A Multisite Neurobehavioral Assessment of FASD Sarah N. Mattson, Ph.D. Center for Behavioral Teratology Department of Psychology San Diego State University

Can we accurately identify FASD/ARND? 60-90% of children born to alcoholic mothers do not demonstrate the classic facial features of FAS These children are affected by prenatal alcohol exposure in similar ways as those children with FAS but are difficult to identify and thus do not receive adequate treatments or interventions Our goal is to determine which features can be used to accurately identify children affected by prenatal alcohol exposure, including those who do not have FAS Not every child is exposed to alcohol has the facial features of FAS. However, even those children who do not have the classic facial features are affected, particularly in terms of their behavior or cognition. But because they are not dysmorphic (do not show the physical features), they are more difficult to identify. Our goal is to determine which features can be used to identify children who are affected by heavy prenatal alcohol exposure but who do not have FAS. 2

Recent Accomplisments: 2012 5 papers accepted or in press 1 paper under review 5 papers in preparation

Background Prenatal exposure to alcohol often results in disruption to cognitive and behavioral domains, including executive function (EF) and adaptive behavior (AB). Although EF and AB have been shown to be related (i.e., poor EF correlates with poor AB), this relation had not been tested in FASD. In the current study, this relation was examined in subjects from the CIFASD dataset. Prenatal exposure to alcohol often results in disruption to discrete cognitive and behavioral domains, including executive function (EF) and adaptive functioning. In the current study, the relation between these 2 domains was examined in children with histories of heavy prenatal alcohol exposure, nonexposed children with a diagnosis of attention-deficit/hyperactivity disorder (ADHD), and typically developing controls.

Methods 3 groups of children (8-16y, M=12.1) from multiple sites were tested Children with histories of heavy prenatal alcohol exposure (AE, n=142) Nonexposed children with attention-deficit/hyperactivity disorder (ADHD, n=82) Nonexposed controls without alcohol exposure or ADHD (CON, n=133) Children completed subtests of the Delis-Kaplan Executive Function System (D-KEFS) Caregivers completed the Vineland Adaptive Behavior Scales-II Data were analyzed using regression analyses As part of a multisite study, 3 groups of children (8 to 18 years, M = 12.10) were tested: children with histories of heavy prenatal alcohol exposure (ALC, n = 142), nonexposed children with ADHD (ADHD, n = 82), and typically developing controls (CON, n = 133) who did not have ADHD or a history of prenatal alcohol exposure. Children completed subtests of the Delis–Kaplan Executive Function System (D-KEFS), and their primary caregivers completed the Vineland Adaptive Behavior Scales-II. Data were analyzed using regression analyses.

Subject Demographics

Results EF measures were predictive of adaptive abilities, and significant group-specific interactions between EF and group were present. For the ADHD group, the relation between adaptive abilities and EF was more general, with 3 of the 4 EF measures showing a significant relation with adaptive score. For the AE group, only nonverbal EF related to adaptive behavior. In the CON group, performance on EF tasks did not predict adaptive scores over the influence of age. Analyses showed that EF measures were predictive of adaptive abilities, and significant interactions between D-KEFS measures and group were present. For the ADHD group, the relation between adaptive abilities and EF was more general, with 3 of the 4 EF measures showing a significant relation with adaptive score. In contrast, for the ALC group, this relation was specific to the nonverbal EF measures. In the CON group, performance on EF tasks did not predict adaptive scores over the influence of age.

Conclusions/Implications As in previous studies of ADHD, EF deficits were predictive of poorer AB. The relation between EF and AB was extended to include children with heavy prenatal exposure to alcohol. The relation between EF and AB differed by group, suggesting unique patterns of abilities in alcohol-exposed children. Specifically, in the AE group, nonverbal EF performance was related to AB while verbal EF performance was not. Laboratory measures of EF can be used to predict deficits in daily function in children; nonverbal EF is more predictive of AB in alcohol- exposed children than verbal EF. Test batteries designed to identify FASD/ARND should include both verbal and nonverbal EF measures. These results support prior research in ADHD, suggesting that EF deficits are predictive of poorer adaptive behavior and extend this finding to include children with heavy prenatal exposure to alcohol. However, the relation between EF and adaptive ability differed by group, suggesting unique patterns of abilities in these children. These results provide enhanced understanding of adaptive deficits in these populations, as well as demonstrate the ecological validity of laboratory measures of EF.

