Thimerosal Exposure in Early Life and Neuropsychological Outcomes 7–10 Years Later

Abstract

Objective The authors used a public use data set to investigate associations between the receipt of thimerosal-containing vaccines and immune globulins early in life and neuropsychological outcomes assessed at 7–10 years. Methods The data were originally created by evaluating 1,047 children ages 7–10 years and their biological mothers. This study developed seven latent neuropsychological factors and regressed them on a comprehensive set of covariates and thimerosal exposure variables. Results The authors found no statistically significant associations between thimerosal exposure from vaccines early in life and six of the seven latent constructs. There was a small, but statistically significant association between early thimerosal exposure and the presence of tics in boys. Conclusions This finding should be interpreted with caution due to limitations in the measurement of tics and the limited biological plausibility regarding a causal relationship.

children, modeling, neuropsychology, public health, structural equation

Introduction

The association between exposure to thimerosal-containing vaccines and developmental outcomes has been debated since 1999 (Bernard, 2008; Clements, Ball, Ball, & Pratt, 2001; Offit, 2007; Rooney, 2008; Sugarman, 2007) when the Food and Drug Administration (FDA) determined that children who received multiple thimerosal containing vaccines at a young age were at risk for exceeding the Environmental Protection Agency’s (EPA) safety limits for methylmercury (Ball, Ball, & Pratt, 2001; Stratton, Gable, & McCormick, 2001). EPA had never determined safety limits for ethylmercuty, the compound found in thimerosal containing vaccines. EPAs methylmercury saftey limits had been determined based on previous studies that found associations between methylmercury exposure and neuropsychological outcomes (Crump, Kjellstrom, Shipp, Silvers, & Stewart, 1998; Grandjean et al., 1999). Specifically, the Faroe Island studies found that high levels of methylmercury exposure due to maternal consumption of mercury-contaminated fish during pregnancy have been associated with children exhibiting lower motor function and verbal skills at 7 years of age (Grandjean et al., 1999). As a precautionary measure, the U.S. Public Health Service recommended the removal of thimerosal from vaccines administered to children early in life and the Centers for Disease Control and Prevention (CDC) proceeded to sponsor several studies investigating the possible associations between exposure to thimerosal-containing vaccines and child developmental outcomes (Thompson et al., 2007; Tozzi et al., 2009; Verstraeten et al., 2003).

In the first study published by the CDC, Verstraeten et al. (2003) analyzed historical medical records from three health maintenance organizations (HMO) and found that higher exposure to thimerosal-containing vaccines was associated with a greater likelihood of the presence of tics in one HMO and was associated with language delay in another HMO. No significant associations were found in the third HMO. The inconclusive findings and the concern regarding the reliability of the case definitions using historical medical records led the CDC to carry out several additional studies using more rigorous and reliable assessment methods. One of these studies assessed children using a 3-hr battery of neuropsychological tests and found several statistically significant associations including associations with verbal language among girls and tics among boys (Thompson et al., 2007). The difficulty in interpreting the results presented in the original paper was that the number of statistically significant findings was no greater than the number expected by chance alone (15 significant findings out of 378 tests). A recent randomized trial by Tozzi et al. (2009) compared children that received 62.5 μg of thimerosal to children that received 137.5 μg within 1 year of birth on 24 neuropschological outcomes. They found that girls with higher thimerosal intake had lower mean scores on the finger-tapping test (dominant hand) and on the Boston Naming Test. Although, it should be noted that the highest amount of thimerosal exposure in this study was considerably lower than the upper limits of exposure found in the Thompson et al. (2007) study (25% received 187.5 μg, within 7 months). Finally, using the same data set as the Thompson et al. (2007) study, a recent study found that receiving vaccines on-time (compared to delayed or not at all) was not related to any negative neuropsychological outcomes after adjusting for thimerosal exposure (Smith & Woods, 2010). This suggests that the timeliness of vaccination does not appear to adversely affect neuropsychological outcomes of children 7–10 years later after controlling for the level of thimerosal exposure.

Recently, the National Vaccine Advisory Committee (NVAC) recommended continued investigation of issues related to the safety of thimerosal-containing vaccines (National Vaccine Advisory Committee, 2009). Specifically, NVAC recommended that additional research be carried out using a public use dataset made available by the CDC to assess in greater detail whether higher exposure to thimerosal is associated with an increased risk for clinically important tics and/or speech and language delays. In addition to NVACs recommendation for additional research, it is important to note that thimerosal continues to be used in many childhood vaccines in other areas of the world, particularly in developing countries (Dorea, 2010), and therefore the value of developing a more thorough understanding of the risks and benefits of thimerosal-containing vaccines is increased.

