that controlled for gender, significant alcohol dependence risk was associated with CSA [OR 1.44 (95% CI 1.16 – 1.79)] and PA [OR 1.60 (95% CI 1.30 – 1.98)].
The significant within same-sex twin pair agreement for both forms for abuse can be viewed as offering evidence of corroboration for these constructs (Hardt & Rutter, 2004
). It is somewhat surprising that nearly identical correlations were found in same-sex twins for CSA and CPA given that each twin is reporting about his or her own experience and CSA is not limited to intra-familial abuse. The lower agreement between opposite-sex twins for CPA and the negative correlation for CSA reported by opposite-sex twins is suggestive of gender differences in risk. These findings are not surprising given that the prevalence of CSA is known to be substantially higher for women (Fergusson & Mullen, 1999
) and some gender differences in patterns of response to abuse screening items have been observed (Thombs et al., 2006
). One investigation that examined corroboration in adult sisters for retrospective reports of CSA, CPA, and neglect found much stronger corroboration when these individuals were concordant for that form of abuse (Bifulco et al., 1997
The significant differences between high-risk and control siblings for CSA (females only) and CPA provide additional corroboration of retrospective reports. Given that families were ascertained on the basis of twins’ responses to screening questions, it is expected that family status-related differences are somewhat attenuated in siblings. The rates observed for both forms of abuse in high-risk siblings also considerably higher than prevalence estimates from the general population (Fergusson et al., 2000
). The lack of a significant difference for CSA in men most likely resulted from lower power due to the lower prevalence. The substantial rates in control twins of both CSA and CPA are presumably reflecting false negative responses due to the limited sensitivity of the screening questions. Other reports have observed similar patterns of response (Fergusson et al., 2000
; Hardt et al., 2006
; Thombs et al., 2006
The relationships between twin pair abuse status and sibling abuse risk provide some evidence that familial risk for abuse is type-specific and that patterns of observed risk differ for these two types of abuse. For CSA, comparable risk is seen for any level of twin endorsement. For CPA, greater evidence is found for cumulative loading of familial risk as well as a suggestion that missing data are non-random and indicative of greater liability. Consistent with this latter finding, 11 of the 37 cohort II study respondents (29.7%) who reported that their co-twin was deceased [mean age of death was 24.2 (SD 5.49)] gave a history of CPA (Odds Ratio 4.84; 95%CI 2.38–9.86). These results are somewhat consistent with Bifulco and colleagues’ (1997)
report observing greater concordance between sisters for CPA and neglect than for CSA.
Several limitations of our study need to be considered. No gold standard exists for measures of retrospectively reported childhood abuse. It seems reasonable to assume that the comprehensive Christchurch Trauma Assessment is more sensitive than the cohort II screening questions, but we cannot definitively demonstrate this. Our data can only provide indirect support for the validity of our abuse constructs. Since kappa statistic values are somewhat dependent on population base rates, our sample’s enrichment for both forms of abuse may have affected estimates to some degree. Although offspring reports of parental alcohol problems could be biased by abuse history, their use is supported by excellent degree of within-pair twin agreement as well as significant associations both with parental self-report and missing parental interview. Maximal parental educational attainment is admittedly a rough proxy for socioeconomic status; however, the relative consistency of findings using this measure is promising. It is possible that our ascertainment on the basis of twins’ responses to cohort II abuse screening items may have introduced some bias in terms of their willingness to participate. One report (Edwards et al., 2001
) found that those with a history of CSA in an HMO population are more likely to participate in research, but found no excess of psychopathology in these participants versus non-participants with a history of CSA. Similarly, the change in design necessitated by currency fluctuations may have influenced our findings by increasing participation rates of high-risk twins. Although our prioritization of males who reported a history of CSA at time 1 may have affected the degree to which our high-risk families are representative of cohort II twins with a history of CSA or CPA, it is encouraging that we found no significant bias involving either parental problems or maximal parental education attainment on time 2 participation of high-risk twins.