Project outline: The Economic Determinants and Costs of Delinquent and Disruptive Behavior

Motivation:

Disruptive and delinquent behavior in children and adolescents is a strong predictor of anti-social and

criminal behavior later in adults. (Sentence on Criminality Rates) While there is a large literature on the

development ofthese traits, risk factors for these traits, and evidence of efficacious treatment to prevent

these behaviors, the use of evidence based interventions is very low as practiced in communities within

major countries like the United States and the United Kingdom. There is debate over what role economic

policy might play in addressing this issue.

Research Plan:

Our group, a task force of Harvard's Center for the Developing Child, has been asked to perform analyses

that show how public policy can impact the development of delinquent and disruptive behavior, as well as

the economic costs of these traits to society. We seek to collaborate with (and support) ALSPAC

researchers in a study of the economic antecedents of and costs from disruptive and delinquent behavior.

Utilizing multiple data sources (including the Add Health and NLSY cohort data in our possession) we

seek to look at "natural" economic experiments where exogenous changes in public policy might impact

the development of these traits. We also seek to estimate the life-long costs to society of individuals with

these traits.

Research Aims:

Aim 1:

Determine the role that changes in economic policy might have on the development of disruptive and

delinquent behavior, as defined by standard research practice from survey responses. These policies

include

a. Health insurance coverage for parents and their children

b. Insurance design to improve access to evidence-based treatment for co-occurring

disorders (i.e. ADD, depression)

c. Welfare poliCies

d. School reforms, including special needs education

To control for endogeneity in outcomes, we propose to use two forms of randomization based on the

various strengths of the available datasets. First, we plan to utilize specific genetic variants linked with

these traits, controlling for parental genotype when possible. Second, we will use variation across time

and geographic areas in welfare, health insurance, and related variables that affect prevention of CD/ODD

and the effective treatment of these disorders. Finally, we hope to interact these policy changes with

temperament measures as a way of understanding individual traits.

Aim 2:

Determine the cost of delinquency and disruptive behavior. We seek to estimate the suspected cost to

society of children who meet criteria for oppositional defiant disorder and conduct disorder, by looking at

predicted future rates of incarceration and health care costs (particularly mental health) thru linking

childhood measures of disruptive and delinquent behavior with future criminal activity and other

economic measures.

Needs from ALSPAC data (primarily to serves as controls for our results):

1) Longitud inal measures of parental inputs (including parenting styles)

2) School -level traits, including data on peer groups

3) Estimates of anti-social behavior in children

5) Targeted SNP-level data for a limited set of markers (MAO-A, primarily)

6) Reported health care utilization and youth services intervention(s) for the sample, including

detailed information on use of behavioral and pharmacological treatments of ODDleD and

related disorders.

Differences in ADL, IQ and academic achievement between children with DCD and children at risk for DCD.

Study Team:

Dr. M.M. Schoemaker, Associate Professor in Human Movement Science, University of Groningen, The Netherlands.

Dr Raghu Lingam Lecturer in Community Child Health University of Bristol

Professor M.J. Jongmans, Department of Pediatrics, University Medical Centre Utrecht, The Netherlands

Professor Alan Emond, Professor of Child Health University of Bristol

Background

Developmental Coordination Disorder (DCD) is a frequently occurring disorder in childhood with a prevalence ranging between 1.4 and 19%. In order to be classified as DCD, children need to meet four diagnostic criteria as listed in the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV-TR, 2000). According to criterion A of the DSM-IV Criteria for DCD, children only meet this criterion if they have a significant impairment in motor coordination development, which interferes with academic achievement and/or activities of daily living (ADL). Over the years, various cut-off criteria have been applied to separate children with motor difficulties from typically developing children, varying from the 5th to the 20th centile. Recently, more strict inclusion criteria for the condition have been defined in the Leeds Consensus Statement (LCS, 2006). Children with scores below the 5th centile on an individually administered motor coordination test fulfill Criterion A; however those children between the 5th and 15th centile are said to have "probable" coordination difficulties and said to be at risk of DCD. Application of the more strict criteria as proposed by the LCS in the ALSPAC study, has recently led to a prevalence of 1,8% (Lingam et al., 2009).