The current study reanalyzes data from the Thompson et al.’s study using measurement models to assess more efficiently the associations between thimerosal exposure and theoretically meaningful neuropsychological constructs using a subset of the outcomes from the original study (Thompson et al., 2007). The items selected from the original study were primarily based on the previous studies that found significant associations between methylmercury exposure and neuropsychological outcomes (Crump et al. 1998; Grandjean et al., 1999) as well as the results from the initial CDC screening study (Verstraeten et al., 2003). The specific purpose of the current study was to assess whether prenatal thimerosal exposure or thimerosal exposure between birth and 7 months of age were associated with any of seven latent nerodevelopmental constructs for boys or girls. We hypothesized that the neuropsychological outcomes would fit well to seven latent constructs and that there would be latent mean differences in these constructs by gender. We hypothesized despite differences in the latent means, measurement equivalence would be maintained across gender. We also hypothesized that the bulk of the findings found in the Thompson et al. (2007) study were likely due to chance and thimerosal exposure was unlikely to be associated with all seven latent constructs. This is because by utilizing latent constructs, rather than using the original 42 individual items, we increased the reliability of the outcome measures and reduced the Type-I error rate, which was a major weakness of the original study. Further, compared to traditional regression techniques, we used structural equation modeling which substantially reduces the overall number of statistical tests, appropriately models measurement error, and increases precision (Nelson, Aylward, & Steele, 2008; Tomarken & Waller, 2005). Furthermore, multi-group analyses were considered important because previous research suggests that the strength of association between the dependent variables and thimerosal exposure (Thompson et al., 2007; Tozzi et al., 2009), as well as with many of the covariates (Biederman, Faraone, & Monuteaux, 2002; Neisser et al., 1996), differs depending on the gender of the child.

Methods

Participants and Sampling Procedure

This study used data collected by the CDC and first presented by Thompson et al. (2007). The public use data set used for this analysis was obtained through the CDCs Immunization Safety Office (http://www.cdc.gov/vaccinesafety/vsd/thimerosal_outcomes/). Children included in this dataset were enrolled in one of four HMOs from birth to 1 year of age, were 7–10 years of age at assessment and participated in the CDCs Vaccine Safety Datalink (Chen et al., 1997, 2000). Recruitment for Thompson et al.’s (2007) study was attempted for 3,648 children, of which, 1,985 refused or were unable to be located, 512 were deemed ineligible and 44 were unable to be assessed. Of the remaining 1,107 children that were assessed, 1,047 were retained for the final sample. The final sample reduction was due to 24 of the assessed children not having weighed at least 2,500 g at birth, seven not having weight data, 23 having an exclusionary medical condition (e.g., encephalitis, meningitis, or hydrocephalus), five having no prenatal exposure data, and one not having care in a sampled HMO during their first year of life.

The final dataset included background and assessment data for 1,047 children and their biological mothers obtained from historical medical records, maternal interviews, and a 3-hr neuropsychological test battery administered between 7 and 10 years of age (M = 9.28; born between January 1993 and March 1997). The child sample was 51% females and was largely comprised of White (49%), African American (30%), and Latino (14%) children. Mothers’ reported socioeconomic status ranged from 20% of the poverty line to 23 times above the poverty line. Mothers with college degrees (51%) and higher income were overrepresented due in part to their greater access to private health insurance. Further details regarding the original study design and population can be obtained from the previously published study (Thompson et al., 2007) and the online technical report (Price, Goodson, & Stewart, 2007).

Thimerosal Exposure from Vaccines and Immune Globulins

Based on recommendations from the expert panel that oversaw the study design and analysis of data (Price, Goodson, & Stewart, 2007), exposure to thimerosal was measured by identifying the amount of thimerosal contained in immune globulins and vaccines administered prenatally to the mothers, and the amount of thimerosal from immune globulins and vaccines that children received from birth to 7 months of age. Total thimerosal exposure from birth to 7 months was estimated by dividing the amount of mercury in each administered vaccine by the participant’s weight at the time of exposure, and then summing the estimated exposures (M = 19.80, SD = 7.092, range 0–38.27). As a point of reference, without taking into account participants’ weight at the time of exposure, participants, on average, were exposed to a total of 118.18 µg (SD = 41.44; range 0–187.50) of ethylmercury between birth and 7 months of age. The 1995 EPA guidelines for methylmercury exposure from birth to 6 months was 89 µg for those at the 50th percentile for weight suggesting that some children may have exceeded the EPAs safety threshold (Ball et al., 2001). On the other hand, other agencies had significantly higher safety thresholds (U.S. Agency for Toxic Substances and Disease Registry (266 µg), the U.S. Food and Drug Administration (354 µg), and the World Health Organization (417 µg) (Ball et al., 2001) and no children are likely to have exceed those. Since a child’s weight at the time of prenatal exposure was not known, prenatal thimerosal exposure was assessed separately and was not adjusted for weight (M = 2.24, SD = 8.31, ranging from 0 to 100 µg; 111 or 11% of the sample had some prenatal exposure to thimerosal). A full description of the procedures utilized to determine the level of thimerosal exposure can be found in Price et al.’s technical report (2007).

Neuropsychological Test Battery Assessed at Age 7–10 Years

Trained evaluators measured a series of neuropsychological outcomes during a 3-hr testing period with the child. In addition, while the target child was being tested, the mother was interviewed and completed assessments on several additional child outcomes. Teachers of the children also completed assessments of the children’s behavior following the visit to the clinic.

