Conceived and designed the experiments: TP NBF HM. Analyzed the data: EC S. Service JW JKS S. Schonauer. Contributed reagents/materials/analysis tools: HT MK MJ MRJ LP JV HM. Wrote the paper: EC S. Service NBF JM TP.
The authors have declared that no competing interests exist.
Investigation of the environmental influences on human behavioral phenotypes is important for our understanding of the causation of psychiatric disorders. However, there are complexities associated with the assessment of environmental influences on behavior.
We conducted a series of analyses using a prospective, longitudinal study of a nationally representative birth cohort from Finland (the Northern Finland 1966 Birth Cohort). Participants included a total of 3,761 male and female cohort members who were living in Finland at the age of 16 years and who had complete temperament scores. Our initial analyses (Wessman et al.,
Measures of early environment, neurobehavioral development, and adolescent behavior significantly predict adult temperament, classified by both cluster membership and temperament dimensions. Specifically, our results suggest that a relatively consistent set of life course measures are associated with adult temperament profiles, including maternal education, characteristics of the family’s location and residence, adolescent academic performance, and adolescent smoking.
Our finding that a consistent set of life course measures predict temperament clusters indicate that these clusters represent distinct developmental temperament trajectories and that information about a subset of life course measures has implications for adult health outcomes.
Understanding the causation of psychiatric disorders will require dissection of the specific genetic and environmental determinants of disease susceptibility. Yet the two components of this task differ enormously in their feasibility. The genetic variations contributing to such susceptibility, although mostly still unknown, are knowable. Aspects of genetic variation are fixed throughout life, and increasingly straightforward to assay; most will likely be identified within the decade, after routine genome re-sequencing provides comprehensive catalogs of genome variants.
Investigation of the environmental influences on human behavioral phenotypes poses more fundamental questions. The environment encompasses a vast array of different components, some of which are distinct and objectively measurable – for example exposure to particular toxins – while others are generally poorly defined and their severity only assessed subjectively – such as stressful life events
Longitudinal birth-cohorts uniquely provide such a framework. They offer the opportunity to assess the influence of multiple early environmental factors on the development of neurobehavioral profiles. Such cohorts also enable examination of the relationship between these profiles and overt expression of psychiatric illness and adult temperament while avoiding problems associated with sampling and recall bias. The Northern Finland 1966 Birth Cohort (NFBC 1966) is well suited to address these types of questions, as more than 10,000 individuals born in the year 1966 in the two most northern provinces of Finland have been followed over the course of their life, starting from before birth, until age 31. The NFBC 1966 database permits longitudinal analyses of sociodemographic characteristics, neurodevelopment, and quantitative neurobehavioral measures
Another example is the Dunedin Multidisciplinary Health and Development Study, from which a number of early childhood factors have been identified that predict the risk of developing post-traumatic stress disorder
Temperament is considered a candidate endophenotype for a wide range of psychiatric disorders, reflecting common genetic factors shared across diagnostic categories
In the second of this pair of analyses (presented here), we set out to further examine whether the temperament patterns seen in adulthood are consistent across the developmental trajectory. Specifically, we examined the relationship between prospective measures capturing the early environment, neurobehavioral development, and adolescent behavior (obtained from the extensive life course data available in NFBC 1966) and temperament clusters assessed in adulthood. Temperament, in our series of analyses, therefore represents a critical phenotype for examining the development of individual differences associated with adult health outcome. By conducting a data-driven investigation to uncover relationships between life course measures and adult temperamental profiles in this rich, longitudinal birth cohort, our approach is in contrast to many analyses of early environmental influences on temperament in longitudinal birth cohorts. First, conducting an exploratory analysis with a range of life course variables enabled us to comprehensively examine all variables, in order to identify suitable targets for future research, rather than limit our focus to a single known predictor. Second, by comparing the relationship between these life course variables and temperament profiles to the relationship between these variables and individual temperament scales, we were able to compare these different (i.e., person-oriented vs. variable-oriented) approaches to representing temperament. Although our data-driven approach did not involve testing a series of hypotheses about each prospective measure, we did hypothesize that:
We would identify early life course measures that could predict temperament clusters, just as we identified health and outcome correlates of temperament in adulthood in Wessman et al. (
We would identify associations between early life course measures and adult temperament clusters that are consistent with previous findings of risk factors for the development of psychopathology.