Background Children with heavy prenatal alcohol exposure often meet criteria for attention-deficit/hyperactivity disorder (ADHD). ADHD research has examined subtype differences in symptomatology, including sluggish cognitive tempo (SCT). This construct is defined by behavioral symptoms including hypoactivity and daydreaming and has been linked to increased internalizing behaviors. The current study examined whether similar findings are displayed in children with prenatal alcohol exposure. Children with heavy prenatal alcohol exposure often meet criteria for attention-deficit/ hyperactivity disorder (ADHD). ADHD research has examined subtype differences in symptomatology, including sluggish cognitive tempo (SCT). This construct is defined by behavioral symptoms including hypoactivity and daydreaming and has been linked to increased internalizing behaviors. The current study examined whether similar findings are displayed in children with prenatal alcohol exposure.

Methods 4 groups of children (8-16y, M=12.2) from multiple sites were tested Children with histories of heavy prenatal alcohol exposure and ADHD (AE+, n=75) Children with histories of heavy prenatal alcohol exposure without ADHD (AE-, n=35) Nonexposed children with attention-deficit/hyperactivity disorder (ADHD, n=60) Nonexposed controls without alcohol exposure or ADHD (CON, n=102) Caregivers completed the Child Behavior Checklist (CBCL) and the SCT Scale CBCL and SCT data were analyzed using 2 (exposure) x 2 (ADHD) ANOVA. Correlations were used to test the relation between CBCL and SCT. Discriminant Function analysis examined whether SCT items could accurately classify groups As part of a multisite study, caregivers of 272 children (8 to 16 years) completed the SCT Scale and Child Behavior Checklist (CBCL). Four groups were included: alcohol-exposed children with ADHD (ALC+; n = 75), alcohol-exposed children without ADHD (ALC; n = 35), nonexposed children with ADHD (ADHD; n = 60), and nonexposed children without ADHD (CON; n = 102). SCT and CBCL scores were analyzed using 2 (exposure) x 2 (ADHD) analyses of variance. Pearson’s correlations measured the relationships between SCT, CBCL, and Full Scale IQ (FSIQ). Discriminant function analysis examined whether SCT items could accurately classify groups.

2 x 2 Design ADHD Dx Yes No Alcohol Exposure AE+ AE- ADHD CON

SCT Items Absent minded Overtired Preoccupied Apathetic Lack of persistence Stares blankly Forgets details Lacks energy Forgets where things are kept Forgetful Leaves things behind Drowsy Confused Forgets instructions Daydreams

Subject Demographics

Results SCT Scores For the SCT-S, the main effects of AE and ADHD and their interaction were all significant All > CON AE+ = ADHD AE+ & ADHD > AE- SCT significantly correlated with CBCL internalizing, externalizing, and attention problems in all groups and with FSIQ in the AE+ group Int Ext Attn FSIQ AE+ .35** .39** .60** .32* AE- .41** .43** .42** -.29 ADHD .49** .58** -.08 CON .45** .54** -.02 No demographic variables were significantly related to SCT scores (p > 0.20); therefore, SCT analyses were continued without covariates. There was a significant main effect of exposure, F (1, 268) = 13.82, p < 0.001 (exposed > nonexposed), and ADHD, F (1, 268) = 177.17, p < 0.001 (ADHD > non- ADHD). There was also a significant Exposure x ADHD interaction, F (1, 258) = 9.04, p = 0.003 (Fig. 1). Pairwise comparisons revealed that ALC+ had significantly higher SCT scores than ALC- (p < 0.001), and both ALC- and ADHD had significantly higher SCT scores than CON (p < 0.001). The ALC+ and ADHD groups had similar SCT scores (p = 0.422). Analyses revealed significant main effects of exposure and ADHD on SCT and internalizing and externalizing scores and significant interaction effects on SCT and internalizing scores. SCT significantly correlated with internalizing, externalizing, and attention ratings in all groups and with FSIQ in ALC+. Discriminant function analysis indicated that specific SCT items could distinguish ALC from CON. **p<.01; *p<.05