In the original study, there were 42 measures used to assess nine different constructs. These 42 items were selected based on the previous CDC screening study (Verstraeten et al., 2003), previous methyl mercury studies (Davidson et al., 1998; Grandjean et al., 1999), and external expert consultants’ recommendations. For the current analysis, we organized 25 of the measures into the following theoretical constructs based on expert opinion and previous literature: general intellectual functioning, verbal memory, fine motor coordination, executive functioning, behavior regulation, tics, and language (Table I). These latent constructs included subscales from the Wechsler Abbreviated Scale of Intelligence (WASI; Wechsler, 1997), Woodcock-Johnson III (WJIII; Woodcock, McGrew, & Mather, 2001), Boston Naming (Kaplan, Goodglass, & Weintraub, 1983), Neuropsychological Assessment (NEPSY; Korkman, Kirk, & Kemp, 1998), Clinical Evaluation of Language Fundamentals-Third Edition (CELF; Semel, Wiig, & Secord, 1995), California Verbal Learning Test-Children’s Version (CVLT-C; Delis, Kramer, Kaplan, & Ober, 1994), Wechsler Intelligence Scale for Children (WISC; Wechsler, 1974), Gordon Diagnostic System (GDS; Gordon, 1988), Conners’ Rating Scales-Revised (CRS-R; Conners, 2001), and the Behavioral Regulation Inventory of Executive Functioning (BRIEF; Gioia, Isquith, Guy, & Kenworthy, 2000). All of these tests have well-established reliability and validity (Gioia, Isquith, Retzlaff, & Espy, 2002; Mayes & Crowell, 2001; Schmitt & Wodrich, 2004). Additionally, a finger-tapping test (fine motor coordination; Letz & Baker, 1988) as well as observation for phonic and motor tics were completed.

Table I.

Descriptive Statistics for all Latent Outcomes

	Boys n = 509	Girls n = 538
	M (SD)	M (SD)
General intellectual functioning
WASI Vocabulary	52.86 (10.03)	54.66 (10.44)
WASI Similarities	54.52 (9.76)	55.30 (8.78)
WASI Block Design	53.56 (11.55)	50.85 (10.66)
WASI Matrix Reasoning	53.04 (9.85)	53.82 (9.54)
WJIII Letter Word	106.56 (11.91)	107.49 (11.37)
Speech and language
Boston Naming (raw)	39.64 (7.87)	39.44 (8.07)
NEPSY Speed Naming	10.13 (2.82)	10.57 (2.66)
NEPSY Comprehension	10.64 (2.90)	11.02 (2.75)
CELF Formulated Sentences	10.78 (2.79)	11.27 (2.59)
CELF Recall Sentences	10.38 (3.11)	10.97 (3.03)
Verbal Memory
CVLT-C Free Recall, No Delay	52.46 (10.15)	55.51 (9.49)
CVLT-C free recall, short Delay (scaled)	0.28 (0.98)	0.57 (0.90)
CVLT-C free recall, long Delay (scaled)	0.36 (0.96)	0.68 (0.83)
CVLT-C Cued Recall, Short Delay (scaled)	0.32 (1.05)	0.70 (0.90)
CVLT-C Cued Recall, Long Delay (scaled)	0.32 (0.97)	0.68 (0.77)
Attention/Executive functioning
WISC III Digit, Forward (raw)	7.98 (1.89)	8.13 (1.86)
WISC III Digit, Backward (raw)	4.52 (1.67)	4.54 (1.60)
GDS Vigilance Task: Correct Responses	40.28 (5.29)	40.63 (4.97)
Fine motor coordination
Finger Tapping: Dominant Hand	40.17 (7.17)	37.49 (6.12)
Finger Tapping: Nondominant Hand	35.97 (6.63)	32.97 (5.48)
Tics
Motor tics: assessor 0/1	0.12(0.32)a	0.06 (0.24)a
Phonic tics: assessor 0/1	0.09(0.29)a	0.05 (0.22)a
Behavior regulation
CRS-R: Parent Rating Inattentive (scaled)	51.68 (9.89)	52.49 (11.00)
CRS-R: Parent Rating Hyperactive (scaled)	53.68 (10.07)	54.48 (11.57)
BRIEF Parent Rating: Behavior Regulation (scaled)	49.42 (10.98)	48.79 (9.99)

^aThis represents the average number of participants expressing tics on a 0/1 scale. This number can be multiplied by 100 to represent the percentage of children reporting tics; analyses estimating tics scores as continuous or dichotomous yielded comparable results.

All outcome indicators were measured on a continuous scale with the exception of tics, the presence of which was coded as dichotomous variables (0/1). Trained interviewers assessed tics by recording whether motor or phonic tics were present at any point during the 3-hr assessment period. The interviewers received verbal instruction and viewed a 30-min training video (“Tourette Syndrome: A Guide to Diagnosis of TS,” 1989). The presence of tics was also assessed by parent report (presence of tics this week or ever) but we found low cross-validation with assessor reports (positive presence of tics from the parent report of tics in the past week and the assessor report of tics agreed only 23% of the time for motor tics and 16% of the time for phonic tics). Consequently, due to the greater potential for self-report biasing the results (e.g., varying definition and perception by parents), and a recent study among children diagnosed with tics that found clinical observations across three consecutive weeks to be a more reliable method for assessing tics than indirect reporting methods (Himle et al., 2006), we choose to use only assessor observations for this investigation. Descriptive statistics for the neuropsychological tests as well as the tic observations can be found in Table I.