We note that the analysis conducted here does not allow us to make conclusions about causality (which is difficult to establish with life course measures and temperament). The analysis does, however, identify specific measures that are associated with the development of temperament features.
The Northern Finland 1966 Birth Cohort (NFBC 1966) is a longitudinal birth cohort, initially comprised of all 12,058 individuals live-born in 1966 from the two northernmost provinces of Finland, Oulu and Lapland
When cohort members were 31-years old, all subjects who were alive at the time and had a known address were asked to complete a subset (107 items) of Cloninger’s Temperament and Character Inventory (TCI) version 9 questionnaire for measurement of four dimensions of temperament (Novelty Seeking: NS, Harm Avoidance: HA, Reward Dependence: RD, and Persistence: P) and their respective subscales
The current study sample contains cohort members who were living in Finland at the age of 16 years, who completed the TCI at the age of 31, who were not mentally retarded, and who provided informed consent (N = 3,761∶ 1,726 male, 2,035 female). All subjects included in the present study gave written consent for their data to be used. The study was approved by the Ethics Committee of the Faculty of Medicine, University of Oulu.
As this was an exploratory analysis to identify any life course measure associated with adult temperament, we did not restrict our choice of variables based on
Predictors | Cluster I | Cluster II | Cluster III | Cluster IV | Scales_F | Scales_M |
Primary parent occupation at birth | M | |||||
Maternal education | F, M | M | F | P | ||
Mother’s age | F, M | M | NS | HA | ||
Mother lived in same region entire life | F | RD | RD | |||
Ratio children/household rooms | ||||||
Home location at birth | F | NS | ||||
Distance to maternity clinic | ||||||
Distance to neighbors | HA | |||||
Distance to city/town center | RD | |||||
Distance to doctor | F | HA | RD | |||
Household has running electricity | RD | |||||
Household has running water | F | |||||
Household has running car | ||||||
Family owns home at birth | F | HA | ||||
Mother worked outside of home during pregnancy | M | HA | ||||
How strenuously the mother worked during the pregnancy | ||||||
Mother exposed to outside information during pregnancy | M | NS | HA | |||
Desirability of the pregnancy | M | M | ||||
Mother’s frame of mind during the pregnancy | F | |||||
Mother smoked during the pregnancy | ||||||
Maternity clinic visits |
Significant predictors are indicated by an F for female or M for male Cluster I–IV membership or the scale name for individual TCI scales. Scales_F: predictors of individual TCI scales for females; Scales_M: predictors of individual TCI scales for males.
Predictors | Cluster I | Cluster II | Cluster III | Cluster IV | Scales_F | Scales_M |
Birth weight | ||||||
Birth height | P | |||||
Ponderal index at birth | ||||||
Weight at one year | ||||||
Height at one year | ||||||
Ponderal index at one year | ||||||
Age of standing | ||||||
Age of walking without support | ||||||
Number of words spoken by age one | F | HA, NS | RD | |||
Child wets self during the day at age one | HA | HA, RD | ||||
Child wets self during the night at age one | NS | |||||
Potty-training age one |
Significant predictors are indicated by an F for female or M for male Cluster I–IV membership or the scale name for individual TCI scales. Scales_F: predictors of individual TCI scales for females; Scales_M: predictors of individual TCI scales for males.
Predictors | Cluster I | Cluster II | Cluster III | Cluster IV | Scales_F | Scales_M |
Father’s occupation in adolescence | F | HA, NS | NS | |||
Family status | ||||||
Home location in adolescence | F, M | NS | ||||
Weight in adolescence | ||||||
Height in adolescence | F | F | ||||
Number of hospital visits from 1966–1987 | ||||||
Number of long-duration illnesses. |
Significant predictors are indicated by an F for female or M for male Cluster I–IV membership or the scale name for individual TCI scales. Scales_F: predictors of individual TCI scales for females; Scales_M: predictors of individual TCI scales for males.
