VAERS Reports

The National Childhood Vaccine Injury Act mandates research to improve the safety of vaccines. In the United States, adverse events following vaccination are monitored through a passive surveillance system known as the Vaccine Adverse Events Reporting System (VAERS) co-administered by the CDC and the FDA. Reports are voluntarily submitted to VAERS by health care providers, vaccinees, manufacturers, and others, and are accepted into VAERS regardless of whether the vaccine plausibly caused the adverse event (Chen, 1994; Singleton et al., 1999; Varricchio, 1998).

VAERS' strengths are its national scope and its potential to detect in a timely manner rare adverse events that were not identified during prelicensure clinical trials (Singleton et al., 1999). The centralized reporting enables detection of events that might otherwise go unnoticed. For instance, VAERS quickly detected cases of intussusception following release of the rotavirus vaccine, which resulted in the voluntary removal of the vaccine from the market. The system can also be used to monitor vaccine lot-specific safety (Varricchio et al., 2004).

While VAERS is useful in detecting some rare adverse events and generating hypotheses, such a passive surveillance system has many limitations. For one thing, it is subject to underreporting and general reporting bias because of its reliance on voluntary reporting. For another, many reports are incomplete; because they are not formal case reports, information is not presented in standardized fashion. Other limitations include inadequate data on the “denominator” (or total population receiving the vaccine), and lack of an unbiased comparison group. Therefore it is usually not possible to determine causal associations between vaccines and adverse events from VAERS reports nor can VAERS be used to calculate incidence or prevalence of an adverse reaction (Varricchio et al., 2004).

From January 1, 1990, through March 1, 2004, VAERS received approximately 167,340 adverse-event reports for all vaccines (FDA, 2004a); 1,348 of those reports included autism as an adverse event following vaccination (FDA, 2004b). Ninety-three percent of reports that listed autism as an adverse event were reported between 1999 and 2004 (FDA, 2004b), much in the same period in which publicity increased regarding the putative links between MMR vaccine and autism and TCVs and autism. Ten of the 1,348 suggest a rechallenge, but only five of these describe a rechallenge followed by autistic symptoms (FDA, 2004c). The other five describe a recurrence of other reactions such as fever and flu-like symptoms (FDA, 2004c). The committee reviewed these reports, but concluded that they were not informative on the issue of causality, given the analytic limitations of VAERS.

Thimerosal-Containing Vaccines and Autism

Controlled Observational Studies

Denmark. Hviid and colleagues (2003) compared rates of autism and other ASDs in Danish children who received TCV with those who received thimerosal-free vaccine. The study was based on data from the Danish Civil Registration System and other national registries, including the Danish Psychiatric Central Register, data from the National Board of Health, and the National Hospital Discharge Register. (The Danish Civil Registration System is a nationwide register that assigns unique identifier codes to new residents and all individuals born in Denmark and stores demographic information for each individual.) This unique identifier is used to link data in the Danish Civil Registration System to individual information stored in other national registries (Madsen et al., 2002).

The whole-cell pertussis vaccine, produced by Denmark's Statens Serum Institut,⁸ was the only TCV recommended in the Danish immunization schedule and was in use from 1970 to 1992. The vaccine was replaced with a thimerosal-free version, which was used until January 1997. This vaccine was itself replaced with acellular pertussis vaccine (which never contained thimerosal) and has been in use ever since. Both versions of the whole-cell pertussis vaccine were administered to children at 5 weeks, 9 weeks, and 10 months of age until they were replaced. The first dose of the TCV contained 50 μg of thimerosal (~25 μg EtHg). The second and third doses each contained 100 μg of thimerosal (~50 μg EtHg).

Information on all children born in Denmark between January 1, 1990, and December 31, 1996, was obtained from the Danish Civil Registration System. Information on receipt of the whole-cell pertussis vaccine was obtained from the National Board of Health. Doses received before June 1, 1992, were considered to contain thimerosal, and doses received after June 1, 1992, were deemed thimerosal free. Children who received thimerosal-free vaccine after one or two doses of TCV were classified according to receipt of TCV. The Danish Psychiatric Central Register provided information on cases diagnosed with autism or other ASDs. Autism and ASD cases were ascertained using the ICD-10 coding scheme. From 1991 to 1994, only inpatients were included in the Danish Psychiatric Central Register. Since 1995, both inpatients and outpatients have been included. The National Hospital Discharge Register provided information on other conditions relevant to autism, including tuberous sclerosis, Angelman syndrome, Fragile X syndrome, and congenital rubella. The Danish Civil Registration System and the Danish Medical Birth Registry provided information on possible confounders.

Follow-up was measured using person-time. Follow-up time began at 1 year of age or January 1, 1991 (whichever occurred last) to diagnosis of autism, other ASDs, or other relevant disorders (tuberous sclerosis, Angelman syndrome, Fragile X syndrome, and congenital rubella), death, disappearance, emigration, 11 years of age, or until December 31, 2000 (whichever occurred first). Poisson regression was used to estimate the rate ratios of autism or other ASDs in those receiving thimerosal-containing whole-cell pertussis vaccine to those receiving thimerosal-free whole-cell pertussis vaccine. In addition, the dose-response relationship between TCVs and autism or other ASDs was estimated as the increase in the rate ratio per 25 μg of EtHg.

A total of 467,450 children were born in Denmark between January 1, 1990, and December 31, 1996.⁹ Of this total, 446,695 children received at least one dose of whole-cell pertussis vaccine. For the cohort receiving thimerosal-free vaccinations, there were 1,660,159 person-years at risk. For the cohort receiving TCVs, there were 1,220,006 person-years at risk. Altogether, 407 children were diagnosed with autism, of which 303 children had received thimerosal-free vaccine and 104 received TCV. Some 751 children were diagnosed with other ASDs, of which 430 children received thimerosal-free vaccine and 321 children received TCV. The adjusted rate ratio for autism (adjusting for age, calendar period, child's sex, child's place of birth, birth weight, 5-minute Apgar score, gestational age, mother's age at birth of child, and mother's country of birth) was 0.85 (95% CI, 0.60-1.20). The adjusted rate ratio for other autistic spectrum disorders was 1.12 (95% CI, 0.88-1.43). There was no increase in the rate ratio for autism per 25 μg of EtHg (0.98 [95% CI, 0.90-1.06] or in the rate ratio for other ASDs per 25 μg of EtHg (1.03 [95% CI, 0.98-1.09]).

The authors reanalyzed the data to examine the possibility of misclassification bias (by excluding children vaccinated between June 1, 1992, and December 31, 1992), the robustness of the results (by including only children born between 1991 and 1993), and the impact of missing values (by replacing a missing value with the most common value of the variable). The results from the reanalysis were similar to the original findings. The authors noted that the validity and completeness of ASDs diagnosed in the Danish Psychiatric Central Register was high. They also mentioned that the diagnoses of other related conditions in the National Hospital Discharge Register may not be complete, but the lack of completeness is unlikely to seriously confound an association between thimerosal content and ASDs. The authors raised another limitation in that the date of diagnosis, not date of symptom onset, was used to calculate incidence rates, but they stated that because the study examines risk factors and not autism incidence, this may not be a problem. Overall, the authors concluded that the results did not demonstrate an association between TCVs and autism and other ASDs, and there was no indication of a dose-response association between autism and the amount of ethylmercury received through thimerosal.

The committee considered the study as having strong internal validity with the findings demonstrating a secular trend increase in autism, even after the removal of thimerosal in the vaccines used in Denmark. The committee identified a few limitations of the study, including its time-series design and the generalizability of the study's findings to the U.S. situation, especially with regard to the different dosing schedule used in Denmark and the relative genetic homogeneity of the Danish population.

Vaccine Safety Datalink (VSD). Verstraeten and colleagues (2003) conducted a retrospective cohort study to examine whether certain NDDs, including autism, are related to exposure to TCVs. Because of the topic of this report, the committee focuses just on the analyses of the autism outcome. The study was based on data from the Vaccine Safety Datalink (VSD), a large-linked database that includes vaccination, clinical, hospital discharge, and demographic data. The VSD, formed as a partnership between CDC and eight health maintenance organizations (HMOs), was initiated in 1991 and covers approximately 2.5 percent of the U.S. population (Verstraeten, 2001).

The study was conducted in two phases. Phase I was designed to screen data for potential associations between exposures to mercury from TCVs and selected neurodevelopmental outcomes, including autism. Phase II was designed to confirm the hypotheses generated in the first phase. The autism outcome was examined only in Phase I.

In Phase I, the study population included children born between January 1992 and December 1998 who were enrolled continuously during their first year of life in either HMO A or HMO B and who received at least two polio vaccinations by 1 year of age. HMO A included outcomes available throughout the study period, and HMO B included data available beginning in January 1995 with follow-up data available through the end of 2000. Excluded from the study were infants born at a weight of <2500 g and infants with diagnoses of congenital or severe perinatal disorders, or born to mothers with serious medical problems of pregnancy. Of the 23,241 infants born in HMO A and 229,285 infants born in HMO B between 1992 and 1998, 13,337 from HMO A and 110,833 from HMO B met the eligibility criteria for inclusion in the analyses.

For each child, cumulative vaccine-related mercury exposure was calculated at the end of the first, third, and seventh months of life. Exposure estimates were based on the mean ethylmercury content of each vaccine in multidose vials: each vaccine that a child received was assumed to contain the mean amount of thimerosal reported by manufacturers to the FDA. Thus the ethylmercury content per dose of each childhood vaccine was assumed to be as follows: diphtheria-tetanus-pertussis (whole-cell or acellular), 25.0 μg; hepatitis B, 12.5 μg; and Haemophilus influenzae type b, 25.0 μg. The level of mercury exposure was categorized in increments of 12.5 μg. The authors also performed analyses in which exposure was modeled as a categorical variable to show the change in risk for each level of exposure and to check the linearity assumption in the analyses examining exposure as a continuous variable. At 3 months, exposure categories were 0-25 μg, 37.5-50 μg, and ≥ 62.5 μg; and at 7 months the exposure levels were 0-75 μg, 87-162.5 μg, and ≥ 175 μg. The outcomes of interest in the study included a range of neurological disorders as defined by specific ICD-9 diagnostic codes. The committee focused on the autism outcome (ICD-9 299.0).

Cox proportional hazard models estimated the risk of autism at each level of mercury exposure. The follow-up time period for each child in HMO A started at the first birthday and ended at the date of diagnoses or last date of follow-up. The follow-up time period for HMO B was either the first birthday or January 1, 1995 (whichever occurred later) to the date of diagnoses or last date of follow-up. For HMO A, the analysis was stratified by sex, year, and birth month. For HMO B, the analysis was stratified by sex, year, birth month, and clinic with most visits. To obtain sufficient power to detect any association, the investigators decided a priori to evaluate the effect of thimerosal exposure on outcomes with 50 or more cases.

For HMO A, there were 21 reports of autism, with a median age of 49 months. Based on a priori considerations, the relative risk for autism was not calculated in the HMO A cohort. For HMO B, a total of 202 children (median age was 44 months) were diagnosed with autism. Computerized diagnosis of autism was verified by reviewing medical charts of 618 children from both HMO A and B. In HMO A, autism diagnosis was confirmed in 92.3 percent of the records reviewed. In HMO B, autism diagnosis was confirmed in 81.3 percent of the records reviewed.

The relative risk of autism by increase of 12.5 μg of Hg exposure from TCVs is shown in Table 3. Relative risks were also calculated for cumulative exposure at 3 and 7 months (using 0-25 μg as a reference dose), and they are shown in Tables 4 and 5. Because no significant associations were found between autism and exposure to TCVs in Phase I, no further analysis was conducted of the possible relationship between autism and TCV in Phase II of the study. Overall, the study's findings showed no association between TCVs and autism.

TABLE 3

Relative Risk of Autism by Increase of 12.5 μg of Mercury Exposure from Thimerosal-containing Vaccines at HMO B.

TABLE 4

Relative Risk of Autism by Category of Cumulative Mercury Exposure at 3 Months in HMO B.

TABLE 5

Relative Risk of Autism by Category of Cumulative Mercury Exposure at 7 Months in HMO B.

Additional analyses were conducted to examine other possible study limitations. This included evaluating the effect of the study exclusion criteria—children not continuously enrolled during the first year of life; low-birth-weight (LBW) children; children with congenital or severe perinatal disorders or born to mothers with serious medical problems of pregnancy; and children who received fewer than two polio vaccinations in the first year of life—on the autism results. Although the data were not shown, the authors report that there was no appreciable effect of the exclusion criteria on the autism results. The authors also noted that the confirmation rate between the computerized diagnoses of autism was high for case ascertainment. Some other limitations noted by the authors included possible underreporting of some vaccinations. Other studies (Mullooly et al., 1999) have estimated that 18 percent of hepatitis B virus (HBV) vaccinations may have been missed in HMO A and 2 percent of HBV missed in HMO B. For other vaccinations, 2 percent is estimated to be missed for Hib and DTwP in HMO B, and 10 percent for DTwP and 9 percent Hib in HMO A. At the committee's meeting, Davis (2004) mentioned plans for a case-control study to examine the possible association between cumulative exposure to thimerosal in vaccines or RhoGAM from the prenatal period up to seven months of age, and autistic disorder.