Seventeen of the original 42 measures were not used in the reanalysis because they did not contribute to the measurement of the constructs of interest, were not measured by more than one item, or had large amounts of missing data. For example, measures of WISC forward and backward recall were utilized in the present study as indicators of attention/executive functioning, so the WISC recall combined score, which was also one of the 42 measures included in the Thompson et al. study (2007), was not included to prevent redundancy. Similarly, the measure of correct responses to the GDS vigilance task was included as an indicator of attention/executive functioning in the current study so we did not include the number of errors by the children on this task as a separate indicator. Other measures, such as stuttering (3) were removed because they were assessed by three different raters with little covariation between them. Finally, measures from teachers, such as those included in the CRS (2) and BRIEF (2), were not included due to large amounts of missing data (25–26%).

Covariates

The current study included covariates obtained from both the maternal interview and a review of available medical records. This included the completion of a brief maternal IQ test (The Kaufman Brief Intelligence Test; KBIT; Kaufman & Kaufman, 1990), questions regarding any history of exposure to environmental or diet related toxins, socioeconomic indicators, and the Home Observation for Measurement of the Environment (HOME) inventory (Caldwell & Bradley, 1984). A complete list of covariates used in the model and basic descriptive statistics can be found in Table II. We used a similar set of a priori covariates to the Thompson et al. (2007) study based on prior methylmercury and thimerosal studies, and recommendations from experts, including pediatricians, psychologists, and neurotoxicologists (Price et al., 2007). The inclusion of these covariates was necessary to adjust for known confounding. Specifically, higher thimerosal exposure has been found to be positively associated with higher parental socioeconomic status, primarily because parents of high socio-economic status are more likely to vaccinate their children on-time (Smith & Woods, 2010). Consequentially, failure to adjust for these covariates can lead to the misidentification of positive associations between thimerosal exposure and neurodevelopmental outcomes.

Table II.

Descriptive Statistics for all Covariates and Exposure Measures

	Boys (n = 509)	Girls (n = 538)
	M (SD)	M (SD)
Exposure
Prenatal thimerosal exposure (µg units)	2.43 (9.46)	2.06 (7.06)
Thimerosal exposure birth to 7 months (kg/µg)	18.81 (7.01)	20.74 (7.05)
Covariates
Child age (years)	9.33 (1.09)	9.23 (1.07)
HOME index	11.86 (1.99)	12.10 (1.90)
Percent of poverty line/100	4.11 (2.82)	4.14 (2.38)
Maternal educationa	2.26 (0.88)	2.28 (0.88)
Child lives in a one-parent householdb (0/1)	0.20 (0.40)	0.19 (0.39)
Maternal IQ (KIBITS)	102.42 (13.31)	102.39 (13.22)
Child computer experiencec	1.61 (0.51)	1.53 (0.51)
Maternal aged	2.08 (0.29)	2.05 (0.24)
Having an older siblingc (0/1)	0.64 (0.48)	0.65 (0.48)
Having a younger siblingc (0/1)	0.41 (0.49)	0.40 (0.49)
No. of center-based daycare settings prior to KGe	0.82 (1.02)	0.80 (1.02)
No. of home-based daycare settings prior to KGe	1.51 (1.57)	1.45 (1.31)
English only used in the homec (0/1)	0.63 (0.48)	0.62 (0.49)
Length of time breast fedf	1.18 (0.77)	1.22 (0.78)
Head circumferenceg (0/1)	0.01 (0.11)	0.04 (0.19)
Five min APGAR score	8.91 (0.52)	8.95 (0.46)
Prenatal exposure to toxins (summed)	2.75 (1.33)	2.69 (1.27)

^aMaternal education was assessed as 0 indicating no high school diploma, 1 indicating a high school diploma or GED, 2 indicating some college, and 3 indicating a college degree_.

^b1 was coded as positive response and 0 indicated a negative response.

^c0 indicates no experience, 1 indicates some experience, 2 indicates much experience; ^dMaternal age was assessed with 1 indicating a maternal age between 0 and 16 years, 2 indicated an age between 17 and 39 years, and 3 indicated an age over 40 years.

^eKG, kindergarten.

^f0 indicates breast fed <1 month, 1 indicates 1–6 months, 2 indicates over 6 months.

^g1 indicates a birth head circumference ±2 SD from M.

Three additional dummy coded variables were also included in the analyses to control for the location of which HMO data was collected from.