Predictors | Cluster I | Cluster II | Cluster III | Cluster IV | Scales_F | Scales_M |
School level classification | M | HA | ||||
Average grades in adolescence | F, M | F, M | F | P | ||
Repeated grade in school | RD | |||||
Admitted to secondary school | F | RD | RD | |||
School admission | F | |||||
Times applied to secondary school | M | M | M | RD | ||
Physical education grades in adolescence | F | M | M | HA | ||
Frequency of sports outside of school | F | HA | HA | |||
Smoking in adolescence | F | F | F | F | NS | |
Drinking in adolescence | HA, NS | |||||
Being drunk in adolescence | M | NS | ||||
Intoxicant use in adolescence | HA, NS |
Significant predictors are indicated by an F for female or M for male Cluster I–IV membership or the scale name for individual TCI scales. Scales_F: predictors of individual TCI scales for females; Scales_M: predictors of individual TCI scales for males.
The following sociodemographic characteristics were selected from a questionnaire completed in the 24th to 28th gestational week (
A selection of variables representing infant developmental milestones and health were selected from data collected during the cohort members’ examination by nurses performed at one year of age (
A selection of variables reflecting family characteristics and adolescent health were selected from a questionnaire mailed to the cohort members at age 14, in addition to information obtained from the national health registry (
Educational attainment and adolescent behavior characteristics were selected from a questionnaire mailed to the cohort members in 1980, in addition to information obtained from the Joint Application System (which is a nationwide application system through which cohort members applied to secondary level education) in 1982 (
Cluster analysis, using the
The resulting clusters obtained from these prior analyses are described briefly here (and in greater detail in our concurrent manuscript, Wessman et al.,
In the present analyses, we attempted, using a multivariate analysis, to identify life course measures that were significantly associated with membership in the above four clusters (I–IV). To reduce the number of variables for consideration in the multivariate models, we first conducted a series of univariate analyses, separately by sex, in order to examine differences between temperament clusters in early life variables. These univariate analyses consisted of one-way analyses of variance for continuous variables and chi-square tests for categorical variables, and were conducted using R statistical software (R 2.9.2) (
After initial univariate analyses, we conducted four stepwise logistic regression analyses for each sex. The outcome in these logistic regression analyses was an indicator variable for membership in one of the four clusters. By entering all variables that significantly predicted cluster differences in univariate analyses (at a
To examine life course measures in relation to temperament dimensions, we followed a similar analysis plan to that employed for the temperament clusters, specifically using univariate analyses to identify candidate independent life course variables followed by a multivariate analysis. The difference between these analyses was that for each of the TCI scales (NS, HA, RD, and P), and for each sex, we used linear models (rather than logistic models used for temperament clusters) to predict the temperament values as a function of each life course variable.