Several concerns have been raised regarding the findings of the study by Verstraeten and colleagues (2003). The committee examined the published paper by Verstraeten and colleagues (2003), the letters in response (Verstraeten, 2004), and criticisms of the study in detail (SafeMinds, 2000, 2003, 2004a,SafeMinds, b,c). Because of the controversy surrounding this study, the committee discusses these concerns, specifically those related to the autism findings, and describes the authors' responses to these allegations in some detail. (Most concerns were addressed by the authors in the published paper [Verstraeten et al., 2003] or at the committee meeting in February [Davis, 2004].)

One overarching concern expressed by some is that the study design and process may have been manipulated in various ways to reach a negative association (SafeMinds, 2000, 2003, 2004a,SafeMinds, b,c). In response, Dr. Davis described the chronology of the VSD study and discussed the differences between the preliminary and final results (Davis, 2004), specifically the relative risks of autism showing a positive but nonsignificant association in earlier analyses versus the relative risk showing no association in the final published study. The difference in preliminary results can be attributed to three major reasons: investigators updated datasets with extended follow-up periods, which allowed for additional cases to be identified; they modified exclusion criteria based on scientific input from the IOM report (IOM, 2001b) and CDC and VSD investigators; and they improved adjustments for health-care-seeking behavior. Other reasons cited for the differences were a modification to the time of exposure, and inclusion of additional variables in the model (Davis, 2004).

The committee notes that it is commonplace for large and important studies to be reviewed along the way, with adjustments often made to improve the eventual validity of the results; thus, it finds nothing inherently troubling in the fact that the VSD study underwent this process. The committee also notes that preliminary results are often misleading and can change substantially as methods are adjusted and more cases and controls are assembled. Indeed, the fact that a conference was held to discuss preliminary findings (Simpsonwood Transcript, 2000) would typically be interpreted as an attempt by the researchers and their sponsors to “get it right,” given the high level of interest in the findings.

In these kinds of conferences, it is common for participants to speak candidly of strengths and weaknesses in the research that, if taken out of context, may be misleading. The term “bias,” for example, has a specific meaning in research that simply refers to ways in which factors may lead to systematic errors in the true effect estimates, where in lay terms “bias” usually refers to a conscious attempt to distort findings. This type of dual-meaning may lead to confusion and an interpretation that has an entirely different meaning than was intended and understood by the participants. The committee examined the justifications for the analytical approaches and changes made during the VSD study process and found that they conformed to standard epidemiological practice.

Other concerns have been raised regarding specific study methods, including choice of inclusion and exclusion criteria, ascertainment of the outcome (autism) and exposure (thimerosal), potential bias due to health-care-seeking behavior, the dose-response relationship between autism and thimerosal exposure, potential confounders, the limitations of automated databases, and other issues (SafeMinds, 2000, 2003, 2004a,SafeMinds, b,c). These are each discussed in turn below:

Exclusion Criteria. Concerns about the appropriateness of the exclusion criteria used in the VSD study have been raised, specifically regarding how excluding subjects may have led towards a bias to the null hypothesis—i.e., shifting the estimates of the size of an association between vaccines and autism toward a modest one or none at all. Subjects were excluded if they did not have documentation in the HMO databases of receiving at least two polio vaccines by 1 year of age, and the authors reasoned that the exclusion criteria ensured that subjects were actually users of the HMO. Subjects with congenital or perinatal disorders or who were born to mothers with serious medical problems during pregnancy, and infants with low birth weight (less than 2,500 g) were also excluded because these outcomes are associated with autism and could confound the study's results.

Dr. Davis presented a reanalysis of the HMO B data that examined the effect of the exclusion criteria on the autism risk estimates. Risk estimates were calculated by including subjects who were originally excluded from the analysis. The analysis was based on an earlier cohort from HMO B, which included 150 subjects (about 75 percent of the size of the final cohort). The data were also presented to the committee at their 2001 meeting. Table 6 depicts the unadjusted relative risk estimates by increase of 12.5 μg of ethylmercury based on exclusion criteria. The risk estimates for autism in the final study cohort did not appreciably differ from the risk estimates of groups after removing exclusion criteria. Autism was not examined in the published LBW analysis.

TABLE 6

Effect of Exclusion Criteria, Unadjusted Relative Risk of Autism by Increase of 12.5 μG of Ethylmercury, HMO B (based on earlier analysis).

Autism Diagnosis. Concerns about the ascertainment of cases with autism were raised. To assess the validity of the computerized diagnoses, the authors reviewed 120 medical records of autism cases with an ICD-9 diagnosis and found that 81 to 92 percent of the diagnoses were made by either a clinician or behavioral specialist. The relative risks for autism by increase of 12.5 μg of ethylmercury in chart-validated cases were calculated. There was no evidence of an increased risk for autism. See Table 7.

TABLE 7

Unadjusted Relative Risk by Increase of 12.5 μg of Ethylmercury Chart Verified Diagnosis.

Several concerns were raised about the possibility of misclassification of cases with autism because of the way the age of the child was handled in the analyses. One issue related to the published study including different age groups and with varying periods of follow-up time for each subject, leading to the possibility that some cases of autism may have been missed with shorter follow-up. In the analysis, the data were adjusted for month and year of birth and time of follow-up. The statistical-analysis technique should therefore take care of this concern. Another related concern was that inclusion of a younger group (who are less likely to be diagnosed with autism) in the study would bias the thimerosal effect toward zero (or in other words, an odds ratio of 1). Adjusting for age would reduce, but not eliminate, this tendency. However, if there were an effect of thimerosal, one still would anticipate a trend of increasing effect with age—that is, increasing odds ratios with age, and each odds ratio exceeding 1.0, albeit with wide confidence intervals. In this study, there was no such association, even in the older age groups.

Another concern was that children at younger ages with early symptoms of autism might be incorrectly diagnosed with other conditions, such as speech or language delay or “misery disorder.” The authors attempted to address this by determining the association between thimerosal and neurodevelopmental outcomes and found no consistent significant associations. However, if there are multiple pathways leading to these disorders, it would be difficult to detect the effect of any one cause—unless it occurred with high frequency and the sample size was large—because the tendency of misclassification of outcome is to dilute measures of effect such as the odds ratio.

Thimerosal Exposure. The authors raised the possible misclassification of thimerosal exposure in using data from the VSD. They cited a study by Mullhooly and colleagues (1999) that evaluated the reliability of automated vaccination data in the VSD. In HMO B, 2 percent of TCVs, specifically neonatal HBV and both DTwP and Hib vaccines, may have been missed. This type of misclassification could lead to a bias toward the null hypothesis; however, in view of the small amount of missing data, the effect is likely to be small.

Health-Care-Seeking Behavior. In their published paper, the authors raised a general concern about how care-seeking behavior would affect the study's results. Speech therapy is not covered in HMO B, and the authors suggested that providers may be less likely to screen for speech and language disorders among young children, and that parents may be a more important factor in the ascertainment of these disorders. Thus parents who are more concerned with neurodevelopmental delays may also be more likely to adhere to a timely vaccination schedule, leading to falsely elevated estimates between thimerosal exposure and certain disorders. To address the effect of health-care-seeking behavior on the risk of autism, Dr. Davis presented an analysis to the committee that implicitly adjusted for care-seeking behavior. They limited this analysis to those who had a complete vaccination schedule—that is, similar health-seeking behavior—but were administered DTP separately versus combined. The combined vaccine has less thimerosal. Using data on autism cases from HMO B (n = 150), they calculated the relative risk of autism between those who received DTP separately (DTP and Hib administered separately) versus those who received DTP combined (combined DTP and HIB administered). In terms of a 25 μg difference in exposure, there was no evidence of an increased risk.

Dose-Response Relationship. In Dr. Davis' presentation, the relative risk of autism was plotted against the levels of cumulative ethylmercury exposure from TCVs administered during the first 7 months of life at HMO A and HMO B from 1992 to 1999. The confidence intervals of the relative risks all went through 1. Dr. Davis noted that there is no evidence of dose-response or an increased risk of autism at any particular exposure.

Autism Incidence. Concerns have been raised that the use of VSD underestimated the true autism incidence because the study only captured those who sought medical attention. It is important to note that the study was designed to examine the relative risk of developing NDD (including autism) at increasing levels of thimerosal exposure, not to examine the incidence of autism and other NDDs.

Other analyses using VSD. In a separate unpublished analysis of the VSD data, Geier and Geier (2004b) examined the relationship between thimerosal-containing DTP vaccines and autism. Children were included if they received four doses of the DTaP vaccine. These inclusion criteria were used to limit the possibility of differences in health-seeking behavior and to ensure that children were old enough to be diagnosed with an NDD. Overall, 28,056 children received thimerosal-free DTaP vaccine and 55,335 children received only thimerosal-containing DTaP vaccines. The authors analyzed both inpatient and outpatient diagnoses, discarded duplicate cases, and only included children with ICD-9 299.0 (autistic disorder) diagnostic code.

An incidence rate was calculated for each cohort, and the relative risk was determined by dividing the incidence rate in the thimerosal-containing DTP cohort by the incidence rate in the thimerosal-free DTP cohort. The authors “determined the attributable risk by subtracting the incidence rate of each condition examined in the cohort that only received thimerosal-containing DTaP vaccine from the incidence rate in the cohort that only received thimerosal-free DTaP vaccine.”

To determine statistical significance, the authors used likelihood ratio chi-square and Fisher's exact test. The authors' assumption is that “the number of conditions and the number of children in the cohort receiving thimerosal-containing DTaP are the observed values. In this analysis, the statistical package contained in Simple Interactive Statistical Analysis (SISA) and Fisher's exact were used, and a double-sided p-value < 0.05 was accepted as statistically significant. In addition, p-values were determined.”

They report a relative risk for autism of 9.6 (attributable risk = 3.04 per 10,000 children, p < 0.005). To check for possible biases, the authors examined the hypersensitivity diagnosis (ICD-9 995.2). They assumed that because of a correlation between hypersensitivity and thimerosal, children who received thimerosal-containing DTaP would have a higher incidence of hypersensitivity. They found a hypersensitivity relative risk of 1.8 (attributable risk = 26 per 10,000 children, p < 0.0001) when comparing the risk between children who received thimerosal-containing DTaP and those who received thimerosal-free DTaP. “An overall sampling of diagnoses made in the thimerosal-containing DTaP vaccine cohort in comparison to the thimerosal-free DTaP vaccine cohort was designed to determine whether there were significant differences in the health-care-seeking behaviors of children in the vaccine cohorts, because such a difference could account for significant differences in diagnoses within each cohort examined.” The relative risk for health-care-seeking behavior in those who received thimerosal-containing DTaP vaccine was 0.77 (p < 0.0001).

Dr. Geier and Mr. Geier presented to the committee an unpublished analysis of the VSD data. Only one slide depicted this information, and it demonstrated an increasing relationship between autism relative risk and amount of thimerosal. The basis of this calculation was not provided and additional data and methods were not described. Overall, the committee found the results of their analyses using VSD data uninterpretable, primarily due to the lack of a complete description of their methods, specifically in how they determined whether individuals belonged to the unexposed or exposed group. Given this lack of clarity, it is unclear how the incidence rate and the estimate of the relative risk could be calculated. The committee finds the results uninterpretable and, as such, noncontributory with respect to causality.

Other Issues. Verstraeten and colleagues (2003) discussed in their paper general limitations of analyzing VSD data. For example, they were unable to control completely for other potential confounding factors. In HMO B, the clinic that a child attended may have acted as a confounder. Also, the HMO databases did not provide information on other possible confounders, such as maternal smoking, lead exposure, or fish consumption. The authors noted, however, that it is unclear how these variables relate to the child's vaccination status and thereby confound the results.

The authors also cited several limitations when using automated databases, particularly the inability to separate the effects of thimerosal from other vaccine ingredients. The authors conducted a subanalysis by comparing risks associated with DTwP or DTaP vaccine and Hib vaccine given separately or combined. The thimerosal content of these vaccines differed in that the combined vaccine contained 25 μg of Hg and the two vaccines administered separately totaled 50 μg of Hg. There was no difference in estimates of speech or language delays (speech delay RR = 1.04 [95% CI, 0.85-1.27]; language delay RR = 1.29 [95% CI, 0.91-1.82]). There was a significant decreased risk for attention deficit disorder (ADD) (RR = 0.70 [95% CI, 0.52-0.95]).

In summary, the VSD study has several strengths and limitations. Strengths of the study include the following: vaccine information was available on the individual level, a subset of autism diagnoses were validated by medical record review, related diagnoses were examined to address possible misclassification, and a sensitivity analysis was conducted to determine the effect of exclusion criteria on study results. Limitations include the study's ability to answer whether thimerosal in vaccines causes autism because the study tests a dose-response gradient, not exposure versus non-exposure. Also, the small number of cases and instability of some of the risk estimates may affect the findings. Given these strengths and limitations, and with the effect estimates, although nonsignificant, near or below 1, the study showed no association between TCVs and autism.