Statistical Analysis

Similar to the progression of studies for assessing the effects of prenatal methylmercury exposure via fish consumption, which first investigated effects using multiple-regression techniques (Grandjean et al., 1999) and then subsequently used structural equation modeling (SEM) to improve the reliability of the outcomes and reduce Type-I errors (Budtz-Jorgensen, Keiding, Grandjean, & Weihe, 2002), we designed the current study as a follow-up to the regression analyses carried out in the Thompson et al. (2007) study. SEM provides an efficient means to test simultaneously the associations between exposure variables and multiple dependent variables while appropriately modeling measurement error (Kline, 2005; Tomarken & Waller, 2005). Use of factor analysis within the SEM framework also has the advantage of reducing the possibility of Type I errors, which was an inherent limitation of the Thompson et al. (2007) study. The current study was completed by first identifying a valid and well-fitted measurement model composed of seven latent constructs. Next, a multi-group Multiple Indicator, Multiple Indicator Cause (MIMIC) model design (Figure 1) was used to examine independent associations of thimerosal exposure with each of the constructs.

Figure 1.

The structural MIMIC model utilizing seven latent variables. The covariates are represented by one box for display purposes but were entered statistically as individual items.

Results

All endogenous indicators were first screened to inspect their distributional properties. No indicator was notably skewed (skew index > 3.0) with the exception of motor tics, which slightly exceeded this threshold (skew index = 3.29) and no indicators were notably kurtosis (kurtosis index > 8.0) with the exception of motor tics which slightly exceeded this threshold (kurtosis index = 8.87) (Kline, 2005). Using LISREL 8.7 (Joreskog & Sorbom, 2004), a measurement model was first tested using the entire sample without the inclusion of covariates or measures of thimerosal exposure. This was done to test the fit of the hypothesized seven-factor model to the data. The measurement model fit well, χ² (253) = 1061.39, p < .001, RMSEA = .055 (90% confidence interval: 0.052–0.059) (Browne & Cudeck, 1992; Kline, 2005), CFI = 0.97 (Bentler, 1990; Hu & Bentler, 1999), NNFI = 0.96 (Bentler & Bonett, 1980), with all standardized factor loadings (λ) falling between .46 and .93 (Table III). This was followed by testing a single group SEM that regressed the seven latent constructs on the a priori covariates and the two thimerosal exposure variables. This model fit marginally well, χ² (688) = 3634.85, p < .001, RMSEA = .064 (90% confidence interval: 0.062–0.066), CFI = 0.92, NNFI = 0.87. The structural model yielded one significant association between the latent factor for tics and the thimerosal exposure variable for the birth and 7-month period (γ = .11, p < .05). No other significant relations with thimerosal exposure were found. Additional analyses using Mplus 5.2 (Muthen & Muthen, 1998–2007) were then conducted to test whether designating the tic indicators as both a continuous (linear) and dichotomous (logistic) variable affected the association between thimerosal and the observation of tics. These two analyses yielded comparable results (standardized results between models changed from .11, when treated as a continuous variable, to 12 when treated as a dichotomous variable), and due to the availability of approximate fit indices and comparative indices when utilizing all continuous indicators, the authors chose to present findings that treated the tics indicators as a continuous outcome.

Table III.

Standardized Factor Loadings by Sex and Fixed Across Sex

	Boys n = 509	Girls n = 538	Fixed
	Std. λ	Std. λ	Std. λ
General intellectual functioning
WASI Vocabulary	.80	.82	.81
WASI Similarities	.74	.70	.72
WASI Block Design	.52	.50	.52
WASI Matrix Reasoning	.56	.53	.55
WJIII Letter Word	.66	.65	.66
Speech and language
Boston Naming (raw)	.68	.76	.71
NEPSY Speed Naming	.60	.55	.58
NEPSY Comprehension	.70	.66	.69
CELF Formulated Sentences	.70	.65	.68
CELF Recall Sentences	.77	.77	.77
Verbal memory
CVLT-C Free Recall, No Delay	.77	.76	.74
CVLT-C Free Recall, Short Delay (scaled)	.85	.81	.84
CVLT-C Free Recall, Long Delay (scaled)	.85	.82	.86
CVLT-C Cued Recall, Short Delay (scaled)	.90	.91	.90
CVLT-C Cued Recall, Long Delay (scaled)	.85	.86	.87
Attention/Executive Functioning
WISC III Digit, Forward (raw)	.59	.50	.57
WISC III Digit, Backward (raw)	.57	.54	.56
GDS Vigilance Task: Correct Responses	.43	.50	.46
Fine Motor Coordination
Finger Tapping: Dominant Hand	.88	.89	.88
Finger Tapping: Nondominant Hand	.94	.80	.94
Ticsa
Motor tics: assessor 0/1	.56	.56	.56
Phonic tics: assessor 0/1	.61	.62	.61
Behavior regulation
CRS-R: Parent Rating Inattentive (scaled)	.77	.70	.76
CRS-R: Parent Rating Hyperactive (scaled)	.90	.90	.92
BRIEF Parent Rating: Behavior Regulation (scaled)	.84	.71	.80

Note. The residual errors of cued recall, short and long delay, WASI, block design and matrix reasoning, and the WISC III Digit, forward and backward were correlated.

^aindicates the frequency of a positive (0/1) score for the presents of motor or phonic tics during the assessment; Analyses treating tics scores as a dichotomous or continuous yielded comparable results.