Discriminant Function Results DFA was used to determine whether individual SCT items could be used to distinguish groups. 4 SCT items could be used to distinguish AE- from CON subjects Forgets details, confused, forgetful, drowsy 78% classification accuracy (62% AE-/83%CON) 3 of 4 items do not overlap with ADHD vs. CON analysis (below) indicating specificity 3 SCT items distinguished AE- from ADHD (AE+ & ADHD) Lack of persistence, leaves things behind, forgets instructions 77% classification accuracy (79%AE-/76%ADHD) Significant items were redundant with main analyses (ADHD>AE-) and therefore not clinically meaningful 2 SCT items distinguished ADHD from CON Lack of persistence, forgets details 92% classification accuracy (91%ADHD/92%CON) These analyses test whether SCT items can distinguish alcohol-exposed subjects without ADHD from typically developing children and from children with ADHD, respectively. The third analysis was included to determine the specificity of the discriminating items in the first analysis.

Conclusions/Implications Results supported the hypothesis that prenatal alcohol exposure is related to elevations of SCT. Both alcohol-exposed groups demonstrated elevations in SCT relative to nonexposed controls. The alcohol-exposed group without ADHD had significantly lower SCT scores than the alcohol-exposed group with ADHD, but were still elevated over controls indicating that ADHD is not necessary for elevated SCT. 4 items successfully discriminated the AE- and CON groups and 3 of these items did not overlap with the 2 items that could distinguish ADHD from CON, suggesting that these items may be useful in differential diagnosis, particularly for alcohol- exposed children without ADHD. Alcohol-exposed children exhibited elevated SCT scores. Elevations were related to increased parent ratings of internalizing and externalizing behaviors and attention. These findings are observed in alcohol-exposed children regardless of ADHD symptoms and specific SCT items proved useful in distinguishing exposed children, suggesting clinical utility for this measure in further defining the neurobehavioral profile related to prenatal alcohol exposure.

Background Previous studies have examined the rates psychiatric disorders, including ADHD, in children with histories of heavy prenatal alcohol exposure No study had addressed the rates of these disorders in relation to ADHD We compared the prevalence of psychiatric disorders and behavioral problems in children with and without alcohol exposure and with and without ADHD Background: This study examined prevalence of psychiatric disorders and behavioral problems in children with and without prenatal alcohol exposure (AE) and attention-deficit/hyperactivity disorder (ADHD).

Methods 4 groups of children (8-16y, M=12.3) from multiple sites were tested Children with histories of heavy prenatal alcohol exposure and ADHD (AE+, n=85) Children with histories of heavy prenatal alcohol exposure without ADHD (AE-, n=52) Nonexposed children with attention-deficit/hyperactivity disorder (ADHD, n=74) Nonexposed controls without alcohol exposure or ADHD (CON, n=133) Caregivers completed the Child Behavior Checklist (CBCL) and the Computerized Diagnostic Interview Schedule for Children-IV (C-DISC- 4.0) Frequency of specific psychiatric disorders, comorbidity, and CBCL behavioral scores were analyzed by Chi-square and ANCOVA Methods: Primary caregivers of 344 children (8 to 16 years,M = 12.28) completed the Computerized Diagnostic Interview Schedule for Children-IV (C-DISC-4.0) and the Child Behavior Checklist (CBCL). Subjects comprised 4 groups: AE with ADHD (AE+, n = 85) and without ADHD (AE_, n = 52), and nonexposed with ADHD (ADHD, n = 74) and without ADHD (CON, n = 133). The frequency of specific psychiatric disorders, number of psychiatric disorders (comorbidity), and CBCL behavioral scores were examined using chi-square and analysis of covariance techniques.

Subject Demographics

Results Dx AE+ AE- ADHD CON GAD 5.9% 1.9% 12.2% 0% MDD 7.1% 5.8% 16.2% Clinical groups had greater frequency of all psychiatric disorders, except for anxiety, where the AE- and CON groups did not differ. AE+=AE- for MDD and GAD Synergistic effect of AE&ADHD on Conduct disorder ADHD increased comorbidity (regardless of AE) but AE did not. For CBCL scores, significant interactions indicated a combined effect of AE and ADHD on Externalizing, Total Problems, and Attention Problems. There were main effects of AE and ADHD on all scales (not shown) Dx AE+ AE- ADHD CON GAD 5.9% 1.9% 12.2% 0% MDD 7.1% 5.8% 16.2% ODD 48.2% 11.5% 45.9% 3.8% CD 18.8% 5.4% Mn #Dx 0.80 0.23 0.04 Frequencies are presented as % of group. Results: Clinical groups had greater frequency of all psychiatric disorders, except for anxiety, where the AE- and CON groups did not differ. There was a combined effect of AE and ADHD on conduct disorder. For comorbidity, children with ADHD had increased psychiatric disorders regardless of AE, which did not have an independent effect on comorbidity. For CBCL scores, there were significant main effects of AE and ADHD on all scores and significant AE x ADHD interactions for Withdrawn/ Depressed, Somatic Complaints, Attention, and all Summary scores. There was a combined effect of AE and ADHD on Externalizing, Total Problems, and Attention Problems. Scale AE+ AE- ADHD CON Attn 73 60 69 51 Ext 66 56 43 Tot 68 57 63 42