Studies of Passive Reporting Data

United States. Geier and Geier conducted three similar studies (Geier and Geier, 2003a,b,d) to investigate the role of thimerosal-containing DTP vaccine with autism and other adverse outcomes. These studies compare the frequency over time of reported adverse events assumedly associated with administered doses of thimerosal-containing DTP vaccine versus the frequency of reported adverse events assumedly associated with administered doses of thimerosal-free DTP vaccine. They hypothesized that the DTP vaccine, whether containing thimerosal or not, should have a similar incidence rate of adverse events over time.

In each of these studies, data on adverse events were obtained from the descriptions in the VAERS database. The authors also analyzed several control outcomes, defined as “adverse events that were not biologically plausible [sic] linked with thimerosal” (Geier and Geier, 2004b), using data from VAERS. Information from the CDC BSS, which receives voluntary reports from all manufacturers of distributed and returned vaccine doses (Rosenthal et al., 1996), was also used as a proxy for the number of DTP vaccine doses administered in a year.

In the first study (Geier and Geier, 2003d), the authors examine two neurodevelopmental outcomes, autism and speech disorders. Information on the adverse event was obtained from VAERS. They estimated the number of administered doses of thimerosal-containing DTaP and DTwP vaccines from 1992 to 2000, and doses of thimerosal-free DTaP from 1997 to 2000 using data from the BSS. They estimated the number of each adverse event for two groups of individuals: those who assumedly received an average of 37.5 μg of mercury and those who assumedly received an average of 87.5 μg of mercury. They do not explain how they categorized individuals into the two exposure groups, nor is it clear how they could.

They then calculate incidence rate¹⁸ of the adverse events following vaccination. Relative risks were calculated by dividing the incidence rate of a particular adverse event following administration of TCV (DTaP or DTwP) by the incidence rate of a particular adverse event following thimerosal-free (DTaP) vaccine. Relative risks were calculated for the 37.5 μg and the 87.5 μg mercury exposure groups. The relative risks were then plotted against the amount of mercury each child assumedly received. The authors assumed a baseline mercury exposure of zero with a relative risk of 1. The relative risk estimates were graphically depicted in a plot, but were not reported. An R-square was calculated to estimate the association between each adverse event and mercury exposure. The R-squared calculation in the analysis comparing thimerosal-containing DTaP vaccine with thimerosal-free DTaP vaccine was 0.99 for autism. They also report a R-squared of 0.99 for autism in the analysis comparing thimerosal-containing DTwP with thimerosal-free DTaP. The regression models and other calculations were not specified in the paper. The authors concluded that “the analyses showed increasing relative risks for neurodevelop-ment disorders with increasing doses of mercury.” In an unpublished paper submitted to the committee (Geier and Geier, 2004b), they also state that this study found “consistent increasing risk dose-response relationships for autism following both of our thimerosal-containing vaccines in our comparison to our thimerosal-free DTaP vaccines.” In addition, they report that the control adverse events did not follow an increasing risk dose-response relationship with additional doses of mercury from TCVs.

In a second paper (Geier and Geier, 2003a), the authors conduct a very similar analysis using slightly different outcomes and time period for estimates of administered doses. In this study, they examine several adverse event outcomes from VAERS, including autism, personality disorder, and mental retardation. Control outcomes included febrile seizures, fevers, pain, edema, and vomiting. They estimated the number of administered doses of thimerosal-containing DTaP and DTwP and thimerosal-free DTaP from 1997 to 2001. Reports of adverse events were again categorized as either following an average exposure of 37.5 μg of mercury or an average exposure of 87.5 μg of mercury, but again no information was provided on how these averages and the exposure groups were derived.

The authors calculated ORs by comparing the incidence rate of a particular adverse event following DTP TCV with the incidence rate of a particular adverse event following DTP thimerosal-free vaccine. ORs were calculated for the 37.5 μg and the 87.5 μg mercury exposure groups. The ORs were then plotted against the assumed average amount of mercury received. The authors assumed a baseline exposure of zero with a relative risk of 1. A linear regression coefficient and slopes¹⁹ were reported for each adverse event. The regression model was not described in the paper.

They reported an overall OR for autism of 2.6 and that the OR of autism increased by 0.029 per μg of mercury. Standard errors were graphically depicted in a plot, but were not reported. They reported an R-squared of 0.98 for autism in the analysis comparing thimerosal-containing DTaP with thimerosal-free DTaP. The authors concluded that “the dose-response curves show that increases in odds ratios of neurodevelopmental disorders from VAERS … [were] closely linearly correlated with increasing doses of mercury from thimerosal-containing childhood vaccines and that for the overall odds ratios examined statistical significance was achieved” (p. 100).

In a third paper, Geier and Geier (2003c) conduct the same analysis using data again from the BSS to estimate the administered doses of thimerosal-containing DTaP vaccine from 1992 to 2000 and administered doses of thimerosal-free DTaP vaccines from 1997 to 2000. Outcomes examined were autism, mental retardation, and speech disorder. Control outcomes included deaths, seizures, vasculitis, emergency department visits, total reactive reports, and gastroenteroritis.

The authors state that 6,575 adverse events reports after thimerosal-containing DTaP vaccines and 1,516 adverse events reports after thimerosal-free DTaP vaccines were reported in the VAERS database. Relative risks were calculated by dividing the incidence rate of a particular adverse event following DTP thimerosal containing vaccine by the incidence rate of a particular adverse event following DTP thimerosal-free vaccine. Other risk estimates included attributable risk,²⁰which was calculated by subtracting one from the relative risk, and percent association,²¹ which was calculated by dividing the relative risk by the relative risk plus 1 and multiplying this sum by 100. Statistical significance was calculated using Fisher's exact test. A p-value of 0.05 was considered significant. The authors provided these estimates and the statistical significance in their paper (see Table 8), and concluded that “an association between neurodevelopmental disorders and thimerosal-containing DTaP vaccines was found.”

TABLE 8

Risk Estimates and Statistical Significance Testing in Geier and Geier 2003b.

These three studies have serious methodological limitations that make their results uninterpretable. A major problem with these studies is that they rely on analyses of VAERS data to draw conclusions about causality. VAERS is a passive reporting system that accepts voluntary reports from health care providers, vaccine recipients, manufacturers, and others regarding potential adverse events. Adverse event reports are not formal case reports, but rather are descriptions of symptoms that are temporally associated with receipt of a vaccine or vaccines. All adverse event reports are accepted into VAERS regardless of whether the vaccine plausibly caused the adverse event. VAERS has inherent limitations that include variability in reporting standards, reporting bias (e.g., due to other factors such as media attention), unconfirmed diagnoses, lack of information on people who were immunized but did not report an adverse event, lack of an unbiased comparison group, and variable and potentially significant underreporting of adverse events. While VAERS is useful in generating hypotheses, further validation and studies are required before conclusions on causality can be drawn (Varricchio, 2004).

In addition, the methods they use to analyze VAERS data, which they state follow methods developed by CDC and FDA (Rosenthal et al., 1996), in fact differ in important ways from methods used by the agencies. Rosenthal et al. (1996) evaluated the safety of acellular pertussis vaccine versus whole-cell pertussis vaccine using VAERS data. They calculated the number of doses distributed for each birth cohort using data from the National Health Interview Survey to obtain immunization coverage in children. These data provided information on age and dose-specific vaccination rates. In addition, the authors adjusted the administered doses for seasonal effects by using data from the VSD vaccination records. By contrast, the Geiers' papers do not make these adjustments, thus limiting the utility of the BSS data on doses administered. In addition, the Geiers' studies do not define a specific time period of interest; and they do not define a period of symptom onset (that is, it is unclear if autism or some other adverse event occurred after receipt of the vaccine) or the time period between vaccine receipt and autism diagnosis.

These studies also do not clearly delineate the unit of analysis (individual vs. group) nor do they explain how individuals are allotted into exposed and unexposed groups or how groups-level rates are constructed. They also make significant assumptions about individuals' vaccine histories, which cannot be accurately known from the VAERS or BSS data, introducing an unknown bias to their results.

The articles also lack a complete and transparent description of their methods and underlying data, making it difficult to confirm or evaluate their findings. That is, no information was provided on specification of regression models, the frequency distribution of variables was not provided, and calculations of statistics were not clear or were not reported. Furthermore, as noted when the measures appear in the text, the epidemiological measures “incidence rate,” “relative risk,” and “attributable risk” are applied incorrectly in these studies.

The results of their studies are likewise improbable. For instance, in one study (Geier and Geier, 2003a), the R² was 0.98, suggesting that the thimerosal content of DTP vaccines (excluding all other TCVs) explains 98 percent of the variation in autism rates. The R² is improbably high and means that all other potential factors (e.g., genetic or environmental factors) only explain 2 percent of the variance in autism rates. Furthermore, the R² explains variation at the individual level; the authors' calculations, however, are a line through averages. Using averages instead of individual points results in the disappearance of most of the variability in a data set (Nordin, 2004). As a result of these significant methodological limitations, the committee finds the results of their studies to be uninterpretable and, as such, they are noncontributory with respect to causality.

Unpublished Controlled Observational Study

United Kingdom. Miller (2004) presented to the committee a cohort study that examined whether certain NDDs, including autism, are related to exposure to TCVs. Regarding this study, forthcoming in the journal Pediatrics, the committee focused only on the analyses examining the autism outcome. The study was based on data from the U.K.'s General Practice Research Database (GPRD), which is similar to the U.S. VSD; About 500 general practices in the United Kingdom contribute data to the GPRD, which is about 5.7 percent of the population. The study's data cover 1988 to 1999 and include patient consultations, referrals, and vaccine information.

The only TCVs administered routinely to children in the United Kingdom are DTP and DT-containing vaccines. These vaccines contain 50 μg of thimerosal (25 μg of ethylmercury). Since 1990-1991, children have received doses of the vaccine at 2, 3, and 4 months of age (and at 3, 5, and 10 months of age before then). With the 2-3-4 month schedule, children could have received a maximum of 50 μg of mercury at 3 months of age and 75 μg of mercury at 4 and 6 months of age. This amount is less than the maximum amount received by U.S. children. U.S. children could have received 75 μg of mercury after 3 months, 125 μg after 4 months, and 187.5 μg after 6 months. Most children in the United Kingdom receive their vaccines in a timely manner.

Children were included into the study, based on the following criteria: born between 1988 and 1997 into a GPRD practice, enrolled in a GPRD practice for at least 2 years, exact birth data available, and data met “up to standard” criteria of GPRD. A total of 107,152 children were in this cohort, but children were further excluded if their records had data errors, for example, if the vaccination date was prior to the birth date. To reduce the possibility of confounding, individuals with the following conditions were excluded: various congenital, prenatal, perinatal conditions; postnatal conditions during the first 6 months; or other outcome events during the first 6 months of life. In addition, children who did not receive three doses of DTP by 1 year of age or who had incomplete vaccination records were excluded in this analysis. (For the main analysis, they were included.) The final cohort included 103,043 children. Of these children, 2,471 were preterm, and data were not presented on these children because none of them were diagnosed with autism. The final cohort included 100,572 term children.

Exposure to mercury was classified in three categories: those receiving doses of DTP or DT-containing vaccines by 3 months of age (93 days), with a maximum cumulative exposure of 50 μg; by 4 months of age (124 days), with a maximum cumulative exposure of 75 μg; and age-specific exposure by 6 months of age. The researchers examined autism and other neurodevelopmental outcomes that were also examined in the VSD study (Verstraeten et al., 2003). Autism diagnoses in the GPRD were converted from Read and OXMIS codes (coding used by GPRD) to ICD codes. Validation of autism diagnosis was not conducted, but other studies using GPRD data have demonstrated a high degree of reliability in the autism diagnosis (Black et al., 2002; Kaye et al., 2001). A Cox proportional hazards survival was the method of analysis. Survival began at 183 days of age to age of event, censoring occurred at age of event or last date of follow-up, and the results were reported as a hazard ratio per DTP or DT dose.

Overall, there were 104 reports of autism. Cases had a median age at first recorded diagnosis of autism of 4.4 years, at least 8 years of follow-up, and 89 percent of all cases occurred in males. The hazard ratio for autism after receiving doses by 3 months was 0.89 (95% CI, 0.65-1.21; p = 0.46); after receiving doses by 4 months was 0.94 (0.73-1.21; p = 0.66); and after receiving all doses by 6 months of age was 0.99 (0.88-1.12; p = 0.89). Estimates were adjusted for sex and year of birth. The results do not suggest that exposure to thimerosal at these doses increases the risk of autism. A reverse Kaplan-Meier plot was used to illustrate that the results did not suggest an increased risk of autism, and it was used to compare the risk of autism at zero, one, two, and three doses of receiving the DTP vaccine. Miller concluded that there was no evidence that increased exposure to thimerosal at a young age increases the risk of autism.