The measurement model was then tested again, using a multi-group model to examine sex differences. The model was tested for measurement equivalence (Little, 1997) by first introducing a model that allowed the factor loadings for the seven constructs to vary by sex, χ² (505) = 1291.51, p < .001, RMSEA = .055 (90% confidence interval: 0.051–0.058), CFI = 0.97, NNFI = 0.97, and comparing these results against those in which the factor loadings were forced to be equal across groups (see Table III for a breakdown of factor loadings by sex), χ² (525) = 1356.81, p < .001, RMSEA = .055 (90% confidence interval: 0.051–0.058), CFI = 0.97, NNFI = 0.96, Δ χ² (20) = 65.30, p < .01, ΔRMSEA = .00, ΔCFI = 0.00. Although the change in chi-square was significant, sex differences in the factor loadings were modest (Table III) and changes in key fit indices were very small. These findings support the use of a multi-group model due to a lack of change in key fit statistics (ΔCFI < 0.01) (Cheung & Rensvold, 2002; Little, 1997). Latent mean differences between the sexes were examined and are presented in Table IV. Girls tended to have higher scores on the language and verbal memory factors while boys had higher scores on fine motor skills and were observed having more tics. There were no significant differences between boys and girls on measures of intelligence, executive functioning, and behavior regulation.

Table IV.

Latent Mean Differences Between Boys and Girls

Latent constructs	α	SE
Intelligence	1.07	0.57
Language	0.45**	0.16
Verbal memory	0.32***	0.05
Executive functioning	0.09	0.08
Behavior regulation	0.41	0.64
Fine motor	−2.96***	0.40
Tics	−0.05***	0.02

Note. Reference group is boys. **p < 0.01, ***p < 0.001

Finally, a structural model was tested to find whether the strength of the relationships between exposure to thimerosal and the outcomes of interest differed by sex. This model fit the data well, χ² (1295) = 2869.07, p < .001, RMSEA = .048 (90% confidence interval: 0.045–0.051), CFI = 0.95, NNFI = 0.92. There were no statistically significant effects of thimerosal exposure for boys or girls for six of the seven constructs (Table V). However, thimerosal exposure during the first 7 months of life was associated with the presence of tics in boys (γ = .17, p = .03) between the ages of 7 and 10 years, but not in girls (γ = .05, p = .48). A sex difference in the strength of association between thimerosal exposure and tics was examined by fixing all paths to be equal for boys and girls, χ² (1449) = 3070.59, p < .001, RMSEA = .046 (90% confidence interval: 0.044–0.049), CFI = 0.95, NNFI = 0.93, and then comparing this result with a model in which only the path from thimerosal exposure from birth to 7 months and tics is allowed to be estimated freely, χ² (1448) = 3065.56, p < .001, RMSEA = .046 (90% confidence interval: 0.044–0.049), CFI = 0.95, NNFI = 0.93, Δ χ² (1) = 5.03, p < .05. ΔCFI = 0.00, ΔCFI = 0.00. The significant change in chi-square suggests that the relationship between exposure to thimerosal from birth to 7 months and the presences of tics 7 years later was different for girls and boys but it should be noted that these differences did not substantially change the approximate fit indexes. A graphic presentation of the differences between boys and girls is presented in Figure 2.

Figure 2.

The percent of boys and girls presenting phonic or motor tics, split below and above the median of exposure to thimerosal per kilogram weight from birth to 7 months of life.

Table V.

Exposure to Thimerosal and Covariates’ Relation to Seven Outcome Factors for Boys and Girls