Although a large group of clinicians and researchers have consented with the use of the 5th centile as cut-off criterion, it is acknowledged that this cut-off criterion is arbitrary and merely defined on statistical grounds. But as resources are limited for children with developmental problems, it is legitimate to offer those services in particular to those who are most severely in need of them. At present, in the UK more intensive services are provided for children with scores on a motor coordination test below the 5th centile than to those with scores between the 5th and 15th centile. The underlying assumption is that motor performance of children with scores below the 5th centile will interfere more with their academic achievement and ADL than those with less severe motor difficulties. However, this assumption is not evidence based. The decision to use the 5th centile as cut-off criterion will be more empirically grounded, if it is supported by research findings.

The data of the ALSPAC population based study offers an ideal opportunity to investigate whether children with more severe coordination difficulties are indeed the ones with the most severe problems in areas of academic achievement and/or ADL. In the ALSPAC study, 7,058 7 to 8 year old children were assessed with the ALSPAC coordination test, which was derived from the Movement Assessment Battery for Children. With this test, 324 children were identified as children with severe coordination difficulties (scores below 5th centile), and another 975 children were identified as children with moderate coordination difficulties (scores between 5th and 15th centile). Data regarding handwriting and reading, as measures of academic achievement, IQ as well as ADL were gathered for all of these children in the ALSPAC study, which enables a comparison of both groups of children regarding the amount of interference of the motor difficulties on the core aspects of the disorder. The large sample of children in ALSPAC offers sufficient statistical power to answer the research question in a reliable way.

Research question:

Are children with severe coordination difficulties (scores on coordination test below 5th centile) more impaired on measures of ADL and academic achievement than children with moderate coordination difficulties (scores on coordination test between 5th and 15th centile)?

Methods:

Participants:

The sample will consist of 6990 7-8 year old children who were assessed with the ALSPAC coordination test and who completed the ADL questions or handwriting test. Children with known visual, developmental, or neurologic condition will be excluded as well as children with an IQ less than 70. The ALSPAC coordination Test will be used to divide the sample into three groups: a group of children without coordination difficulties (scores on coordination test above the 15th centile), a group of children with moderate coordination difficulties (scores on coordination test between 5th and 15th centile) and a group with severe coordination difficulties (scores on coordination test below 5th centile).

The three groups of children will be compared on the following outcome variables:

Measure of ADL

ADL derived variable

Measures of academic functioning:

- Language (Wechsler Objective Language Dimensions-WOLD)

- Spelling (Nunes & Bryant)

- Reading (NARA-II and Wechsler Objective Reading Dimensions -WORD)

- Short-term memory (Children's Test of Nonword repetition)

- non-verbal skills (Diagnostic Analysis of Nonverbal Accuracy Scale-DANVA)

- IQ (WISC-III)

- attention/concentration (Development and Well-Being Assessment-DAWBA)

Analysis of variance will be carried out to investigate a main effect of group division, followed by post-hoc analyses to investigate differences between specific groups. Regression methods will account for potential confounding and mediating factors identified from the literature. Adjustment will be made for the bias introduced by missing data using multiple imputation by chained equation.

Power calculation:

The prevalence of motor coordination difficulties has been reported to be around 5% in literature. Taking 7000 to be the study population (approximate children that attended focus at 7) and our outcome variable to be the lowest 10% of a given developmental trait, the relative risk that it will possible to be 80% sure of detecting at the 5% level of significance is 1.59.

Not available

I am currently preparing a grant application for the MRC Methodology panel with a computer scientist from York, James Cussens (PI), to investigate the potential of powerful graph-learning algorithms for reconstructing pedigrees with maximum likelihood from genetic marker data. This is a problem that I have been working on for a few years with a Norwegian collaborator, Thore Egeland, but our emphasis has been very much on forensic identification applications and we have taken a more traditional statistical genetics approach. A pedigree is a special type of graph in that nodes representing individuals can only have two parents and parents must be labelled as being of opposite sex. Consequently, general graph learning algorithms are not very efficient unless they are suitably modified, although we do not have the additional problem encountered in other applications of having to learn the probabilities on the edges (arrows) between nodes as these derive from our knowledge of the genetic model under consideration. James has been in communication with us for some time and has considered a number of approaches to this specialised graph learning problem. He has decided that an approach using constrained optimisation using integer linear programming is the most promising. Indeed, initial results for a simplified situation whereby all individuals are observed and are totally ordered by age are promising, and much faster than competing methods, so we are quite optimistic.