Unpublished Ecological Study

United States. An unpublished ecological analysis comparing aggregate trends in autism rates in California with the trends in mercury exposure from TCVs was presented to the committee (Blaxill, 2001). To estimate the average level of mercury exposure for children 19-35 months of age in a given year, the mercury content of all recommended vaccine doses (assumed to be 37.5 μg for three doses of hepatitis B, 75 μg for three doses of Hib, and 75-100 μg for three or four doses of DTP) was weighted by estimates from the National Health Interview Survey of coverage rates for the specific vaccines for the specified year. Estimates of the number of children with autism were obtained from annual caseload data from the California Department of Developmental Services database. The presenter concluded that the increasing trends seen in these data, both in the prevalence of autism and the levels of mercury exposure from TCVs, are consistent with the hypothesis that mercury exposure had a direct role in an increasing incidence of autism in the 1990s.

As noted in previous IOM reports (IOM, 1994a, 2001b), a positive ecological correlation constitutes only weak evidence of causality, and additional research would be needed to establish a causal association. The analytical value of the California data is limited by the inability to account for any changes over time in diagnostic concepts, case definitions, or age of diagnosis (Fombonne, 2001a). As a result, the committee cannot assess trends in the prevalence of autism from these data. The committee concludes that this unpublished ecological analysis is uninformative with respect to causality.

Several studies show that the prevalence of autism has increased in the past 20 to 30 years, but it is unclear whether the increase derives from changes in case definition, changes in case-finding methods, simply an increased awareness of autism, or other causes (Croen et al., 2002; Fombonne, 1999, 2002, 2003; Gillberg and Wing, 1999; Gurney et al., 2003; Wing and Potter, 2002). Incidence rates are more sensitive measures than prevalence rates for determining changes in the factors causing autism, but only a few incidence studies have been conducted, and the methods used in these studies limit their ability to test this hypothesis (Fombonne, 2003; Wing and Potter, 2002). Further study and clarification about the epidemiology of autism, especially studies examining autism incidence, would be helpful in elucidating why autism has been increasing in the past 30 years.

Causality Argument

Epidemiological studies examining TCVs and autism, including three controlled observational studies (Hviid et al., 2003; Miller, 2004; Verstraeten et al., 2003) and two uncontrolled observational studies (Madsen et al., 2003; Stehr-Green et al., 2003), consistently provided evidence of no association between TCVs and autism, despite the fact that these studies utilized different methods and examined different populations (in Sweden, Denmark, the United States, and the United Kingdom) (see Table 9). Other studies reported findings of an association. These include two ecological studies (Geier and Geier, 2003a, 2004a), three studies using passive reporting data (Geier and Geier, 2003a,b,d), one unpublished study using VSD data (Geier and Geier, unpublished), and one unpublished uncontrolled study (Blaxill, 2001). However, the studies by Geier and Geier cited above have serious methodological flaws and their analytic methods were nontransparent making their results uninterpretable, and therefore noncontributory with respect to causality (see preceding text for discussion). The study by Blaxill is uninformative with respect to causality because of its methodological limitations. Thus, based on this body of evidence, the committee concludes that the evidence favors rejection of a causal relationship between thimerosal-containing vaccines and autism.

TABLE 9

Evidence Table: Exposure to Thimerosal-containing Vaccines and Autism.

This conclusion differs from the committee's finding in its 2001 report on TCVs and NDDs which was that the evidence was “inadequate to accept or reject a causal relationship between exposure to thimerosal from childhood vaccines and the neurodevelopmental disorders of autism, ADHD, and speech and language delay.” (IOM, 2001b. 66) The committee's conclusion in 2001 was based on the fact that there were no published epidemiological studies examining the potential association between TCVs and NDDs, and the two unpublished, epidemiological studies that were available (Blaxill, 2001; Verstraeten, 2001) provided only weak and inconclusive evidence of an association between TCVs and NDDs. Furthermore, the conclusion in the 2001 report pertained to a broader set of NDDs, while this report's conclusion applies only to autism.

MMR and Autism

In 1998, Wakefield and colleagues published a case series describing 12 children with pervasive developmental disorder characterized by GI symptoms and developmental regression (Wakefield et al., 1998). In eight of the cases, parents or physicians reported the onset of these symptoms within 2 weeks of the child's MMR vaccination. This study put forth a hypothesis that a new variant of autism characterized by GI symptoms and developmental regression could be associated with the MMR vaccine. Ten of the original 13 authors have since formally retracted that interpretation: “We wish to make it clear that in this paper no causal link was established between MMR vaccine and autism, as the data were insufficient. However, the possibility of such a link was raised and consequent events have had major implications for public health. In view of this, we consider now is the appropriate time that we should together formally retract the interpretation placed upon these findings in the paper, according to precedent” (Murch et al., 2004).

In response to public concern about the safety of the MMR vaccine following the 1998 study, researchers initiated a number of epidemiological studies to examine the association between the vaccine and ASD. In addition to case reports from VAERS and other sources, the committee identified 16 epidemiological studies related to MMR and autism that were published since the original Wakefield paper. Ten of these studies (DeStefano et al., 2004; DeWilde et al., 2001; Farrington et al., 2001; Fombonne and Chakrabarti, 2001; Geier and Geier, 2003c, 2004a; Madsen et al., 2002; Makela et al., 2002; Takahashi et al., 2003; Taylor et al., 2002) were published since the committee's last review in 2001. These studies addressed four main questions: (1) Are rates of ASD higher among children who are vaccinated with MMR than in those who are not? (2) Is there an increase in ASD as a result of the MMR vaccine? (3) Is there a new variant form of autism associated with the MMR vaccine? and (4) Is the development of ASD temporally associated with receipt of the MMR vaccine? (Wilson et al., 2003).

The studies were conducted in five different countries (the United States, the United Kingdom, Finland, Denmark, and Sweden), using 11 distinct data sources (one data source was used three times and another was used twice). The studies employed a variety of study designs, including time-series analysis, cross-sectional analysis, ecologic analysis, case control, and retrospective cohort. They also differed in how ASD was diagnosed, though most of them relied on service records of children with autism or children with disabilities or on ICD codes (Wilson et al., 2003). All the epidemiological studies of MMR and autism, including those reviewed in the previous IOM report (IOM, 2001a), are described in detail below and are summarized in Table 10.

TABLE 10

Evidence Table: Exposure to MMR Vaccine and Autism.

Controlled Observational Studies

United States. DeStefano and colleagues (2004) conducted a case-control study in the metropolitan Atlanta area comparing ages at first MMR vaccination of a population-based sample of children with autism and school-matched controls who did not have autism. The authors stated that implicit in the exposure comparison was the assumption that if MMR vaccination increases the risk of autism (typical onset is 24 months), children who receive the vaccine at earlier ages would have a higher risk of developing autism. The researchers chose this study design because they did not have full information on onset of initial parental concern, date of first diagnosis of autism, or onset of regression.

Children with autism who were enrolled in the study were identified from the Metropolitan Atlanta Developmental Disabilities Surveillance Program (MADDSP), a population-based surveillance program that monitors the occurrence of selected developmental disabilities among children in the Atlanta region. Autism cases were identified through screening and abstraction of files at multiple sources, including schools, hospitals, clinics, and specialty providers. In 1996, the first year that autism was included as part of the MADDSP, 987 children 3-10 years old with autism were identified. School immunization records were located for 660 of those children; the rest either moved out of the state or transferred to a school outside the Atlanta public school system. Three control children were matched for 97 percent of the children with autism. Controls were selected from the regular education programs and were matched on the basis of age, gender, and school. Data on immunization history was abstracted from immunization forms required for school entry. Case and control children with missing or incomplete vaccination forms²² were excluded. After all exclusions, 624 cases and 1,824 controls remained in the study.

A developmental pediatrician reviewed the MADDSP data files and classified children with autism into three subgroups with potentially different environmental susceptibilities to development of autism. These non-mutually exclusive groups included (1) children with no indication of developmental delay before 1 year of age (i.e., before recommended age of the first MMR vaccine) or a preexisting condition (e.g., major birth defect, co-occurring developmental disability, or major perinatal or postnatal insult that could have contributed to developmental delays); (2) children with loss of age-appropriate developmental skills (regression) or appropriate skills that failed to progress (plateau); (3) children with and without coexisting mental retardation (defined as an IQ of less than 70).

In addition to vaccination records for all children, researchers obtained additional developmental disability-related information from MADDSP, including birth defects, epilepsy, IQ level, and presence of other developmental disabilities, for children with autism. Demographic data were collected for all case and control children. Approximately 56 percent of cases and controls were matched with Georgia state birth records, which provided additional information about birth and maternal factors (e.g., birth weight and gestational age, and mother's parity, age, race, and education).

The overall distribution of ages at MMR vaccination was similar to that of matched control children; 70.5 percent of cases and 67.5 percent of control children were vaccinated between 12 and 17 months of age. The authors analyzed associations between MMR vaccine and autism using three age cutoffs: (1) < 18 months, as an indicator of on-time vaccination; (2) < 24 months of age, the age at which developmental problems have become apparent in most children with autism; and (3) < 36 months of age, the age by which characteristics of autism must have developed in order to meet DSM-IV criteria. Similar proportions of children received the MMR vaccine by the recommended age of 18 months and before 24 months, the age at which developmental delays are normally recognized in children with autism.

Conditional logistic regression models were used to estimate adjusted ORs. The rates of vaccination prior to 36 months were higher among cases (93.4 percent) than among controls (90.6 percent) (OR = 1.49; 95% CI 1.04-2.14 in the total sample; OR = 1.23; 95% CI 0.64-2.36 in the birth certificate sample).²³ This association was strongest in the age group of 3- to 5-year-olds. The authors suggested that this likely reflected immunization requirements for early intervention programs.

The authors identified several strengths of the study. It was a large population-based study with a well-defined case and control population. In addition, a panel of autism experts reviewed case records to confirm that cases met the DSM-IV criteria. Immunization records were obtained from standardized school records, thus eliminating possibility of recall bias. Data on clinical and behavioral factors and vaccination data were obtained from independent sources, thereby reducing potential information bias. The study also controlled for confounding demographic and birth factors.

One limitation of the study was the lack of vaccination records for one-third of the 987 case children, owing to transfers to schools outside the Atlanta public school district. In addition, a subanalysis of children with Georgia birth certificates was conducted to evaluate potential confounders in terms of birth and maternal characteristics. The birth certificate sample did have lower ORs, which could represent random fluctuation or potential bias related to being born outside of Georgia.

United Kingdom. In a population-based, self-matched case series study (or case-crossover study) of children in eight health districts of the North East Thames region in the United Kingdom, Taylor and colleagues (1999) investigated trends in the incidence of ASD before and after the introduction of MMR vaccine (in 1988) and tested for postvaccination clustering of diagnoses or other indicators of onset. Children with ASD born since 1979 were identified from computerized specialized-needs/disability registries and from records in special schools. Information was extracted on the age at which the disorder was diagnosed, the recorded age at which parents first became concerned about the child's developmental course, and the age at which regression became obvious (if it occurred). The authors identified 498 children with ASD: 261 with typical (core) autism, 166 with atypical autism, and 71 with Asperger's syndrome. When ICD-10 criteria were applied to available records, these diagnoses were confirmed for 82 percent of core autism cases, 31 percent of atypical autism cases, and 38 percent of Asperger's syndrome cases. While not all cases were verified according to ICD-10 criteria, most were documented as having been assessed by a specialist clinician. All 498 cases were included in the analyses. Data on the vaccination histories of these children were obtained from a separate regional information system.

Three statistical analyses were undertaken. First, in a time-series analysis, Poisson regression was used to fit an exponential trend to the number of cases diagnosed by age 60 months for children born between 1979 and 1992. Children born later were excluded because a diagnosis might not have been made by the time of the study. The trend showed a significant increase in cases for core and atypical autism but produced no evidence of a sudden “step up” after MMR vaccine was introduced in the United Kingdom.

The second analysis focused on children born after 1987. Among the 389 study subjects in this group, the proportion who had received MMR vaccine by their second birthday (86.4 percent) was reported to be similar to that of the North East Thames region in general. A further analysis was limited to 356 children who were diagnosed with core or atypical autism at age 18 months or later. Of these cases, 233 received the MMR vaccine before this age, 64 never received it, and 59 received the vaccine at 18 months or later. Children with Asperger's syndrome were excluded because of the small numbers involved and their older age at diagnosis. There were no statistically significant differences in mean age at diagnosis between those vaccinated before 18 months of age, those vaccinated afterward, and those never vaccinated. There was no temporal association between changes in the incidence of autism by birth cohort since 1987 and changes in vaccine coverage.