Conclusions/Implications Results support increased risk of additional psychiatric disorders in children with ADHD, regardless of alcohol exposure. Examining independent effects of AE and ADHD allowed us to document increases in externalizing disorders and behavior problems, particularly for conduct disorder and attention problems, which occurred at the greatest rates in the group with both alcohol exposure and ADHD. In the AE groups, internalizing disorders were less affected by presence of ADHD These results, along with others presented, indicate that more than one profile may exist in children with prenatal alcohol exposure, at least in the behavioral domain. Behavior profiles may vary dependent on presence of ADHD Conclusions: Findings indicate that ADHD diagnosis elevates children’s risk of psychiatric diagnoses, regardless of AE, but suggest an exacerbated relation between AE and ADHD on conduct disorder and externalizing behavioral problems in children. Findings affirm a poorer behavioral prognosis for alcohol-exposed children with ADHD and suggest that more than 1 neurobehavioral profile may exist for individuals with AE.

Background A main goal of CIFASD has been to define the neurobehavioral profile of FASD Our previous study suggested a profile that could distinguish subjects with alcohol exposure from controls, regardless fo the presence of FAS The current study aimed to determine whether a neurobehavioral profile exists that can distinguish alcohol- exposed subjects from subjects with ADHD Background: Heavy prenatal alcohol exposure (AE) results in a broad array of neurobehavioral deficits. Recent research has focused on identification of a neurobehavioral profile or profiles that will improve the identification of children affected by AE. This study aimed to build on our preliminary neurobehavioral profile to improve classification accuracy and test the specificity of the resulting profile in an alternate clinical group.  

Methods 3 groups of children (8-17y, M=12.3) from multiple sites were tested Children with histories of heavy prenatal alcohol exposure (AE, n=209), including 79 with FAS Nonexposed children with attention-deficit/hyperactivity disorder (ADHD, n=74) Nonexposed controls without alcohol exposure or ADHD (CON, n=185) Children completed a standardized neuropsychological test battery focused on executive function Data were analyzed using 3 latent profile analyses AE with FAS & CON; AE without FAS & CON; AE & ADHD Methods: A standardized neuropsychological test battery was administered to 3 groups of children: subjects with AE (n = 209), typically developing controls (CON, n = 185), and subjects with attentiondeficit/ hyperactivity disorder (ADHD, n = 74). We assessed a large sample from 6 sites in the United States and South Africa, using standardized methodology. Data were analyzed using 3 latent profile analyses including (i) subjects with fetal alcohol syndrome (FAS) and controls, (ii) subjects with AE without FAS and controls, and (iii) subjects with AE (with or without FAS) and subjects with ADHD.

Subject Demographics

Results Results: Classification accuracy was moderate but significant across the 3 analyses. In analysis 1, overall classification accuracy was 76.1% (77.2% FAS, 75.7% CON). In the second analysis, overall classification accuracy was 71.5% (70.1% AE/non-FAS, 72.4% CON). In the third analysis, overall classification accuracy was 73.9% (59.8% AE, 75.7% ADHD). Subjects that were misclassified were examined for systematic differences from those that were correctly classified. In these analyses, controls who were younger, with lower IQ scores, and/or from South Africa were more likely to be misclassified as AE. In contrast, AE subjects from the United States, with higher IQ, were more likely than other AE subjects to be misclassified as controls. It is difficult to disentangle the factors that led to misclassification as some of the factors may be related to the country of origin.