After initial univariate analyses, we conducted four stepwise linear regression analyses. By entering all variables that significantly predicted dimension scores in univariate analyses (at a
To examine the relative correlation of life course measures with temperament cluster membership as compared to temperament dimensions, we present a generalized r2 for the logistic models
Our univariate analyses reveal multiple variables that significantly differed between clusters (
Stepwise logistic regression analyses revealed sets of variables that significantly predict group membership for each cluster and each sex separately (
A comparison across clusters reveals that a relatively consistent set of life course measures predicts group membership, including maternal education, characteristics of the family’s location and residence, adolescent academic performance, and adolescent smoking. In particular, the odds of being in Cluster I for both males and females decreases, while the odds of being in Cluster II for males and Cluster III for females increases, with increasing maternal education. In terms of the prenatal sociodemographic environment, for females the odds of Cluster I membership decrease as households become more rural and distant from key resources, the odds of Cluster II membership decrease as families report not having running water, and the odds of Cluster IV membership decrease as families report not owning their own home. In terms of educational milestones and behavior at age 14, the odds of being in Cluster I (females and males) increase, while the odds of being in Cluster III (females and males) and IV (females) decrease, with increasing grades. The odds of being in Cluster II increase with increasing physical education grades for females, while the odds of being in Cluster IV decreases with increasing physical education grades for males. Finally, with increasing smoking reported by females in adolescence, the odds of either Cluster II or III membership increases, but the odds of either Cluster I or IV membership decreases. The presentation of
A comparison of generalized r2 values from the logistic models (
Furthermore, in terms of the measures that significantly predict temperament scale scores, there is little consistency of variables across scales or across sexes. The only measures that consistently predict scale scores for both sexes are mother’s lifetime residence for RD, whether the child wets him/herself during the day at age one for HA, and sports frequency in adolescence for HA. In contrast to the suite of life course measures that predict more than one cluster, there are no shared variables that significantly predict scores across temperament scales. This contrast is highlighted in
This is the first report to demonstrate that life course measures (assessed as early as before birth) significantly predict adult temperament (assessed at age 31). Although some prior evidence has suggested specific developmental pathways with implications for psychopathology leading from early environment to adult temperament, such evidence derives from studies using limited age ranges or retrospective data. In the series of analyses reported here, we first observed that stable and robust clusters of temperament differ on a number of variables that were assessed at age 31 in the NFBC 1966, including lifestyle, working capacity, socioeconomics status, and mental health (Wessman et al.,
The goal of these analyses was to identify sociodemographic, developmental, and behavioral correlates, as measured prenatally, in infancy and into adolescence, of adult outcome as indicated by temperament profiles. By conducting a data-driven investigation using a longitudinal birth-cohort, we are able to demonstrate that a set of life course measures predict adult temperament clusters, revealing both novel relationships and confirming similarly reported associations. Although we do not make any claims about causation, based on our findings we propose that these clusters represent distinct temperament profiles and that information about a subset of life course measures has implications for adult health outcomes.
These findings have implications for our understanding of the development of individual differences in temperament, as well as mental health outcome in adulthood. It has been shown that specific early environmental risk factors influence psychiatric susceptibility
In support of our first hypothesis – that we would identify early life course measures that predict adult temperament clusters – we were able to demonstrate that life course measures assessed as early as the prenatal period are associated with membership in distinct clusters organized according to temperament in adulthood, and that these differences seen across the life course are consistent with differences between clusters seen in habits, socioeconomic status, and health in adulthood (Wessman et al.,
The results of our multivariate analyses suggest that a relatively consistent set of life course measures are associated with adult temperament profiles, including maternal education, characteristics of the family’s location and residence, adolescent academic performance, and adolescent smoking. In support of our second hypothesis – that we would identify associations between early life course measures and adult temperament that are consistent with previous risk factors for the development of psychopathology – the set of life course measures identified in our analyses are in line with previous reports. For example, maternal education has been associated with children’s problem behavior, such that increasing maternal education protects against the development of problem behaviors at ages 2 and 5
Adolescent smoking has also been implicated as playing an important role in the developmental trajectory as it is predicted by early life measures (particularly family socioeconomic status)
The comprehensive assessment of life course measures in this cohort therefore allows for the elucidation of a set of correlates that potentially play an important role in the development of individual differences in temperament. The set of variables that consistently predicts adult temperament across the four clusters reflects the growth environment (such as maternal characteristics or the nature of the home environment) or the early, emerging temperament of cohort members (such as academic performance). The variables related to the growth environment may reflect the background of emerging temperament. Alternatively, as the development of temperament is under genetic control, these variables may interact via mechanisms of genetic correlation, as the genetic background of the parents (with whom the offspring shares genes) affects the growth environment.