The third analysis focused on the timing of diagnosis, first parental concern, and regression in children who received one or more MMR vaccine doses. With one exception, the authors found no clustering of diagnosis, parental concern, or autistic regression in the periods following vaccination. (The authors investigated periods within 12 or 24 months after vaccination for diagnosis; periods within 6 or 12 months after vaccination for parental concern; and periods within 2, 4, or 6 months following vaccination for regression.) A statistically significant clustering of parental concern was found within 6 months of vaccination (relative incidence 1.48; 95% CI, 1.04-2.12), and this was attributed to a large peak in recorded parental concern at 18 months and a peak in MMR vaccination at 13 months. Because the convergence of parental reports around age 18 months may reflect recall uncertainty and age 13 months corresponds to the recommended vaccination schedule, this association was interpreted by the authors as an artifact related to the difficulty of precisely defining the onset of symptoms. The authors concluded that their analyses did not support a causal association between MMR vaccination and autism or the onset of autistic regression.

In an extended analysis using the same data as Taylor and colleagues (1999), Farrington and colleagues (2001) assessed the temporal association between MMR vaccine and ASD. This analysis used a self-matched case-series method employing the same age groups as the previous study. The Farrington study found no increase in the incidence of diagnosis of autism, regressive autism, or parental concern at 24, 36, or 60 months following vaccination. There was no increased likelihood of ASD, regression, or parental concern when comparing pre- and postvaccination groups.

United Kingdom. Taylor and colleagues (2002) updated their 1999 analysis in five of the eight health districts studied in North East London to examine the association between the MMR vaccine and bowel problems and developmental regression in children. Using computerized health registers of children with disabilities, information from special schools, and psychiatric records, the investigators identified children with autism who were born between 1979 and 1998. Information abstracted from clinical records was then linked to independent vaccination records. The researchers also recorded information on bowel problems (when they exceeded 3 months in duration), onset of parental concern about the child's development, and regression (if there was documented decline in the child's development or parents reported loss of skills).

A total of 473 children with autism (278 with childhood autism and 195 with atypical autism) were enrolled in the study. Of the 473 children, 81 (17 percent) were reported as having bowel problems. Regression was reported by parents in 118 (25 percent) of the 469 children who had developmental records. The proportion with bowel problems and regression was similar among the children with childhood autism and those with atypical autism.

The researchers conducted single and multivariate logistic regression analyses to examine the relationship between exposure to the MMR vaccine and the onset of autism and the presence of bowel problems and regression. Adjustments were made for several potential confounders (sex, year of birth, district, age of parental concern, type of autism), but the adjustments did not significantly affect the results. No significant trends were found by year of birth in the proportion of children with autism with bowel symptoms (OR = 0.98; 95% CI, 0.93-1.04; p = 0.47) or regression (OR = 0.98; 95% CI, 0.93-1.03; p = 0.50) during the 20-year period from 1979 to 1998.

There was no significant difference in the rates of bowel symptoms or regression between three groups: children with autism who received the MMR vaccination prior to onset of concerns about development (where MMR might have caused the autism); those who received MMR vaccination after the onset of parental concerns; and to those who did not receive the MMR. These results found neither evidence of an association of these two symptoms with the MMR vaccine (p = 0.57) nor a change in the proportion with these symptoms by year of birth (OR = 0.98; 95% CI, 0.89-1.07; p = 0.58).

The authors did find an association between bowel problems and developmental regression in the study population. Thirty-one of the 118 children [26 percent] with regression and 49 of the 351 children [14 percent] (p = 0.002) without regressive autism reported bowel symptoms. Single and multivariable logistic regression models, however, showed no association with these factors and MMR vaccine. The authors suggested that the occurrence of the bowel problems and regression together might reflect particular dietary problems, but that the findings do not support the hypothesis of a MMR-induced new variant form of autism with regression and bowel symptoms.

United Kingdom. Fombonne and Chakrabarti (2001) conducted a cross-sectional study to examine whether a new phenotype of autism, characterized by regression and bowel symptoms, is associated with MMR. The authors stated that if this new phenotype of autism exists, then at least one of the following six predictions should be supported by empirical data: (1) childhood disintegrative disorder (CDD)²⁴ has become more prevalent; (2) the average age at first parental concern for children with autism who received MMR is closer to the average immunization age than in children who did not receive MMR; (3) developmental regression in children with autism has become more common in MMR-vaccinated children; (4) the age of onset for regressive autism clusters around the MMR immunization date and is different from that of children with autism without regression; (5) children with regressive autism have different symptom and severity profiles than those without regressive autism; or (6) regressive autism is associated with GI symptoms and/or inflammatory bowel disorder.

This study used three samples to examine these questions. The primary sample for this study, the Stafford sample, was drawn from a population-based epidemiologic survey of pervasive developmental disorder (synonymous with ASD) in children born between 1992 and 1995 in Staffordshire, United Kingdom. In this population, 96 children were confirmed to have ASD,²⁵ with 99 percent having received their first MMR vaccine (median age of 13.5 months) and 65.6 percent having received their second MMR vaccine (median age 43.6 months). The investigators compared that sample with two other clinical groups, the Maudsley Hospital Clinic (MHC) and the Maudsley Family Study (MFS) samples. The MHC sample consisted of 68 children with autism born between 1987 and 1996 who had been exposed to the MMR vaccine and had been seen in a clinical setting. The MFS sample consisted of 98 individuals with autism born between 1954 and 1979 seen in a clinical setting, none of whom had been exposed to MMR. All patients were assessed using the standardized ADI, which provided reliable data (e.g., on age at first parental concern, rates of regression) across samples. Data on bowel symptoms and disorders were obtained from the epidemiological (Stafford) survey and immunization records were obtained from computerized records.

The investigators tested all six hypotheses noted above. First, only one child in the epidemiological (Stafford) sample had CDD, resulting in a very low prevalence of CDD (0.6/10,000 [95% CI, 0.02-3.6/10,000]). This is consistent with previous studies and does not suggest an increase in this form of autism in samples of children vaccinated with MMR. In a second analysis, there was no difference across three samples (two MMR-exposed [Stafford and MHC] samples and the one unexposed [MFS] sample) in the mean age at which parents first became concerned about autistic symptoms in their child (F_2,250 = 0.02; not significant). In a third analysis, investigators compared the rates of regression in the two samples for which information on regression was available to see whether the rate of regression increased. The results showed no difference in the rates of regression in the epidemiological sample of 96 children (Stafford sample) who had been exposed to MMR as compared to unexposed comparison group (X² = 3.65; df = 2; p > .15). Fourth, age at parental recognition of first symptoms among children with autism with developmental regression (19.8 months) was similar to that of children with autism without developmental regression (19.3 months). Mean intervals between MMR immunization and parental recognition of autistic symptoms were similar among children with and without regression (248 days versus 272 days; not significant). Finally, in the epidemiologic (Stafford) sample, where data on GI symptoms were available, such symptoms were reported in 18.8 percent of the sample. There was no association between regression and GI symptoms in this study population (OR: 0.63; 95% CI, 0.06-3.2; not significant). The investigators concluded that their study does not support a new phenotype of MMR-induced ASD with features of regression and bowel symptoms.

This study was limited by the use of samples from two different time periods and was thus susceptible to bias caused by secular changes in the diagnostic methods and referral patterns for autism (Wilson et al., 2003). However, these limitations should not affect the inferences regarding regression and bowel symptoms.

United Kingdom. DeWilde and colleagues (2001) conducted a case-control study of children in the United Kingdom to examine whether a temporal association between MMR vaccine and autism existed. The authors hypothesized that onset of developmental regression following MMR vaccine would be reflected in the increased consultations with the child's general practitioner. The Doctor's Independent Network, a general practice database, was used to examine whether children who were subsequently diagnosed with autism had more frequent consultations following MMR vaccine than children who were not vaccinated. Seventy-one children with ASD and 284 matched controls (4:1) were identified during the time period 1989 to 2000. Controls were matched for age, sex, month of MMR vaccination, and general practitioner practice. The authors found no significant difference in consultational behavior between the patients with autism and controls. Furthermore, only one case of ASD was diagnosed within 6 months of MMR vaccination (median time was 1,053 days). A significant limitation of this study is that it included any consultation rather than only consultations specific to autism (Wilson, 2004).

Denmark. Madsen and colleagues (2002) conducted a population-based retrospective cohort study comparing the rates of ASD among children in Denmark who were vaccinated with MMR with those who did not receive the MMR vaccine. Study subjects included all children in Denmark born between 1991 and 1998. The cohort was established using data from the Danish Civil Registration system, which assigns a unique identification number to each live birth. This unique identification number is used to link an individual's record with all other national registries.

MMR vaccination status was obtained from data reported to the Danish National Board of Health (NBH) by general practitioners, who administer all MMR vaccinations in Denmark and who are reimbursed for these vaccinations upon reporting them to the NBH. The MMR vaccine was introduced in Denmark in 1987 and was the same vaccine that was used in the United States at the time.²⁶ Single antigen measles virus has not been used in Denmark. The Danish national vaccine program recommends vaccinating children with MMR at 15 months and 12 years of age and provides the vaccine free of charge. No changes to the MMR vaccination schedule were made during the study period. The authors obtained information only on the exposure at 15 months because that was the only one relevant to the outcome measurement in this study.

Information on diagnoses of autism was obtained from the Danish Psychiatric Central Register, which contains information on diagnoses received by patients in inpatient psychiatric hospitals and departments and in all outpatient clinics. Over 93 percent of children were treated on an outpatient basis. Diagnoses were based on ICD-10 codes and included all children with autism (ICD-10 code F84.0 and DSM-IV code 299.00) or another ASD (ICD-10 code F84.1-F84.9). Experts in child psychiatry and autism conducted a record review of 40 patients to validate the autism diagnoses. Thirty-seven (92 percent) of these children met the operational criteria for autistic disorder. The three children who did not meet the operational criteria for autistic disorder were classified as having another ASD. Data on birthweight and gestational age and other confounders were also obtained from other national registries.

Children were followed from 1 year of age until a diagnosis of autism, other ASDs, or associated genetic condition,²⁷ emigration, death, or end of the follow-up period (December 31, 1999), whichever came first.²⁸ Children were assigned to the unvaccinated group until they received the MMR vaccine. Incidence-rate ratios (referred to as relative risks) for autism and other ASDs in the MMR-vaccinated group compared with the unvaccinated group were analyzed using a log-linear Poisson regression model. Risk estimates were adjusted for potential confounders, including age, sex, calendar period, socioeconomic status, mother's educational level, gestational age, and birth weight.

A total of 537,303 children were included in the cohort and followed for 2,129,864 person-years. A total of 440,655 children (82 percent) received the MMR vaccine for a total of 1,647,504 person-years of follow-up, compared with 482,360 person-years of follow-up for children who did not receive the MMR vaccine. Of these, 316 children were identified with an autistic disorder and 422 with a diagnosis of another ASD. Mean age at time of MMR vaccination was 17 months, and 98.5 percent of vaccinated children received the MMR prior to 3 years of age. The same proportion of boys (82 percent) were vaccinated as were girls.

After adjusting for potential confounders, the relative risk of autistic disorder among vaccinated children compared with unvaccinated children was 0.92 (95% CI, 0.68-1.24) and the relative risk for other ASD was 0.83 (95% CI, 0.65-1.07). There was no association between age at time of MMR vaccination, time since vaccination, and date of vaccination and the developmenta of autistic disorder.

The results showed that the risk of autism was comparable both in vaccinated and unvaccinated populations. In addition, there was no temporal clustering of autism cases following immunization. Finally, MMR vaccine was not associated with autism or other ASD. The authors concluded that the study provided evidence against a causal association between vaccines and autism.

Finland. Makela and colleagues (2002) conducted a retrospective cohort study to investigate the association between MMR vaccine and several neurological disorders, including autism, among children in Finland.²⁹ The study examined two hypotheses: (1) whether there was a clustering of cases of hospitalization for autism after receipt of the MMR vaccine; and (2) whether any recipients of the MMR vaccine who were diagnosed for autism were also hospitalized with inflammatory bowel disease.

In Finland, MMR vaccination at ages 14 to 18 months and 6 years began in 1982. Researchers obtained vaccination records from a national vaccine surveillance study conducted between 1982 and 1986. The register contained records for approximately 86 percent of all children scheduled to be vaccinated between 1982 and 1986. Of the enrolled vaccinees, 535,544 children were 1 to 7 years old at time of the vaccination. Vaccination data on these children was linked with data from the national hospital register. Three hundred fifty-two of them were hospitalized with an autism diagnosis (defined by ICD 8 or 9 codes). Of these, 309 were hospitalized for autism following their MMR vaccination.

Because of the lack of a specific risk period for autism following vaccination, the authors assessed whether there were changes in the overall number of hospitalizations for autism after MMR vaccination for the study as a whole. They found no clustering of hospitalizations for autism following vaccination. Intervals between MMR vaccination and hospitalizations for autism ranged from 3 days to twelve and a half years. Hospitalizations for autism remained relatively stable for the first 3 years and then increased as the child aged, as expected. In addition, of the children hospitalized with autism, none were hospitalized for inflammatory bowel diseases from 1982 to 1995. The authors concluded that the results showed no association between MMR vaccination and autism.

The primary limitation of this study was that it relied only on hospitalization records and thus was unable to identify individuals who were not hospitalized but rather seen on an outpatient basis. While the authors stated that it is common in Finland for children with autism to be admitted to the hospital for observation and testing, a diagnosis of autism does not always involve hospitalization.