Conclusions/Implications Results indicate that the neuropsychological effects of AE are clinically meaningful and can be used to accurately distinguish alcohol-affected children from both typically developing children and children with ADHD. In combination with other recent studies, these data suggest that approximately 70% of children with heavy prenatal alcohol exposure are neurobehaviorally affected, while the remaining 30% are spared these often-devastating consequences, at least in the domains studied. Refining the neurobehavioral profile will allow improved identification and treatment development for children affected by prenatal alcohol exposure. Conclusions: The results of this study indicate that the neuropsychological effects of AE are clinically meaningful and can be used to accurately distinguish alcohol-affected children from both typically developing children and children with ADHD. Further, in combination with other recent studies, these data suggest that approximately 70% of children with heavy prenatal alcohol exposure are neurobehaviorally affected, while the remaining 30% are spared these often-devastating consequences, at least those in the domains under study. Refining the neurobehavioral profile will allow improved identification and treatment development for children affected by prenatal alcohol exposure.

NEUROPSYCHOLOGICAL DEFICITS ASSOCIATED WITH HEAVY PRENATAL ALCOHOL EXPOSURE ARE NOT EXACERBATED BY COMORBID ADHD Leila Glass; Ashley L. Ware; Nicole Crocker; Benjamin N. Deweese; Claire D. Coles; Julie A. Kable; Philip A. May; Wendy O. Kalberg; Elizabeth R. Sowell; Kenneth Lyons Jones; Edward P. Riley; Sarah N. Mattson, and the CIFASD Submitted 12/6/2012

Background Neuropsychological functioning of individuals with ADHD or heavy prenatal alcohol exposure has been well documented independently but the interaction between these factors have not been addressed Our recent studies suggest that behavioral profiles of alcohol exposure may differ depending on the presence of ADHD. The current study examined the interaction between prenatal alcohol exposure and ADHD on cognitive performance Objective: Neuropsychological functioning of individuals with attention-deficit/hyperactivity disorder (ADHD) or heavy prenatal alcohol exposure has been well documented independently. This study examined the interaction between both factors on cognitive performance in children.

Methods 4 groups of children (8-16y, M=12.3) from multiple sites were tested Children with histories of heavy prenatal alcohol exposure and ADHD (AE+, n=90) Children with histories of heavy prenatal alcohol exposure without ADHD (AE-, n=38) Nonexposed children with attention-deficit/hyperactivity disorder (ADHD, n=80) Nonexposed controls without alcohol exposure or ADHD (CON, n=136) Children completed a standardized neuropsychological test battery focused on executive function Neuropsychological data were analyzed using 2 (exposure) x 2 (ADHD) ANOVA. Method: As part of a multisite study, 344 children (8-16y, M=12.28, SD=2.52) completed a comprehensive neuropsychological battery. Four subject groups were tested: children with histories of heavy prenatal alcohol exposure (AE) and ADHD (AE+, n=90), alcohol-exposed without ADHD, (AE-, n=38), non-exposed with ADHD (ADHD, n=80), and non-exposed without ADHD (CON, n=136).

Subject Demographics

Results Interactions between AE and ADHD were found on 6 measures. Follow-up contrasts demonstrated no difference between AE+ & AE- AE+/- < ADHD on VCI and PRI Main effects of ADHD and AE were also found on overall performance Results: Separate 2(AE) x 2(ADHD) MANCOVAs revealed significant main and interactive effects of ADHD and AE on overall WISC-IV, D-KEFS, and CANTAB performance. Individual ANOVAs revealed significant interactions on 2 WISC-IV indices [Verbal Comprehension (VCI), Perceptual Reasoning (PRI)], and four D-KEFS and CANTAB subtests [Design Fluency, Verbal Fluency, Trail Making, Spatial Working Memory]. Follow-up contrasts demonstrated no difference between AE+ and AE- groups on any measures. The combined AE+/- group demonstrated more severe impairment than the ADHD group on VCI and PRI, but there were no other differences between clinical groups.

Conclusions/Implications Results support use of a combined AE+/AE- group for neuropsychological research Both alcohol exposure and ADHD lead to neuropsychological impairment. However, in some cases the neuropsychological effects seen in ADHD are altered by prenatal alcohol exposure The effects of alcohol exposure on verbal comprehension and perceptual reasoning were greater than those related to ADHD along Unlike the behavioral studies, reported earlier, there is no exacerbation of alcohol’s effects in the presence of ADHD Conclusions: These results support a combined AE+/- group for neuropsychological research and indicate that, in some cases, the neuropsychological effects seen in ADHD are altered by prenatal alcohol exposure. The effects of alcohol exposure on verbal comprehension and perceptual reasoning were greater than those related to having ADHD without alcohol exposure, although both conditions independently resulted in cognitive impairment compared to controls. Clinically, these findings demonstrate task-dependent patterns of impairment across clinical disorders.