	Intellectual functioning		Speech and language		Verbal memory		Executive functioning		Fine motor coordination		Tics		Behavior regulation
	Boys	Girls	Boys	Girls	Boys	Girls	Boys	Girls	Boys	Girls	Boys	Girls	Boys	Girls
Thimerosal exposure
Prenatal thimerosal exposure	0.04	0.00	0.05	0.05	0.02	−0.04	−0.06	−0.07	−0.05	0.01	0.03	0.02	0.03	−0.06
Exposure birth to 7 months	0.03	0.06	−0.02	−0.05	−0.02	−0.06	0.08	0.08	0.07	−0.05	0.17*	0.05	0.04	−0.01
Covariates
Child age	0.01	0.02	0.15*	0.11*	0.16*	0.06	0.49*	0.59*	0.60*	0.43*	−0.08	−0.07	−0.04	0.07
HOME index	0.04	0.11	0.05	0.12*	0.02	0.05	0.08	−0.07	0.00	−0.08	−0.02	−0.08	−0.10*	0.02
Percent of poverty line/100	0.05	0.05	0.10*	0.02	0.03	0.01	0.06	0.06	0.04	0.02	0.09	−0.04	0.03	−0.04
Maternal educationa	0.04	0.13*	−0.06	0.13*	−0.13*	0.09	−0.03	0.13*	0.02	0.06	0.12	0.03	0.02	−0.10
Child in a one-parent houseb	−0.10*	−0.10*	−0.05	−0.10*	−0.03	−0.06	−0.04	−0.02	0.11*	−0.08*	0.00	0.04	0.10*	0.06
Maternal IQ (KIBITS)	0.37*	0.20*	0.47*	0.20*	0.22*	0.02	0.30*	0.19*	−0.03	0.01	−0.13	−0.01	−0.06	0.04
Child computer experiencec	0.16*	0.07	0.14*	0.04	0.01	0.00	0.14*	−0.02	0.08*	0.10*	0.01	0.13*	−0.04	−0.02
Maternal aged	0.00	0.06	0.00	0.08	0.02	0.01	0.00	−0.01	−0.09*	0.05	−0.03	0.06	−0.03	0.05
Having an older siblingc	−0.17*	−0.13*	−0.14*	−0.13*	−0.10	0.06	0.05	0.02	−0.03	−0.01	−0.01	0.03	0.11*	0.07
Having a younger siblingc	−0.05	−0.16*	0.02	−0.09*	0.12*	0.02	0.06	−0.06	−0.09	−0.02	−0.09	0.05	−0.02	0.05
No. of center daycare prior to KG	0.01	0.08	0.04	0.05	−0.06	0.00	0.02	0.00	−0.12*	−0.03	0.05	0.01	0.13*	0.07
No. of home daycare before KG	−0.02	−0.04	−0.02	−0.07	−0.06	−0.05	−0.01	0.02	0.07	−0.03	0.05	0.17*	0.08	0.09
English only used in the homec	−0.14*	−0.05	−0.02	0.07	−0.13*	0.06	−0.05	−0.05	−0.09*	0.00	0.01	0.02	0.05	0.09
Length of time breast fede	0.17*	0.25*	0.18*	0.18*	0.02	0.04	0.23*	0.11*	0.05	0.03	−0.02	−0.07	0.00	−0.02
Head circumferencef	−0.01	−0.02	−0.06	−0.03	−0.19*	−0.07	−0.06	−0.05	0.05	−0.04	0.13	0.00	−0.04	0.05
Five minute APGAR score	0.02	0.04	0.02	0.05	0.03	0.04	0.04	0.17*	−0.06	0.02	−0.09	0.01	0.01	0.01
Prenatal exposure to toxins	−0.06	−0.02	−0.03	−0.02	0.00	0.04	−0.06	−0.07	0.01	−0.01	0.02	0.04	0.05	0.09
Total R2	0.40	0.38	0.46	0.45	0.12	0.07	0.44	0.63	0.38	0.34	0.14	0.08	0.09	0.07

Note. KG = kindergarten; Statistically significant (p < .05) associations are bolded.

^aMaternal education was assessed as 0 indicating no high school diploma, 1 indicating a high school diploma or GED, 2 indicating some college, and 3 indicating a college degree_.

^b1 was coded as positive response and 0 indicated a negative response.

^c0 indicates no experience, 1 indicates some experience, 2 indicates much experience.

^dMaternal age was assessed with 1 indicating a maternal age between 0 and 16 years, 2 indicated an age between 17 and 39 years, and 3 indicated an age over 40 years.

^f0 indicates breast fed <1 month, 1 indicates 1–6 months, 2 indicates over 6 months.

^g1 indicates a birth head circumference ± 2 SD from M; three additional dummy coded variables were also included in the analyses to control for the location of which HMO data was collected from; all relationships presented are standardized estimates.

*p < .05.

Discussion

The current study utilized multi-group SEM to test simultaneously the associations between exposure to thimerosal-containing vaccines and seven latent neuropsychological outcomes. This is the first known investigation to assess the association between thimerosal exposure and neuropsychological outcomes using SEM techniques and the use of factor analytic methods significantly reduced the potential for Type I errors. This is a marked improvement in the statistical methodologies employed in previous studies using the same dataset (Smith & Woods, 2010; Thompson et al., 2007).

We found no support for an association between thimerosal exposure from vaccines and immune globulins administered between birth and 7 months for six of the seven neuropsychological constructs we examined. We did find one statistically significant association between exposure to thimerosal-containing vaccines and the presence of tics among boys, however, this association was not replicated in girls. Previous associations between thimerosal containing vaccines and tics were found by Verstraeten et al. (2003) and Andrews et al. (2004) but the findings were not sex specific. Our tic finding was also consistent with the tic finding reported in the original study (Thompson et al., 2007).

The results of this study were consistent with two previous studies that reported an association between tics and thimerosal exposure in early life (Kurlan et al., 2001; Verstraeten, et al., 2003), but differed from a recently published study that reported no significant tic findings (Tozzi et al., 2009). Differences in these findings may be due to the much lower prevalence rates for tics in the latter study; the Tozzi et al. (2009) study identified motor tics in < 3% of the children and phonic tics in < 1% of the children (Tozzi et al., 2009). This suggests that the sensitivity of their tic measures was low and potentially unreliable. Differences between the results of these studies could also be due to differing levels of thimerosal exposure; the children in the Tozzi et al.’s, 2009 study only received a maximum of 137.5 μg of thimerosal before 12 months of age, while 25% of the children in the current study were exposed to significantly higher levels of thimerosal (e.g. up to 187.5 μg, within 7 months).