Classical approaches to pedigree reconstruction developed 40 years ago, such as the sequential algorithm of Thompson ? which is a "greedy algorithm", in our terminology ? typically search for "good" but not necessarily optimal pedigrees by starting with sets of individuals that are most likely to be siblings, then finds the most likely parents for these sibships and moves on from there. The resulting reconstruction in heavily dependent on the decisions made about nuclear families at any given stage. Information on the maximal size of a sibship (which will vary across human populations and differ widely from animal and plant populations), the minimal number of years to reach sexual maturity (generation gap), age differences between parents etc. is incorporated in these algorithms at various stages in order to ensure that a realistic structure is produced. These are all "constraints" and one of our first tasks will be to search the statistical literature for those that are known to be important and then find a way to express them as linear constraints. All these algorithms assume complete data which we will typically not have. Searching over all possible pedigrees with missing individuals is much harder and is the interesting problem for the graph learning community. We will begin with a likelihood approach but will then move to a Bayesian framework where ""hard" prior information e.g. on certain known relationships and "soft" prior information e.g. this population is known to favour first cousin marriages etc. can all be incorporated. We will start with simulated data, moving from classic unlinked microsatellite-type data to linked markers and dense sets of SNPs: the likelihood calculations become more problematic then and we will need to employ Markov chain Monte Carlo methods to approximate likelihoods that are intractable with exact methods.

Besides the fact that this is a really nice graph theory problem, the main medical application is the potential for finding sets of related individuals from large Biobank populations for later linkage studies, or for homozygosity mapping where distantly related individuals are better because the shared environmental effects won't mask the relatively small effects of the rarer variants we're trying to find. Identifying the correct pedigree/relationship is crucial for forensic and population wildlife management applications, of course. It is also important for linkage analysis but is overkill for homozygosity mapping where you just need individuals with reasonable IBD-sharing probabilities. However, these general machine-learning algorithms are extremely efficient so if we can tailor-make one for pedigrees that will work quickly, we can obviously get distant relatives easily. We will then need to check sensitivity to getting the right relationship and investigate how best to choose relatives for homozygosity mapping from any constructed extended pedigree.

This is a methodological project with some very hard problems to solve before we get close to using it for anything. There are obviously ethical issues as to when and where you could do this in practice, but for now, we just want to see if it can be done. We will clearly do most of the work with simulated data but we will need a real dataset to check that what we are finding is not totally due to how we simulated. The main application is aimed at UK Biobank which has referred to the desirability of being able to infer relatives at a later point in its protocol. However, as they are a long way off genotyping, we will need a back-up data set to test things on if we get funded and if Biobank takes more than three years to deliver. Other existing cohort studies are probably not large enough to capture any relatives by chance. Obviously there are very close relationships in ALSPAC which are a bit too easy to find but, given that every birth in Avon over a two year period was included, we thought that there might be siblings or even a few cousins there. Our main intention would be to see if we can recover relationships you already know about - not to find hidden ones - although there is a possibility that the ones you know about are not quite as stated. Obviously all ethical issues will have to be handled as deemed appropriate at the time. In particular, we are not interested in any analysis of the ALSPAC data for relatedness, per se, so nothing will be published without clearing everything beforehand. Also, authorship of any paper that uses the ALSPAC data in any way at all, be it for the machine learning, statistical genetics, forensics or medical communities, will include appropriate ALSPAC investigator(s) and this will be discussed with you as beforehand.

Paul Burton (Leicester) and George Davey Smith (Bristol) are both coming on board as full investigators, not because of their great desire to learn about constrained optimisation but because we feel we need their input and experience to ensure a) that the focus of the research stays on medical applications and b) that anything to do with real data will be handled correctly. Knowing them both, I am confident that they will advise us well. I am being deliberately vague about what data we require because we do not know exactly what we need yet, it will be a long time before we need it and much more information than the above will have to be provided then. George mentioned that Illumina quad GWAS data would be available so once we have the machinery in place to handle dense SNP arrays, access to some of that data would be something to discuss. Right now, for the purposes of the grant application, I would like to be able to say, with your permission, that we are in discussion with ALSPAC collaborators for real data which will be used to test the methods in an appropriate way when we finally get to that stage.

Not available

Background: A number of studies have shown that maternal stress during pregnancy has effects on

offspring behaviour independent of postnatal measures of maternal stress. However, animal studies have

shown that behavioural development is influenced by the interaction of antenatal and postnatal

environments (Francis et al. 2003). Data are needed to test the co-action of antenatal and post-natal

environments in humans.