Overall our findings suggest that a suite of life course measures predicts membership across temperament clusters. While these measures may be either shared between temperament clusters (maternal education) or unique to a given cluster (number of words spoken at age one), the measures that predict temperament dimensions are unique to particular temperament scales. In the accompanying report, we demonstrate that these temperament clusters are significantly related to adult outcome across a number of lifestyle and health domains and that the proportion of variables significantly associated with clusters is similar to the proportion of variables significantly associated with any subscale, suggesting that these clusters capture as much information about adult outcome as individual scale scores alone (Wessman et al.,
One advantage of organizing adult temperament according to such clusters is that this strategy reduces the number of variables to be tested. An additional advantage is that it provides the opportunity to consider the context of the individual’s temperamental profile and environmental influences, so that it is possible to consider how different combinations of temperament dimensions assort within individuals, rather than requiring the assumption that dimensions operate independently
In our first set of analyses (Wessman et al.,
In the analyses reported here, we extended our analyses longitudinally, and assessed the relationship between the temperament clusters and a broad range of sociodemographic, developmental, and behavioral measures that were measured prenatally, in infancy and into adolescence. Our results suggest that a relatively consistent set of life course measures are associated with adult temperament profiles, including maternal education, characteristics of the family’s location and residence, adolescent academic performance, and adolescent smoking.
Considering these sets of findings together, our results provide additional support for such a person-oriented approach, and increase our understanding of the factors that contribute to the trajectory of individual differences in temperament, which in turn influence adult mental and physical health outcomes.
The primary strength of this report is the use of a longitudinal birth cohort that allowed us to investigate whether sociodemographic, developmental, and behavioral variables that were assessed prenatally and through development predict temperament scores assessed in adulthood. Our analyses of the NFBC 1966 allowed us to examine the influence of multiple life course measures on temperamental profiles while avoiding problems associated with sampling and recall bias. Our approach to analyzing all 54 life course measures in relation to temperament was exploratory: each variable was treated the same (e.g., not ordered), considered independently at the first stage of univariate analyses, and carried forward to the second stage of multivariate analyses if significant. Although there is potentially some overlap in some of the items, we chose to analyze all variables that were available, if they were sufficiently described and if more than 50% of cohort members had data available for that variable, in order to examine as much as the environmental search space as possible.
The primary limitation of this report is that the information available for analysis is constrained by what was collected in the cohort. For example, we did not have a direct index of socioeconomic status available for the families of cohort members, nor did we have a measure of fetal alcohol exposure. In addition, despite the rich dataset collected on this cohort, temperament was only assessed in adulthood. Future work should be aimed at repeated measurement of temperament, as well as socioeconomic status and family characteristics, health conditions, developmental milestones, education and behavior, across the life course.
We also did not have equal sample sizes for males and females, which complicates interpretation of differences in results across sexes. The NFBC 1966 began with a cohort of 12,058 live births; here, our analyses were conducted on a total of approximately 1,400 of those individuals. While it has previously been demonstrated that study participation is lower in individuals with a psychiatric illness as compared to those without, participation does not vary across specific disorders
Finally, as we have already stated, our analysis does not allow us to make conclusions about causality, but identifies specific measures that are associated with the development of temperament features. In addition, it is also possible that associations reported here are indirect, such that an additional, unmeasured variable is responsible for their association. Additional comprehensive and longitudinal cohorts will be critical to uncovering the mechanisms underlying temperament and the development of psychopathology.
Early environment, neurobehavioral development, and adolescent behavior significantly predict adult temperament. Although all multivariate models account for less than 10% of the variation in outcome classifications (both cluster membership and temperament dimension), our results highlight a consistent set of life course measures that predict temperament clusters. Of note, we were able to replicate previous associations between early life variables (e.g., maternal education) and adult temperament, even when considering a large set of life course measures. These results contribute to our understanding of how individual differences in life course correlates are related to individual differences in adult temperament, and support the utility of conducting data-driven research to both uncover novel, and replicate previously reported, associations.
Our results demonstrate significant relationships between life course measures and temperament clusters, particularly in females. There is substantial evidence that risk factors for later psychopathology include parental psychopathology, low socioeconomic status, prenatal stress and the experience of negative life events, maternal smoking, a low maternal age and education
(DOC)
(DOC)
(DOC)
(DOC)
(DOC)
(DOC)
(DOC)
(DOC)
(DOC)