There were several limitations associated with our study. First, although the creation latent constructs resulted in reducing the likelihood of type I error, the strategy also reduced our ability to detect effects on specific indicators of those constructs; it is possible that specific outcomes (indicators in our model) have unique associations with the exposure variables that are not found in other indicators. Second, because this study did not examine all possible outcomes, it was not possible to rule out several of the other statistically significant associations from the previous study because these measures did not have multiple indices available for analysis and they were not theoretically related to the factors that we assessed. Third, the response rate was relatively low with only 30% of the subjects agreeing to participate and complete the study. Putting this potential bias into context, the time commitment for bringing in a child for a 3-hr evaluation was probably more difficult for single parent mothers from low SES homes who might have difficulty finding child care arrangements for their other children during the time that their child was being evaluated. This also may have resulted in greater enrollment of affluent families with available time and interest in participating in the research study. Finally, because this study excluded subjects born with a low birth weight and other confounding medical conditions, we may have excluded the children who were most vulnerable to the effects of thimerosal exposure. This bias would likely have caused the size of the effects to be smaller and less likely to be statistically significant. While this study design issue was necessary to validate the interpretation of the results, it does not allow for generalization of these findings to all populations.

This study was also limited by the relatively crude measurement of tics. All the other outcome measures assessed in this study used reliable and valid measures that have published manuals, which allowed the researchers to provide feedback to parents of the child. Furthermore, all testers underwent a 2-day training session and required them to reach a specific level of reliability in terms of administering the tests appropriately as documented in the published assessment manuals. The tics assessments, however, carried out by the testers did not require them to meet any reliability criteria and the testers had no prior training in neurology or tic assessments. The only training the testers received for tic assessments was based on viewing a 30-min training video (“Tourette Syndrome: A Guide to Diagnosis of TS,” 1989).

We chose to not include the parent self report for tics because cross-validation of parent self-report with those provided by the trained assessors was low and we believed that the assessment of tics by trained assessors was the more valid and reliable measure for this outcome; it should be noted that we had no means of validating either measure against other clinical reports. Furthermore, all expressions of tics during the assessment period were lumped into one of two categories, verbal or motor tics, leaving no ability for the research to differentiate between different types of motor tics, such as eye blinking or shoulder tics. Despite these limitations, the prevalence rates of tics reported in this study were similar to prevalence rates found in a recent study of tics in schoolchildren (Kurlan et al., 2001) and are unlikely to result in spurious findings.

Although we found a significant association between thimerosal exposure in early life and the presence of tics in boys age 7–10 years, interpretation of this finding is difficult for several reasons. First, the magnitude of the potential contribution from early thimerosal exposure in the present study was relatively small. Second, there is evidence that tic disorders are familial with a complex genetic mode of inheritance (Tourette Syndrome Association Internation Consortium for Genetics, 2007), although the level of heritability appears to depend upon the class of tic-like symptoms (Grados, Mathews, & the Tourette Syndrome Association International Consortium for Gentics, 2008). It is possible that there may be gene–environement interactions associated with the disorder but this would require an alternative study designs to assess. Third, no significant relationship was found between thimerosal exposure and tics among girls. Although tics are less prevalent in girls than in boys, there is limited evidence that suggests there would be etiologic differences based on gender. Despite this, it is possible that the association between thimerosal exposure and certain subtypes of tics disorders would be stronger for boys than for girls if the probability of expressing certain tic-like symptoms were gender specific.

The current analysis found no association between thimerosal exposure and latent neuropsychological outcomes with the exception of tics among boys. The overall results suggest that the bulk of the statistically significant findings presented by Thompson et al. (2007) were most likely due to chance alone. This is particularly important due to the continued use of thimerosal as a vaccine preservative in many countries throughout the world (Dorea, 2010). Furthermore, this study may have implications for the use of some inactivated influenza vaccines administered in the United States that still contain thimerosal. If future studies of tics are undertaken, the interpretation of the findings could be improved by incorporating more reliable and valid measures of tics. Specifically, future tic studies would benefit from the inclusion of validated indirect measures of tics, such as the Autism-Tics, Attention deficit/hyperactivitiy disorder and other Comorbidities (A-TAC; Halleröd et al., 2010; Hansson, Svanstrom Rojvall, Rastam, Gillberg, & Anckarsater, 2005; Larson et al., 2010) and increase the reliability of clinical impressions by having trained professionals observe participants for longer periods and across different settings. Furthermore, these studies should explore the potential for gene–environment interactions, particularly among boys (Lit, Enstrom, Sharp, & Gilbert, 2009).

In conclusion, similar to the Thompson et al. (2007) study, we found that the weight of the evidence in this study does not support a causal association between early exposure to mercury from thimerosal-containing vaccines and immune globulins administered prenatally or during infancy and neuropsychological functioning at the age of 7–10 years for the vast majority of the outcomes examined in this study. Given that the association between thimerosal and tics has been replicated across several different studies, it may be informative to consider additional studies examining these associations using more reliable and valid measures of tics.

Acknowledgments

Disclaimer: The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.