A total of 1238 families initially consented to participate in ROOTS and 1143 primary carers (92%) completed the CAMEEI. Non-participants were more likely to be from the moderate/hard pressed ACORN categories (27%) than the sample as a whole (14%; χ2 = 8.8, df = 2, p = 0.01). Of the 1143 interviews, 96% (1092) were biological mothers and 3% (35) biological fathers. The remainder were adoptive mothers (7), both parents (3) and 2 each of extended family members, step-mothers and step-fathers. To assess inter-rater agreement, 48 interviews were observed and independently double-coded. Agreement was high (kappa = 0.7 - 0.9) on those core indicators with sufficient positive endorsements to permit analysis (any family discord, parenting and any financial difficulties).
Characteristics of the study sample
Gender, ACORN classification and maternal age at birth of the proband were not included in the latent class modelling but used for descriptive purposes only. Of the 1143 adolescents, 622 (54.4%) were females; families were classified as wealthy and urban prosperous (62%), comfortably off (24%) and moderate/hard pressed (14%); 4% of mothers were under 20 at the birth of the proband.
Prevalence of family adversities
Table shows the prevalence of reported exposures to each indicator of family adversity.
Characteristics of family adversity indicators from the CAMEEI parent interview (N = 1143)
Prevalence is given for each discrete time period to expose changes over the 15 year life-course. Most indicators showed marked consistency in the proportions of individuals exposed at each time period, but parental divorce increased from early (10%) to later childhood (15%) but thereafter dropped (8%) in early adolescence. Exposure to mild family discord, acute life events, chronic difficulties and sibling psychiatric illness peaked in early adolescence.
Tetrachoric correlations of the adversity indicators, together with the descriptive variables for each time period are given in tables , , . The correlations range from -0.26 - 0.78 in early childhood, with similar levels at subsequent time points.
Tetrachoric correlations of family adversity indicators in early childhood (N = 1143)
Tetrachoric correlations of family adversity indicators in later childhood (N = 1143)
Tetrachoric correlations of family adversity indicators in early adolescence (N = 1143)
Latent class modelling: early childhood to early adolescence
LCA results for models including nine composite adversity indicators are reported from solutions in which up to five classes were estimated and interpreted. The indicators were: i) family loss, ii) family discord, iii) abuse and criminality, iv) financial problems plus unemployment, v) paternal psychiatric illness, vi) maternal psychiatric illness, vii) paternal parenting style, viii) maternal parenting style, ix) maternal lack of affection or engagement. All indicators were modelled as binary (present/absent) with the exception of family discord, which was categorised as i) none, ii) mild, iii) moderate and severe. Log-likelihood values, information criteria and classification accuracy are reported for all models (tables and ). We judged that four classes provided the most parsimonious solution based on joint consideration of the full range of indices and interpretation of the clusters and class size.
Early childhood to early adolescence information criteria for latent class models with 1-5 classes
Assignment probabilities - probabilistic versus modal allocation 4 class model
Figure shows the probability of endorsing each adversity indicator and overall class sizes based on posterior modal allocations. For comparative purposes, the item probabilities of the one class model were plotted to represent the sample average as in table .
Figure 1 a: Early childhood, four class model - probability of endorsing exposure by class membership. b: Later childhood, four class model - probability of endorsing exposure by class membership. c: Early adolescence, four class model - probability of endorsing (more ...)
In early childhood 784 individuals (69%; 436 [56%] females) were allocated to class 1 characterised by a relatively low or zero exposure to any adversities with levels below the sample average. Class 2 comprised 213 individuals (19%; 113 [53%] females) and was characterised by relatively high rates of family loss, mild family discord, paternal atypical parenting, financial difficulties and maternal psychiatric illness. Moderate/severe family discord and maternal lack of affection/engagement were less likely but higher than the sample average. Paternal psychiatric illness was unlikely to be endorsed in this class. Individuals in class 2 were unlikely to have experienced maternal atypical parenting or abuse/criminality. Class 3 consisted of 66 individuals (6%; 31[47%] females) with strikingly high probabilities of moderate/severe family discord (95%), paternal atypical parenting, paternal psychiatric illness. Almost all the abuse/criminality was allocated to this class. The probability of mild family discord in this class was zero. Maternal psychiatric illness was similar to class 2 level, approaching 40%. Compared to other classes, class 3 showed higher probabilities of all other indicators. Finally, 76 individuals were allocated to class 4 (7%; 4 [54%] females) characterised by conspicuously high levels (70%) of atypical parenting from both parents with other indicators close to the sample average except maternal lack of affection/engagement which was slightly elevated but below the levels seen in classes 2 and 3.
Classes 1, 2 and 3 were clearly characterised by severity of exposure. Class probabilities were consistent and distinct across all adversity types. We designated these as low (class 1), moderate (2) and severe (3) adversity classes. In contrast, class 4, characterised almost exclusively by atypical parenting from both parents, was most accurately defined in terms of the nature of the experience rather than severity. This qualitative difference is clear in Figure showing the probabilities of the atypical parenting class membership crossing the 3 severity classes.
The class profiles in later childhood and early adolescence were markedly similar to early childhood. From early through to later childhood there was a rise in the proportion allocated to the moderate class (26% from 19%) and the atypical parenting class (10% from 7%) with a comparatively lower proportion (60% from 69%) allocated to the low class.