Aims: The proposed research has two aims: (1) to identify trajectories of familial risk from the prenatal to

the postnatal period, (2) to 'test whether familial risk trajectories up to age 8 years are associated with

child cognitive abilities and behaviour from 8 to 10 years. The design will provide evidence as to the

relative importance of antenatal versus postnatal environments and test additive versus interactive models

of antenatal and postnatal effects.

Hypothesis 1: We will identify trajectories that reflect (a) consistently low levels of familial risk, (b)

increasing risk from the antenatal to postnatal period, (c) decreasing risk from the antenatal to postnatal

period, (d) stable and high levels of familial risk.

Hypothesis 2a: If antenatal environments have enduring effects irrespective of postnatal environments,

children on the stable, high risk and declining trajectories will have poorer outcomes than children on the

stable, low risk and increasing risk trajectories.

Hypothesis 2b: If postnatal environments have effects on child outcome irrespective of antenatal

environments, then children on the stable, high risk and increasing risk trajectories will have poorer

outcomes than children on the stable, low risk and declining risk trajectories.

Hypothesis 2c: If antenatal and postnatal environments have additive effects, then children on the stable,

high risk trajectory will have the poorest outcomes, children on the stable, low risk trajectory will have

the most positive outcomes, and children on the increasing and declining trajectories will fall intermediate

to the other two groups.

Hypothesis 2d: If antenatal and postnatal environments have interactive effects, then children on the

stable, high risk trajectory will have poorer outcomes than the other three groups.

Analysis: The analysis would involve Cl) estimating trajectories of environmental risk from the antenatal

through the postnatal period up to age 8 years involving a composite index of financial hardship,

domestic violence (as reflected in mothers' reports of conflict with partner and in life events

questionnaire), child harm (as reflected in life events questionnaire), maternal social support, and

maternal depression and anxiety; (2) estimating associations between environmental risk trajectories and

child behaviour (symptoms of antisocial behaviour, depression, and attention deficit/hyperactivity) and

cognitive outcomes (IQ) at 8 to 10 years. Analyses would control for antenatal smoking and alcohol use

as well as parent education and parental history of antisocial behaviour and mental health problems.

We wish to identify those ALSPAC children who have had 3D facial scans, orthoptic assessments, refraction, clinical examination and for whom we have maternal alcohol intake data. We will stratify alcohol exposure as done previously for this cohort into women who drank no alcohol in pregnancy, or less than 1 drink per week, or more than on drink per week (2). In this earlier ALSPCAC paper, effects on child temperament were identified after exposure to greater than 1 drink per week during pregnancy (approx 16% women in that analysis), suggesting there is enough variation in alcohol exposure within the cohort for useful analyses of outcome. We will perform a variety of statistical analyses seeking to examine associations between alcohol exposure and facial anatomical and ophthalmic measurements. For example, category of first trimester alcohol intake as a predictor or modifier of (a) palpebral aperture width (between medial and lateral canthi), (b) distance between the eyes (between the medial canthi) (c) occurrence, direction and magnitude of strabismus (assessed at age 7) and refractive error (outcome data at age 3 for 2000 and at age 15 for approx 5000) and ocular biometry (available for 2900 at age 15).

AIMS

Specific Aim 1: Longitudinal methylomic analyses in each trajectory group

o Perform a hypothesis-free genome-wide screen of DNA methylation in the 200 individuals using the Illumina Infinium 27K Methylation array on bisulfite-treated DNA from samples obtained at birth, age 7 and age 15

o Perform extended fine-mapping verification analyses across specific genomic regions nominated from the microarray screen using Pyrosequening and/or the Sequenom EpiTYPER system in larger numbers of cases and controls.

o Assess the extent to which dynamic patterns of methylation vary across the trajectories

o Is there variation in the epigenetic-regulation of genes in the trajectories relative to controls?