Recent Accomplishments: Data Collection Preparation/Planning Subcontracts were initiated IRB approvals were obtained The CIFASD neurobehavioral team has established the framework for valid and reliable data collection. Test administration kits, record forms, and materials necessary for subject assessment were purchased and disseminated to each of the four sites. In October, at least one psychometrist from each site travelled to SDSU for a multi-day, intensive training on proper test administration and scoring, subject interaction and recruitment procedures. At each site, the psychometrists administered the CIFASD III neurobehavioral battery to a pilot subject, and a recording of that assessment was reviewed by San Diego State University to ensure compliance with uniform administration procedures. Three of the 4 sites have at least one psychometrist who has demonstrated sufficient mastery of the battery to begin testing subjects. The fourth site is being reviewed currently. Initiation of Data Collection As planned, data collection was initiated by the 5th funding month. Three of 4 sites are actively collecting data, starting on 11/14/12. 19 subjects have been tested (13 AE) across both age groups (5-7, 10-16) Subjects will not be formally counted until upload and tallying procedures are in place. USC: 1 old AE UMN: 5 old AE SDSU: 2 young AE 5 old AE 5 old CON 1 old Contrast

Recent Accomplishments: Database Development With the assistance of the informatics core, a database was developed to facilitate the fast and reliable transmission of test results to a centralized, online data repository. The upload tool was deployed to the SDSU site on 1/7/13 for testing. A comprehensive user-guide and a data dictionary were developed to enhance the understanding and utility of these tools and exhaustive safeguards were implemented to ensure data integrity. We are currently making modifications to the existing demographic data collection and subject classification algorithms to fit with the current phase aims. We are also preparing an automatically generated tally of complete participant data that will inform monthly conference calls as to each site’s progress. Subjects who fit into the defined groups and have complete (uploaded) data will be counted on a monthly basis.

Recent Accomplishments: Papers in Preparation We are close to submitting a paper addressing the interaction between alcohol exposure and a diagnosis of ADHD on adaptive behavior. Results indicate that the interaction between these two factors leads to an exacerbation of adaptive behavior deficits. Children with both conditions have worse outcomes than those with either factor alone. This paper is a companion to the recently submitted paper by Glass et al. We examined the relation between parent-report and direct laboratory assessment of executive function. Relations differed by subject group (FASD vs. ADHD vs. CON). We submitted this as an RSA abstract. Using CIFASD I data, we demonstrated that parent reports were consistent with direct laboratory measures of attention but not hyperactivity. We examined the relation between fluency ability and bilateral striatum volumes. Results confirmed poorer verbal and design fluency and reduced bilateral caudate and putamen volume in the AE group and indicated significant negative correlations between right putamen and caudate volumes and VF performance in the AE group (but not controls). We are currently revising a previously submitted paper on the virtual water maze from CIFASD I.