Japan. Takahashi and colleagues (2003) conducted a case-control study of children with autism in the Tokyo area growing up between 1988 and 1992 to assess the association of autism with MMR vaccine compared to monovalent measles, mumps, and rubella immunization. From 1989 to 1993, the Japanese national immunization program recommended that MMR vaccine or monovalent measles vaccine be given at age 12 months. Monovalent mumps and monovalent rubella vaccines were optional for children not receiving MMR.

Twenty-one children (4 female, 17 male) with autism were enrolled as cases in the study. Diagnosis of autism was made based on DSM-IV criteria. Physicians from a specialized psychiatric facility in the Tokyo area requested that caregivers of ASD patients born between 1988 and 2000 give consent to release records (response rate with informed consent was 87.5 percent). A control group of 8 females and 34 males (twice the size of the case group) was selected from the same birth year cohort of children who were attending one of the two pediatric clinics in Tokyo. Controls were randomly selected; only those who did not have neuropsychiatric disorders and whose guardian's had consented to disclosure of their records were enrolled in the study (response rate 58 percent). Immunization history data, including information on adverse events, were collected from records contained in the Maternal and Child Health handbook. Completeness of immunization in the case group versus the control group was 90.5 percent versus 100 percent for measles, 42.9 percent versus 78.6 percent for mumps (p < 0.01), and 52.3 percent versus 83.3 percent for rubella (p < 0.01). Only two cases and four controls received the monovalent measles, mumps, and rubella shots.

The authors used a 1:2 sex-adjusted logistic regression to analyze data collected from the records. Using MMR immunization as a reference, the ORs for the development of ASD were significantly increased for monovalent measles (OR = 5.33; 99% CI, 1.03-27.74), nonmumps immunization (OR = 8; 99% CI, 1.33-48.2), and nonrubella immunization (OR = 8.57; 99% CI, 1.30-56.4). The odds for development of ASD after receipt of monovalent vaccines (mumps and rubella vaccines), compared to the odds of ASD after receipt of the MMR vaccine, were not statistically significant (monovalent mumps: OR = 3.33; 99% CI, 0.45-24.6; monovalent rubella: OR = 3.82; 99% CI, 0.59-24.7). The results suggest a decreased risk of developing autism following MMR compared to monovalent measles, mumps, and rubella vaccines.

The study was limited by the small sample size (N = 21), the potential bias resulting from the informed-consent process,³⁰ and the poor response rate from the control group, all of which may have exaggerated the findings of no association.

Ecological Studies

United States. Dales and Colleagues (2001) examined trends in autism and MMR immunization coverage among young children in California to determine whether a correlation exists. Data on age at first MMR immunization for children born between 1980 and 1994 who were enrolled in California kindergartens was derived from annual reviews of a sample of school records (approximately 600-1,900 children per year). The California Department of Developmental Services provided data on regional service-center caseloads for children born between 1980 and 1994 and having an ICD-9 diagnosis of autistic disorder, which excludes other pervasive developmental disorders.

The authors observed a substantial increase in autism caseloads for successive birth cohorts but relatively stable immunization rates at ages 17 or 24 months. They concluded that these data did not support an association between MMR immunization and an increase in the incidence of autism.

The authors note that data from the California Department of Developmental Studies were not designed to measure trends in autism incidence because other system factors can affect caseloads from year to year. Several methodological limitations in these data have been cited, including the failure to account for changes over time in the population size or composition, in diagnostic concepts, in case definitions, or in age of diagnosis (Fombonne, 2001c). Thus, trends in autism caseload must be interpreted with caution. The authors noted that they were unable to link individual immunization and autism records for the same children. In addition, the data did not provide precise breakdowns of the percentage of children who received the MMR vaccine versus separate administration of monovalent or other combinations of measles, mumps, and rubella vaccines. Historical information suggests though that separate administration was rare in the United States during the period of study.

United States. Geier and Geier (2004a) examined the hypothesized association between MMR vaccine and autism. The number of children with autism was based on data from the DOE regarding children with autism who were enrolled in special education programs. Autism cases were based on the DOE's definition of autism and data collection standards.^31,32 The authors estimated the number of children who developed autism for the birth cohorts in each of the following years: 1982, 1985, and 1991-1996. They estimated the prevalence of autism in each birth cohort by dividing the number of children with autism in each year by the number of live births that year, using data from the CDC's yearly live birth surveillance data. The number of measles-containing vaccines administered in each respective birth cohort was based on CDC estimates of primary pediatric measles-containing vaccine coverage for birth cohorts in 1982 (67 percent coverage), 1984 (61 percent), 1991 (82 percent), 1992 (82.5 percent), 1993 (84.1 percent), 1994 (90 percent), 1995 (87.8 percent), and 1996 (90.5 percent). The authors plotted the estimated number of primary pediatric measles-containing vaccinations versus the prevalence of autism for each birth cohort examined. The authors reported a slope of 4831³³ and a linear regression coefficient of 0.91.³⁴ The authors concluded that a potential positive correlation exists between primary pediatric vaccines and autism prevalence.

This study has serious methodological limitations, similar to the other studies by these authors. First, the data on autism cases comes from the DOE dataset, which has significant limitations. As noted previously, these data are based on children with a diagnosis of autism who receive special education services, which may or may not reflect the true prevalence of autism. Furthermore, the child count is a cross-sectional (i.e., point-in-time) count of students served and the appropriateness of the authors' transformation of the cross-sectional autism-caseload data into birth-cohort prevalence data is questionable (Fombonne, 2001a). Changes in either states' policies or IDEA's reporting practices can affect the caseload numbers from year to year. For instance, autism caseload may vary from year to year as the result of policy changes, such as changes to the states' eligibility criteria for particular disabilities. Similarly, prior to the addition of the autism category in 1990, IDEA served students with autism, but for reporting purposes they were reported in a different disability category (see http://www.ideadata.org/docs/bfactsheetcc.doc).

Furthermore, the exposure is not specific to MMR; that is, the data do not provide information on the percentage of children who received the MMR vaccine versus separate administration of monovalent or other combinations of measles, mumps, and rubella vaccines. Finally, the regression model was not described in the paper. The description of analytical methods was not transparent and important details were omitted. As a result, the results of this study are uninterpretable and therefore noncontributory to causality.

United Kingdom. Kaye and colleagues (2001) used population-based data from the U.K. GPRD to conduct a time-trend analysis for estimating changes in the risk of autism and specifically for assessing the temporal relationship between MMR vaccination in the United Kingdom and the incidence of autism. The authors noted that the GPRD has been used for numerous published studies and is considered to be complete with respect to vaccination records.

From GPRD records, 305 cases of autism in children aged 12 or younger and diagnosed between 1988 and 1999 were identified. Of these cases, 83 percent were male and 81 percent were referred to a specialist for evaluation of the diagnosis. The estimated annual incidence of diagnosed autism had increased sevenfold from 0.3 per 10,000 person-years in 1988 to 2.1 per 10,000 person-years in 1999, with a median age at first recorded diagnosis of 4.6 years. The authors performed further analyses to estimate the 4-year risk of diagnosed autism for each annual birth cohort. These analyses were restricted to 114 boys born between 1988 and 1993 who were first diagnosed with autism between 2 and 5 years of age. The prevalence of MMR vaccination was calculated separately for each annual birth cohort (restricted to children who were registered with the GPRD within 60 days of birth and had at least 2 years of recorded follow-up).

The 4-year risk of autism increased nearly fourfold from 8 per 10,000 person-years in 1988 to 29 per 10,000 person-years in 1993, while the prevalence of MMR vaccination remained constant, at 97 percent. The authors hypothesized that if MMR vaccine were a major cause of the increasing incidence of autism, then the risk of autism in successive birth cohorts would be expected to stop rising within a few years of the vaccine being widely used. However, because the incidence of autism among 2- to 5-year-olds increased markedly from 1988 to 1993 while MMR vaccine coverage remained over 95 percent for successive birth cohorts, the authors concluded that the results did not support a causal association between MMR vaccination and the risk of autism.

The authors noted they did not review full clinical records for the children diagnosed with autism; such a review would be necessary to provide a more detailed characterization of these children and their diagnosis and to explore other possible explanations for the increase in the observed incidence of autism during the past decade.

Sweden. In a brief commentary, Gillberg and Heijbel (1998) reanalyzed data from a population study of autism conducted in the late 1980s in Sweden (Gillberg et al., 1991). A total of 55 children were diagnosed with autistic disorder based on DSM-III-R criteria, and an additional 19 individuals met criteria for atypical autism based on ICD-10 criteria. The MMR vaccine was introduced into Sweden for 18-month-old children in 1982, and coverage soon rose to above 90 percent. The authors divided the subjects into two groups according to era of birth, as a proxy for exposure to the MMR vaccine: children born between January 1, 1975, and June 30, 1980 (pre-MMR), and children born between July 1, 1980, and December 31, 1984 (post-MMR). The authors hypothesized that if autism were associated with MMR vaccination, children born since July 1980, who were 18 months or younger at the time of MMR introduction, would be at increased risk of having developed autism.

The analysis indicated that 47 of the children (34 with autistic disorder, 13 with atypical autism) were born during the earlier period, and 27 (21 with autistic disorder, 6 with atypical autism) were born during the later period. Because the numbers in the later period were much lower than expected had there been a strong effect at the population level of MMR on the prevalence of autism, the authors concluded that this study did not support the hypothesized association between MMR vaccine and autism. One limitation of this study was that children born from 1980 through 1984 did not have the same length of follow-up as those born earlier, with the maximum follow-up age of 4 for those born in 1984. Although most cases of autism are diagnosed prior to this age, if there was some age dependency in the MMR effect, this study would have been unable to detect it.

This and the other ecological studies are limited in that secular trends in the methods for diagnosing autism may confound any finding of an association or lack of an association. In addition, the ecological study design precludes drawing inferences about individuals based on population data (Wilson et al., 2003).

Studies of Passive Reporting Data

United States. Geier and Geier (2003c) examined reports of serious neurologic symptoms to VAERS following primary pediatric MMR immunization in comparison to reports of these outcomes following DTwP vaccination, which they designated as a control group. The serious neurologic symptoms examined in this study included cerebellar ataxia, autism, mental retardation, and permanent brain damage. The authors used data on vaccine doses distributed per year from the CDC's BSS as a proxy for vaccines administered each year. They used CDC estimates of percent vaccine coverage of each yearly birth cohort from 1994 to 2000. The authors calculated incidence rates³⁵ for the adverse events, which were then used to calculate the relative risk, attributable risk,³⁶ and percent association.³⁷ For autism, they reported a relative risk of 5.2, an attributable risk of 4.2, and a percent association of 84; the statistical significance was p < 0.001, and the 95% CI was 3.0-9.2. The authors report that the serious neurological outcomes were “statistically significantly increased following primary MMR vaccination in comparison to DTwP vaccination” (p. 205).

This study has many of the same methodological limitations as the authors' papers examining thimerosal and autism including the following: using VAERS data to infer causality, failure to clearly delineate the unit of analysis (individual vs. group), inappropriate use of epidemiologic measures (incidence rate and attributable risk), and nontransparent analyses. These problems are discussed in detail in the thimerosal causality section. The study's significant methodological limitations make it results uninterpretable, and therefore noncontributory with respect to causality.

Finland. In 1982 the Finnish National Board of Health and National Public Health Institute launched a long-term MMR vaccination program aimed at the elimination of measles, mumps, and rubella from Finland (Patja et al., 2000; Peltola et al., 1998). All children were to be vaccinated twice with MMR, between the ages of 14 and 18 months and at 6 years. In addition to the primary target groups, intermediate age groups were vaccinated in catch-up programs, unvaccinated adolescents were vaccinated during outbreaks, and adult groups at increased risk of exposure to these diseases (e.g., defense workers, health care workers) were also vaccinated. The live-virus MMR vaccine produced by Merck & Co., Inc. (West Point, PA) was primarily used, except in 1992-1996, when 2,570 doses of Trivirten (Swiss Serum and Vaccine Institute, Bern) were administered to individuals with severe hypersensitivity. Vaccine coverage was around 95 percent, with almost 3 million doses distributed and approximately 1.8 million vaccinations by 1996.

Following introduction of MMR, a countrywide passive surveillance system, based on reporting by health care personnel, was established to gather information about the incidence and nature of all severe adverse events following MMR vaccination. A potentially serious adverse event was one that met at least one of three criteria: (1) a potentially life-threatening disorder (e.g., anaphylaxis); (2) a chronic disease (e.g., rheumatoid arthritis, diabetes) that possibly had been triggered by vaccination; or (3) hospitalization for reasons possibly attributable to MMR vaccine. If an event occurred, a report was filed by health care personnel. The first part of a two-part form was sent immediately, with a serum sample if possible. The second part of the form was completed 2 to 3 weeks later and sent with a second serum sample. Reports were evaluated and contacts were made with the hospital or health center treating the vaccinated person if more information was needed. The authors noted that passive surveillance systems may lead to underreporting, and that active surveillance may more reliably detect adverse events. However, awareness of this potential problem prompted organization of an extensive campaign to encourage health care workers and the public to thoroughly report serious events.