? We would estimate "methylation trajectories" conditional on the conduct trajectories in an attempt to sort "noise" from the potentially informative changes in gene regulation

Specific Aim 2: Examine candidate-gene DNA methylation patterns

o Focus on HPA axis regulatory genes known to functionally inter-relate with serotonin

o e.g., corticotropin-releasing hormone (CRH), CRH receptors, CRH binding proteins, pro-opiomelanocortin (POMC), steroid hormone biosynthetic enzymes, etc.).

o Focus on testosterone (androgen) receptors, androgen biosynthesis pathways, monoamine/catecholamine-related genes (e.g., monoamine oxidase, transporters)

o Quantitative analysis of DNA methylation across gene promoter regions using using Pyrosequening and/or the Sequenom EpiTYPER system .

o Perform extended fine-mapping verification analyses across specific genomic regions nominated from the microarray screen using Pyrosequening and/or the Sequenom EpiTYPER system in cases and controls.

o Assess the extent to which dynamic patterns of methylation vary across the trajectories

o Is there variation in the epigenetic regulation of genes in the trajectories relative to controls?

We would estimate "methylation trajectories" conditional on the conduct trajectories in an attempt to sort "noise" from the potentially informative changes in regulation

Specific Aim 3: examine associations between DNA methylation trajectories and environmental

risk exposures from gestation though to adolescence

3.1 Data collection.

DNA from samples obtained at birth, age 7 and age 15

3.2 Existing Data Required

Environmental risk data are listed in the previous section. The genes we propose to genotype for the candidate methylation scans are listed in the next section.

3.4 Data Analysis.

See specific aims (above). Epigentic analysis will take place at the The Institute of Psychiatry Psychiatric Epigenetics Group, King's College London.

Background:

Hyperacusis is defined as oversensitivity to some everyday sounds (lowered threshold of discomfort to sounds). Children may complain that their ears hurt when exposed to some sounds, typically high pitched or unexpected loud noises. For some children, this can be extremely distressing and can have a significant effect on behaviour. Phonophobia is defined as a fear of sound and in the most extreme cases, the lowest level of the particular sound can trigger what appears to be a disproportionate reaction, with physical symptoms of anxiety, tantrums and avoidant behaviour.

Certain groups of patients are known to be at greater risk of experiencing hyperacusis. It is a common and often disabling condition in Williams Syndrome and has been associated with abnormal cochlear function (Gothelf et al, 2006). It is the commonest sensory perceptual abnormality in children who are on the Autistic Spectrum, with prevalence ranging from 15-100% (Gomes et al, 2008). In these children there is little evidence of peripheral audiological abnormality (Gravel et al, 2006). Research has shown that there are more likely to be central causes for the unusual reactions to auditory stimuli in these children (Khalfa et al, 2001; Orekhova et al, 2008; Jones et al, 2009).

A recent study of 506 children between the ages of 5 and 12 years has used a questionnaire to assess the prevalence of hyperacusis in typically developing children (Coelho et al, 2007a). This identified that hyperacusis was reported by 3.2% of children. The same population was also asked questions related to tinnitus (Coelho et al, 2007b). 37% of children reported hearing a noise inside their head and 17% described it as annoying. Risk factors for experiencing tinnitus included young age, hearing loss and hyperacusis.

The ALSPAC data offers a unique opportunity to establish the prevalence of hyperacusis in a large population of children and to evaluate whether there are any significant associations with early life history, including detailed hearing data.

Research aims and objectives:

1. To determine the prevalence of hyperacusis in the ALSPAC cohort at age 11 years

2. To investigate risk factors for hyperacusis, specifically to examine:

a) Child factors: gender, prematurity, birthweight, migraine, special educational needs (ASD, ADHD & other conditions), meningitis & encephalitis, head injury requiring hospitalisation

b) Auditory function: hearing and acoustic reflex thresholds, current and historical otitis media with effusion, tinnitus

c) Socioeconomic factors: socioeconomic status, maternal education and housing tenure

Methods:

At the age 11 Focus clinic, approximately 7000 children were asked about sensitivity to sound, known as hyperacusis. Pilot analysis showed that approximately 3 % of these children reported hyperacusis of which 60% had experienced it for at least a year. Children were also asked whether they experienced noises in their ears (tinnitus) and pilot analysis showed that children with hyperacusis were more likely to report tinnitus than children without hyperacusis.

We intend to explore the characteristics of children reporting hyperacusis to determine possible risk factors for this condition. We will use univariate and multivariate logistic regression to determine whether those with hyperacusis have common factors in their early medical history (see objective 2a). We will also determine whether there are any differences in pure tone and acoustic reflex thresholds. Finally, we will examine socioeconomic factors.