Recent Accomplishments: Specific Aim 1 Specific Aim 1 was to develop a hierarchical approach to data collection & analysis that will streamline data requirements for diagnosis/identification of alcohol-affected children CART analysis Compiling data from multiple domains yielded 1412 variables and 641 subjects (256 with AE, 100 of which have FAS) Only variables with >100 subjects were included Initial analysis included correlation with alcohol exposure. To reduce the number of variables, we retained the 10 variables with the strongest correlations by project (total of 30 variables). Intercorrelated variables were excluded, reducing our variables to only 7. We are revisiting the dataset to start with a larger number of variables. We are aiming for 10 variables (not intercorrelated) from each project Once all domains are examined and variables are narrowed down to the best identifiers, we plan to conduct a classification and regression tree (CART) analysis to establish cutoff scores that will show us what is most predictive of children with prenatal alcohol exposure that do not meet diagnostic criteria for FAS. Cross validation for the CART will be performed in order to validate our CART model, which will help us estimate how accurately the model will perform clinically. Current aims. To address aim 1, we first compiled data from neuropsychological, dysmorphology, 3D facial imaging, parent questionnaire, and brain imaging testing in order to examine the correlations between all variables with the presence of prenatal alcohol exposure. This resulted in analyses for 1412 variables. Subjects included 256 with histories of prenatal alcohol exposure and 385 non-exposed subjects. All Phase II sites except South Africa were included at this point. 100 met criteria for FAS. Variables with double-digit sample sizes were not analyzed due to limited data. This eliminated use of brain imaging data and certain item scores. Currently, we have analyzed everything except the parent questionnaire data for which we are still waiting on the collection of variables from other sites. We initially correlated all variables with alcohol exposure (yes/no). In an effort to reduce the number of variables, we retained the 10 variables with the strongest correlations by project. With this approach, we reduced the number of variables to 30 (10 variables x 3 domains, currently excluding brain imaging and parent data). We further reduced the number of variables included by examining their intercorrelations within domain. Redundant (r>.8) variables were eliminated, resulting in 14 variables retained. Using these 14 variables, individual logistic regression analyses were conducted within-domain to determine which variables significantly discriminated between subjects that have prenatal alcohol exposure and those that do not. At this point, we have identified seven variables through these logistic regressions that were able to significantly differentiate subjects. These variables include: a composite score of working memory, number correct on a verbal fluency task within the first fifteen seconds of performance, a similarities task (measure of verbal concept formation), arithmetic ability, presence of camptodactyly, presence of difficulty in pronation/supination of elbows, and SuperAurioleRight_Y (from 3D facial imaging). We will use these 7 variables supplemented by those from other domains in our initial attempts at a hierarchical approach to subject classification (see plans).

Next Steps: CART Analysis Questions Should we start with more variables per domain? Starting with 10 per domain for 3 domains resulted in only 7 non-redundant, discriminating variables. How to handle variables that have relatively small sample sizes (e.g., brain imaging and teacher variables)? Are we using the correct approach to reduce the number of variables? Should we have started with variables that make sense clinically instead of those that best discriminate the groups? Validation that we should not be using physical features consistent with FAS in this analysis. The goal is to identify non-dysmorphic children, thus we chose to exclude those variables that were consistent with this diagnosis. Similarly, should we include or exclude cases without FAS? If our goal is to better identify cases without FAS, perhaps they are not relevant in this analysis. There are currently 100 cases with FAS in the analysis. Should we include or exclude the South African data in this analysis? In the past, these data have proved problematic in terms of generalizability due to demographic differences. In relation to our first specific aim, and the ongoing project described above, correlations between the presence of prenatal alcohol exposure and parent questionnaire data will be conducted once all data is retrieved. This will allow us to determine which variables, based on parent behavioral assessment of children, should be used as the best identifiers in this domain. The same steps will be taken with parent data, including examination of intercorrelations to eliminate redundancy and conducting a logistic regression with these variables. Once all domains are examined and variables are narrowed down to the best identifiers, we plan to conduct a classification and regression tree (CART) analysis to establish cutoff scores that will show us what is most predictive of children with prenatal alcohol exposure that do not meet diagnostic criteria for FAS. Cross validation for the CART will be performed in order to validate our CART model, which will help us estimate how accurately the model will perform clinically. Issues that we are considering in this ongoing project include (1) whether we should start with more variables per domain. Starting with 10 per domain for 3 domains resulted in only 7 non-redundant, discriminating variables. Perhaps increasing our initial catch to 20 variables will all for us to end up with 10 total per domain. (2) How to handle variables that have relatively small sample sizes. Brain imaging and teacher variables are among those that are at issue here. (3) Are we using the correct approach to reduce the number of variables? Should we have started with variables that make sense clinically instead of those that best discriminate the groups? (4) Validation that we should not be using physical features consistent with FAS in this analysis. The goal is to identify non-dysmorphic children, thus we chose to exclude those variables that were consistent with this diagnosis. (5) Similarly, should we include or exclude cases without FAS. If our goal is to better identify cases without FAS, perhaps they are not relevant in this analysis. There are currently 100 cases with FAS in the analysis. (6) Should we include or exclude the South African data in this analysis. In the past, these data have proved problematic in terms of generalizability due to demographic differences.

Goals for upcoming months/year Initiate and continue data collection at all sites. Our goal is to have 100 subjects tested by funding-year end Continue to analyze data from previous phases Proceed with CART analysis (Specific Aim 1)