From 1982 through 1996, a 14-year surveillance period, adverse events were reported for 437 vaccinations, and 173 events were considered serious according to the criteria noted above. Age at the time of vaccination ranged from 13 months to 23 years. The interval from MMR vaccination to onset of symptoms ranged from a few minutes to 80 days, with peaks during the first 24 hours and at 7 to 10 days. These cases were grouped into several categories: death, likely allergic reactions, neurologic disorders, and miscellaneous reactions. The neurologic disorders included febrile seizures, epilepsy, undefined seizure, encephalitis, meningitis, Guillain-Barré syndrome, gait disturbance, and confusion during fever.

Patja and colleagues (2000) reviewed all 173 serious adverse events reported during this period. Of these, there was one death, 73 cases (42 percent) of likely allergic reactions, 77 cases (45 percent) of neurologic disorders, and 22 cases (13 percent) of miscellaneous reactions. Peltola and colleagues (1998) followed up surveillance-system reports on 31 children, aged from 1 year 2 months to 13 years at vaccination, who developed gastrointestinal symptoms, all except one after the first vaccine dose. Hospital or health records were reviewed, or local public health nurses were interviewed. The interval between the reported event and follow-up ranged from 16 months to over 15 years (median 10 years and 8 months). Neurological symptoms originally reported in these 31 children included febrile seizures (five cases), headache (two cases), and ataxia (one child).

During the 14 years of MMR vaccination surveillance, no cases of ASD were reported or identified during the follow-up of the 31 children for whom GI disorders were reported following vaccination. Similarly, no cases of ulcerative colitis, Crohn's disease, or any other chronic disorder affecting the GI system were reported. The authors conclude that there is no evidence to support the hypothesis of an association between MMR vaccine and ASD or inflammatory bowel disease.

This study is of limited value to causality because it relies on a passive-surveillance system. Furthermore, the authors found no cases of autism among the 1.8 million vaccinees, a rate that is well below the normal incidence of autism, thereby suggesting a potential reporting bias (Wilson et al., 2003).

Case Series

United Kingdom. Wakefield and colleagues (1998) examined 12 children (11 males, 1 female), aged between 3 and 10 years, consecutively referred to the pediatric gastroenterology department of the Royal Free Hospital and School of Medicine in London. These children each had a history of normal development followed by a loss of acquired skills, including language, and of intestinal symptoms (diarrhea, abdominal pain, bloating, and food intolerance).

Eleven subjects were found to have chronic or acute and chronic nonspecific colitis (non-Crohn's disease or ulcerative colitis). Eight of the 12 subjects were reported to have reactive ileal lymphoid hyperplasia, with 3 of them also having colonic lymphoid hyperplasia, and 1 subject had just colonic lymphoid hyperplasia alone. In addition, nine of the subjects were found to have lymphoid nodular hyperplasia of the terminal ileum. Urinary methylmalonic-acid excretion was significantly elevated in eight children who were tested. The authors reported that there was no clear correlation between the endoscopic appearances and the histologic findings, although none of the findings from the 12 subjects were seen in a series of five ileocolonic biopsies from age-matched and site-matched controls with normal mucosa. In a later study, Wakefield and colleagues (2000) further examined the endoscopic and histopathological features of patients with developmental disorders and bowel symptoms. The cohort of 60 children included the 12 described above.

The authors reported the following behavioral diagnoses for 10 of the children they examined: “autism” for eight subjects; “autism? disintegrative disorder?” for one subject; and “autistic spectrum disorder” for one subject. Two subjects were diagnosed with “post-vaccinial encephalitis?” or “post-viral encephalitis?” The methods used to assess behavioral problems were not clearly stated.

Parents or doctors identified the MMR vaccine as the exposure that was linked to the onset of behavioral problems in six of the eight subjects with definitive autism and in the subject with suspected “post-vaccinial encephalitis.” The other two subjects diagnosed with autism had received MMR vaccine, but no specific exposure was linked to the onset of behavioral symptoms. Recurrent otitis media was the exposure identified for the subject with ASD; this subject had previously received MMR vaccine. Measles infection was the identified exposure in the subject with suspected “post-viral encephalitis”; this subject also had previously received MMR vaccine. For the subject with a diagnosis of either autism or disintegrative disorder, the MMR vaccine was linked to deterioration in behavior, and this subject was also reported to have shown slowed development following an earlier exposure to monovalent measles vaccine.

The time between suspected exposure and first clinical and behavioral symptoms ranged from 24 hours to 2 months, with a median of 1 week. Of the eight subjects for whom MMR had been identified as the exposure linked to the onset of behavioral problems, five had early adverse reactions (fever, rash, convulsions). Self-injury behavior was reported for three subjects; gaze avoidance (n = 2), repetitive behavior (n = 1) and loss of self-help (n = 1) were also reported. One subject was reported to have had recurrent viral pneumonia for 8 weeks following vaccination. For the 11 of the 12 subjects for whom age at onset was reported, the range in age at onset of first clinical and behavioral symptoms was 12 months to 4.5 years, with a median of 15 months. The age at first bowel symptoms was reported for 6 of 12 subjects and ranged from 18 to 30 months, with a median of 19 months.

Although these findings may identify a GI condition in a set of children diagnosed with ASD, or showing symptoms of ASD, who received MMR vaccine, they are uninformative in assessing the hypothesized causal association between MMR vaccine and autism and subsequent studies have not been able to replicate or prove this hypothesis. It is difficult to identify a specific time of onset of developmental and GI problems in young children because of the overlap in timing between the typical age at which ASD symptoms are initially suspected and the schedule for MMR and other vaccinations. In addition, given the relatively high vaccine coverage rates, many children with such problems will have received the MMR vaccine within months of the onset of symptoms.

Causality Argument

Studies examining the association between MMR and autism (see Table 10), including nine controlled observational studies (DeStefano et al., 2004; DeWilde et al., 2001; Farrington et al., 2001; Fombonne and Chakrabarti, 2001; Madsen et al., 2002; Makela et al., 2002; Takahashi et al., 2003; Taylor et al., 1999, 2002), three ecological studies (Dales et al., 2001; Gillberg and Heijbel, 1998; Kaye et al., 2001), and two studies based on a passive reporting system in Finland (Patja et al., 2000; Peltola et al., 1998), consistently showed no association. Two studies reported findings of a positive association between MMR and autism. The first was an ecological study (Geier and Geier, 2004a) that reported a potential positive correlation between the number of doses of measles-containing vaccine and the cases with autism reported to the special education system in the 1980s. The second was a study using passive reporting data by the same authors (Geier and Geier, 2003c) that reported a positive correlation between autism reports in the VAERS and estimated administered doses of MMR. However, these two studies are characterized by serious methodological flaws and their analytic methods were nontransparent, making their results uninterpretable, and therefore noncontributory with respect to causality (see text for full discussion). The case-series study by Wakefield and colleagues (1998), which originally raised the hypothesis linking MMR and autism, is uninformative with respect to causality and subsequent studies have not been able to replicate or prove this hypothesis. Based on this body of evidence, the committee concludes that the evidence favors rejection of a causal relationship between MMR vaccine and autism. This conclusion is consistent with the finding in the committee's previous report on MMR and autism (IOM, 2001a).

MMR Vaccine and Vaccine-Strain Virus in the Intestine

As discussed in a previous portion of this report, a case series published in 1998 described an unusual pathology, ileal lymphonodular hyperplasia, in the intestines of 12 children with developmental disorders, some of whom were described as being autistic (Wakefield et al., 1998). Subsequent work of that laboratory identified the measles virus genome in the ileal lymphoid tissue of 75 out of 91 developmentally disabled children and of 5 out of 70 developmentally normal control children (Uhlmann et al., 2002). Unpublished work by some of those same investigators reported the presence of measles virus F-gene in the cerebral spinal fluid (CSF) of three children with autism in whom MV genome had been previously detected in biopsies of lymphoid nodular hyperplasia (presumably these were subjects in the previously described paper) (Bradstreet, 2004; Sheils et al., 2002). Two of the three children had detectable anti-myelin basic protein (MBP) and measles virus antibodies in the CSF and all three children had detectable serum anti-MBP antibodies. Furthermore, the authors claim that the MV genome was consistent with vaccine strain. However, the specificity of the assay used for indicating vaccine-strain has been questioned (Parks, 2001; Salisbury, 2004; Sheils et al., 2002). A transcript from a court proceeding in the United Kingdom submitted to the committee suggests that strain identity as consistent with vaccine could not be confirmed by other laboratories in the three cases and that virus was absent from three cases not described in the submitted paper (Salisbury, 2004).

Nonetheless, a previous report of this committee concluded that studies documenting the presence of measles virus in the gut of children with autism are not directly relevant to the arguments (causal argument or biological mechanisms argument) regarding vaccine-based adverse events. Even if the measles virus in the gut of some developmentally disabled children were vaccine-strain (which is questionable), it would remain a matter of conjecture that the connection were causal. It is equally likely that autism leads to behaviors (e.g., abnormal eating patterns that lead to intestinal distress) or is associated with other conditions (e.g., immune deficiency) by which measles virus (vaccine-strain or wild-type) hones to altered intestines. Moreover, the finding of measles particles in the gut is not specific to individuals with NDDs; they are also found in neurologically intact individuals with severe bowel disease. Thus, the presence of virus could be secondary to autism or the bowel disease, not a predisposing or causal factor.

The implications of this work for the treatment of children with autism should not be lost in the controversies over the MMR vaccine. If some developmentally disabled children have a previously unrecognized intestinal pathology causing distress, pain, and a cycle of diarrhea and constipation, treatment could lead to a better quality of life and possibly to behaviors conducive to better learning. This could lead to very important advances in the medical and behavioral management of autism. An ongoing study that attempts to replicate the findings of ileal lymphonodular hyperplasia and viral presence in children with autism (described in a subsequent section of this report) might shed light on the role of intestinal disease in autistic manifestations.

Immune Dysregulation

A large number of studies have suggested that immune dysregulation occurs in autism (Korvatska et al., 2002; Krause et al., 2002). Decreased lymphocyte responsiveness in the lymphocyte blastogenesis assay to PHA, ConA, and Pokeweed mitogen has been reported in people with autism as compared with controls (Stubbs and Crawford, 1997; Warren et al., 1986). A study by Warren et al. of 31 autistic patients and 23 healthy control subjects showed that about 40 percent of the patients with autism had significantly reduced natural killer (NK) cell activity (Warren et al., 1987). Gupta et al. noted a decreased proportion of IFN-gamma- and IL-2 (Th1 cytokine)-staining CD4+ T cells in the serum of children with autism compared with healthy controls (Gupta et al., 1998). In contrast, IL-4-(Th2 cytokine)-staining CD4+ T cells were significantly increased in autism. The authors thus suggested an imbalance of Th1 and Th2 like cytokines in autism with consequent depressed cell-mediated immunity in autism. However, the proportion of CD4+ T cells that stained other Th2 cytokines such as IL-6 and IL-10 were not different between the two groups, making it unlikely that a simple Th1-to-Th2 shift had occurred. Moreover, it is difficult to interpret the physiological relevance of differential intracellular cytokine staining of a mixed population of lymphocytes with a whole range of antigenic specificities. Evaluating for Th1/Th2 imbalance is generally more appropriate when considering T cells with a particular antigenic specificity.

In contrast to the abovementioned study, several papers have reported increased levels of plasma IFN-gamma and IL-2 (Th1 cytokines) in people with autism as compared to healthy controls (Singh, 1996; Singh et al., 1991). Increased production of serum IL-12, IL-6, tumor necrosis factor-alpha and IFN-gamma, and increased urinary neopterin have also been reported in people with autism (Croonenberghs et al., 2002; Singh, 1996), suggestive of a generalized activation of the immune system. Consistent with this, Jyonouchi et al demonstrated that PBMCs (peripheral blood mononuclear cells) from patients with autism, both at baseline and after stimulation with LPS (lipopolysaccharide) and PHA (phytohemagluttinin), secreted significantly more pro-inflammatory cytokines (TNF-alpha, IL-1beta, and IL-6) than those from healthy controls and normal siblings (Jyonouchi et al., 2001).

Other groups have shown hypergammaglobulinemia in people with autism, with the IgG2 and IgG4 subclasses being particularly elevated (Croonenberghs, 2002). Increased IgE levels have been reported by some groups in people with autism (Gupta et al., 1996), but the physiological relevance of this is unclear, as atopy is not increased in autism. IgA deficiency has also been reported in autism. In summary, although several studies have reported abnormalities of components of the immune systems, they have often had contradictory results, making it difficult to achieve a consensus on any specific immune abnormality that might characterize autism. More fundamentally, it is not clear how these abnormalities might explain the CNS defects in autism or whether they could be secondary to GI or other complications of developmental disability.