Cross-tabulation of class membership with the inactive covariates revealed that chronic social problems, acute disturbances and sibling psychiatric disorder were elevated only in the severe adversity class. Medical illnesses (parental and sibling) were distributed across all 4 classes, with no specific class associations.
Assignment probabilities for the low, severe and atypical parenting classes of the four class model were high (0.85-0.95) (table ). The moderate class was less well discriminated (0.79-0.81) with these individuals also having non-zero probabilities for membership in the low class.
Class membership over time
To understand the extent of mobility between classes over time, pathways were plotted for each class. For ease of presentation, the plots were separated according to class allocation in early childhood. For example, Figure plots the temporal pathways of individuals allocated to the low class in early childhood and similarly, Figure plots the pathways for those starting in the moderate class.
Pathways of early family environment.
Class membership was fairly stable over time with 55.3% (630) of the total sample remaining in the same class across all time periods (patterns 111, 222, 333, 444).
Stability of class membership over time ranged from 63.9% (501) in the low to 42.4% (28) in the severe and 28.6% (61) in the moderate class. A further small proportion from each class who had switched class in later childhood had reverted to their class of origin by early adolescence. In total 77.2% (606) from the low, 51.5% (45) from the severe and 39.4% (84) from the moderate classes in early childhood were similarly classified in early adolescence.
In a proportion of young people adversity increased over time: 19.9% (156) from the low class in early childhood had been allocated to the moderate or severe class by early adolescence. A further 25 (11.7%) from the moderate class had been elevated to the severe by age 14.
Of those in the moderate class in early childhood, 45.5% (97) had moved out of an adverse environment by early adolescence. In the severe class the figure was 22.8% (15) with a further 16.6% (11) living in a less, though still moderately severe, adverse family environment.
Over half of those in the atypical parenting class (45/76 [59.2%]) in early childhood were similarly classified in early adolescence. Of those 45, 89% (40) remained stable throughout.
Final model estimation - longitudinal class
Final model estimation computed a longitudinal class based on the modal allocation at each time point. The four class model provided the optimum and most stable solution shown in tables and .
Longitudinal model estimation: information criteria 1 - 6 classes
Longitudinal model estimation: probabilistic versus modal classification 4 class model
The low exposure class (754, 66%) represented individuals with a strong probability of remaining stable throughout, although a small number, allocated longitudinally to this class, had experienced moderate, but not severe, adversity at some point. The longitudinal moderate exposure class (236, 21%) showed more fluctuation over time. Here individuals had a strong probability of being in the moderate adversity class during later childhood, but a lower probability at either early childhood or early adolescence. The severe class (88, 8%) is predominately characterized by persistently high levels of exposure to multiple adversities with moderate/severe family discord and abuse/criminality being prominent. The atypical parenting class (62, 5%) was predominantly characterized by atypical parenting throughout.
Assignment probabilities for the low, severe and atypical parenting classes of the four class model were very high (> = .93). The moderate class was less well discriminated (0.84) with these individuals also having non-zero probabilities for membership in class 1 (table ).
There were no significant gender differences in the latent class profiles either at discrete time-points or longitudinally (χ2 = 2.7, 3df, p = 0.44). However, there were significant differences according to socio-economic group. The longitudinal severe adversity class comprised 33% of individuals classified as moderate means/hard pressed compared to 10% in the low adversity class and 14% of the overall sample (χ2 = 46.3, 6df, p <0.001). The mother's age at the birth of the proband was also differentially represented by the class profiles; longitudinally 13% of the high exposure class were born to teenage mothers compared to 2% in the low, 7% in the moderate adversity classes and 5% in the atypical parenting class (χ2 = 26.0, df = 3, p <0.001). These associations are in the expected direction and consistent with prior studies.
Association of longitudinal class with psychopathology
The lifetime prevalence and differential probabilities of DSM IV diagnoses by longitudinal class, adjusted for gender and ACORN classification, are shown in table .
Association of longitudinal class with psychopathology by age 14
The odds ratios increased across the three severity classes for psychopathologies. Compared to the low adversity group, severe adversity over time (class 3) was associated with an eightfold increase in odds ratio (OR) for disruptive behaviour disorders (conduct disorder, oppositional defiant disorder and ADHD), a 4.8 times increase for depressive disorders and approaching a two-fold increase for anxiety and non suicidal self injury (NSSI). Exposure to atypical parenting (class 4) suggested an increase in the OR for NSSI and depressive disorders although it should be noted that these associations were not significant by conventional statistical tests. Individuals in the moderate and severe classes were significantly more likely to have had more than 1 diagnosis over their lifetime (11% and 17% respectively) compared to 4% in the low and atypical parenting classes and 7% of the sample as a whole (χ2 = 48.8, p <0.001).
Gender, psychopathology and longitudinal class
We found an interesting association between psychopathology, longitudinal adversity class and gender. In the low adversity class significantly more females than males showed disorder (21% vs 9%, χ2 = 22.4, p <0.001). However, this levelled off as severity of adversity increased (moderate class: 35% vs 26%, χ2 = 3.7, p = 0.06) until, in the severe adversity class, around 40% of both males and females reported an episode of psychopathology with a significant gender by class interaction (p = 0.02) for the severe class compared to the low adversity class. The atypical parenting class resembled the low class (29% vs 11%, χ2 = 3.2, p = 0.07).