A large number of serum autoantibodies have been detected at a higher frequency in children with autism compared to controls. The antigens against which these autoantibodies are directed include a number of CNS antigens, such as myelin basic protein and neuron-axon filament protein, but they also include a whole host of other proteins, such as nerve growth factor, serotonin receptor, alpha-2-adrenergic receptor, tubulin, heat shock protein 90, and chondroitin sulfate (Connolly et al., 1999; Cook et al., 1993; Evers et al., 2002; Singh et al., 1988, 1993, 1997; Todd and Ciaranello, 1985; Vojdani et al., 2002). Antibodies to chlamydia pneumonia, streptococcal M protein, tubulin, and milk butyrophilin have also been found at increased levels in children with autism compared to controls. This suggests that rather than there being a specific antibody response to CNS antigens, generalized hypergammaglobulinemia resulting from polyclonal B cell activation occurs in autism. Of note in the abovementioned studies, these antibodies were all also found in the serum of healthy controls, albeit at lower levels—i.e., they were not specific to autism, making their pathogenic significance questionable.

A study by Comi et al. highlighted the association of autism with autoimmune disorders in the patients' relatives (Comi et al., 1999). In this study, 46 percent of patients with autism had two or more family members with autoimmune disorders, compared to 26 percent of controls. The most common autoimmune disorders in both groups were type I diabetes, adult rheumatoid arthritis, hypothyroidism, and systemic lupus erythematosus. The increase in familial autoimmune disorders suggests a possible link, in some genetically susceptible families, between autoimmunity and autism. Of note in this study, however, is that there was no increase in the incidence of autoimmune disease in the patients with autism themselves. A major criticism of this study is that it was retrospective and thus suffered from recall bias. As stated by the authors, the families of children with autism were more likely to exhaustively assess the medical histories of all their relatives to look for a reason for their child's autism. Similar findings of family history of autoimmune disorders were reported by Sweeten et al. (2003), but were not found in another study (Micali et al., 2004).

Several autoimmune disorders have known linkage to particular major histocompatibility complex (MHC) haplotypes. An association of autism with the MHC has been reported, with an increased frequency of the extended haplotype B44-SC30-DR4 in people with autism, their mothers, or both (40 percent) as compared to controls (2 percent) (Daniels et al., 1995; Warren et al., 1992). The null allele of the C4B complement protein gene and the third hypervariable region (HVR-3) of certain DRB1 alleles, both known to be part of the ancestral haplotype B44-SC30-DR4, also are associated with autism (Warren et al., 1991, 1996). Although the possession of such a haplotype may deleteriously affect antigen presentation to lymphocytes and have immunological consequences, it is also possible that another nonimmune related gene linked to or contained within this extended MHC haplotype is more relevant to the development of autism by, for example, modifying synaptic plasticity. The relevance of this MHC haplotype is thus unclear.

Torrente et al. reported additional possible evidence for autoimmunity in autism, in the form of immune-mediated enteropathy (Torrente et al., 2002). Through immunohistochemistry, the authors demonstrated extensive lymphocyte infiltration in both the epithelium and lamina propria of the small intestine in children with regressive autism (23 out of 25 patients) compared to normal controls and patients with cerebral palsy. Epithelial IgG and complement C1q deposition was also seen in 23 out of 25 children with autism, and not seen in the other groups. It was suggested that this autoimmune enteropathy may lead to altered cognitive functioning through failure to detoxify neuroactive substances originating from the gut. In other studies, ileal lymphoid hyperplasia and lymphocytic colitis, with striking infiltration of CD8 and T cells, have been reported in children with regressive autism (Wakefield et al., 1998, 2000). It should be pointed out that these GI studies have all been in children with regressive autism, which is an uncommon³⁸ subtype of autism. It is thus difficult to know how these findings apply to the majority of people with autism. More importantly, it is not clear whether the autism is the cause or the consequence of the GI abnormalities.

Although a number of studies have suggested immune dysregulation in autism, there is as yet no evidence that these findings are directly related to the pathogenesis of autism. Unlike what is known about neuroimmunological disorders that affect the brain, such as multiple sclerosis and acute disseminated encephalomyelitis, there is no evidence of immune activation or an inflammatory process within the autistic brain. Neuropathological studies of autism have revealed no evidence of cerebral inflammatory lesions or microglial activation, which is a common feature in immune-mediated encephalitis (Bauman and Kemper, 1997). However, there are very few autopsy studies of brains from people with autism and this has not been fully investigated. Analysis of CSF from young children with autism, including screening for sensitive inflammatory markers such as quinolinic acid and neopterin, has also found no evidence of inflammation (Comi et al., 1999). The sample size in these studies is small, however.

As mentioned, autoantibodies to cerebral antigens, including MBP, have been found at higher titers in children with autism compared to controls. It is important to note, however, that these studies all evaluated serum and not CSF or brain tissue. Furthermore, it should be realized that not all autoantibodies are pathogenic—indeed, they are found in healthy individuals with no evidence of autoimmune disease. Some of the factors that determine the pathogenicity of a particular autoantibody include its affinity or avidity of binding to the autoantigen, its access to the autoantigen, and its ability to fix, complement, and facilitate cell-mediated death. With respect to the pathogenicity of serum anti-MBP antibodies in people with autism, no signs of demyelination have been found either on MRI or neuroanatomically in children with autism (Rumsey and Ernst, 2000). It is thus debatable how relevant serum anti-MBP antibodies may be to the pathogenesis of autism. It is not clear whether the dysregulation precedes the autism or is part of the syndrome.

Some investigators have tried immunotherapies, such as corticosteroids and intravenous immunoglobulin, in people with autism on the assumption that immune dysfunction plays an important role in the etiology of autism. There have been anecdotal reports both of treatment success and treatment failure (DelGiudice-Asch et al., 1999; Gupta et al., 1996; Plioplys, 2000). In any event, regardless of the results of these studies, they cannot be used to conclusively determine the importance of the immune system in autism pathogenesis, as they were not done in a double blind, placebo-controlled (i.e., nonbiased) fashion and did not use standardized behavioral tests as outcome measures.

It is clear that genetics plays a strong role in the etiology of autism. There is a 90 percent concordance rate in monozygotic twins and 5-10 percent concordance in same-sex dizygotic twins (Bailey et al., 1995; Ritvo et al., 1989). In addition, the rate of autism among the siblings of an affected child is 3-6 percent, which is 50-100 times higher than in the general population (Rutter, 1999). The significant sibling risk versus the rapid falloff in the risk for distant relatives is consistent with the involvement of multiple susceptibility genes.

Rett's syndrome, which is in the same family of pervasive developmental disorders as autism, is characterized by a course of symptoms and regression that are analogous to some children with autism (Zoghbi, 2003). The regression is seen after about a year of normal development. It was recently discovered that Rett's syndrome is caused by mutations (mainly de novo) in the methyl-CpG-binding protein 2 (MECP2) gene (Amir et al., 1999). Rett's syndrome thus provides an example of dynamic and altered brain development in a genetically determined condition. MECP2 specifically binds methylated DNA and functions as a general transcriptional repressor (Nan et al., 1998). MECP2 mutations are thought to result in at least a partial loss of function by altering the normal developmental regulation of imprinted genes, some of which are responsible for the specific neurologic phenotype of Rett's syndrome. Immune dysregulation similar to that described in autism, such as immunological abnormalities in T cells and decreased activity of natural killer cells, has been noted in Rett's syndrome (Fiumara et al., 1999). This raises the possibility that MECP2 mutations, in addition to causing the neurodevelopmental abnormality that characterizes Rett's syndrome, also cause immune abnormalities. By analogy to Rett's syndrome, similar epigenetic mechanisms may be operating in autism that simultaneously lead to abnormal development in the immune and central nervous systems (Zimmerman, 2000). However, the deviations from expected levels in various in vitro laboratory assays in both these conditions may represent only a secondary effect of the developmental or behavioral abnormalities.

Thimerosal and Autism

Apoptosis

Several investigators have explored the possibility that autism is associated with altered apoptosis, the cellular pathway of programmed cell death that is important in brain development, among other physiological processes. This has been a subject of interest because of histological findings in autistic individuals' brains of altered cerebellar purkinje cells, although a definitive role for altered apoptosis in these alterations is not established. Postmortem tissue from people with autism and matched controls has been used to measure markers of Bcl-2, a regulatory protein shown to prevent apoptosis and that has also been shown to be reduced in schizophrenic brains. One study found decreased Bcl-2 levels in autistic cerebella, suggesting a vulnerability to pro-apoptotic stimuli and to neuronal atrophy as a consequence of decreased removal of other cell types (Fatemi et al., 2001). Another study from the same laboratory reported decreases of Bcl-2 and increases of p53, also related to apoptosis, in the parietal cortex, superior frontal cortex, and cerebellar cortex of people with autism compared to matched control brain tissues (Araghi-Niknam and Fatemi, 2003). However, there is currently no evidence that such abnormalities are causally related to the development of autism. These biochemical alterations could as likely be secondary to the cause of autism.

Studies show effects of thimerosal on apoptosis in cell culture. Incubation of cultured human frontal cortex (from patients with seizure disorder), fibroblasts, and Jurkat cells (T-cell lymphoblasts) with thimerosal led to DNA damage, nuclear membrane damage, caspase 3 activation, and apoptotic morphology (Baskin et al., 2003). Jurkat cells were incubated with thimerosal in another study, leading to apoptosis, and reduced intracellular glutathione (Makani et al., 2002). Caspase 3 is an enzyme activated in apoptosis. In an extension of that work, lymphocytes taken from children with autism, and from their unaffected siblings as controls, were immortalized and exposed to thimerosal (unpublished data; Baskin, 2004). In 3 of 10 pairs of cell cultures (from cases and matched control), the samples from the child with autism showed increase in caspase 3 expression and decreases in an assay of general metabolic activity and calcine-AM esterase activity. A known neurotoxin, thimerosal has many effects in cell culture (see following section), but the relevance of these findings to a causal pathway in autism is theoretical. The absence of changes in caspase 3 expression in the majority (7 of 10) of cell culture-pairs suggests these biochemical pathways are not necessarily relevant to the expression, let alone the cause, of autism.

Methylation Pathways

Methionine synthase (MS) is an enzyme that is involved in the conversion of homocysteine to methionine. Such methylation is involved in many enzymatic pathways, including gene regulation; methylation of DNA, for example, modulates gene expression and the integrity of DNA. A recently published paper reported a new finding that dopamine and insulin-like growth factor-1 (IGF-1) stimulate methionine synthase activity and folate-dependent methylation of phospholipids in cultured human neuroblastoma cells (Waly et al., 2004). Dopamine is a neurotransmitter associated with movement, attention, memory, and many other brain functions. IGF-1 has been shown to have anti-apoptotic effects in many cells, including oligodendrocytes (those cells responsible for myelination in the brain). The authors also reported that several known neurotoxins, including thimerosal, ethanol, lead, and aluminum, inhibited these effects of dopamine and IGF-1. The authors hypothesize that disruption of this pathway by thimerosal leads to autism, ADHD, and other developmental disorders. However, the committee is aware of no evidence that autism is caused by alterations in this biochemical pathway. In addition, the evidence that several important toxicants disrupt this pathway and that it is involved in many physiological effects weakens the argument that thimerosal might cause autism through this mechanism.

Rodent Models of Autistic-like Pathology and Behavior

Three rodent models of autism-like behaviors and neuropathology have been developed. Two of the models involve infection at an early age with an infectious agent and the development of behavioral, pathological, and neurochemical abnormalities that the investigators construe as similar to autism. The third model uses thimerosal as the environmental trigger of results that are similar but not identical to the infectious models. None is as yet completely described. The thimerosal model is as yet unpublished.

The bornavirus model of autism was reviewed in the first report of the committee (IOM, 2001a). Neonatal rats infected with bornavirus develop abnormal behaviors that might be analogous to behavior expressed by children with autism, including hyperactivity, impairments in righting reflex, and abnormal play behavior (Carbone et al., 2002; Hornig et al., 1999). Histological examination shows damage to the granule cells of the dentate gyrus and cerebellar Purkinje cells. Other abnormalities include reduced (less complex) branching of some hippocampal neurons and other evidence of alterations in a components of the glutamate neurotransmitter pathway in the hippocampus (Hornig, 2004).

Increased mRNA levels for pro-apoptotic proteins (Fas, caspase-1), decreased mRNA levels for the anti-apoptotic bcl-x, and in situ labeling of fragmented DNA implicate a role for apoptosis in the cell loss. Although inflammatory infiltrates are observed transiently in frontal cortex, glial activation persists for several weeks in concert with increased levels of proinflammatory cytokine mRNAs (interleukins 1alpha, 1beta, and 6 and tumor necrosis factor alpha) and progressive hippocampal and cerebellar damage (Hornig et al., 1999).

Another rodent model of infection-mediated neuronal and behavioral abnormalities is the group A beta-hemolytic streptococcal (GABHS) model, which is being used to study a relatively new syndrome, pediatric autoimmune neuropsychiatric disorder associated with streptococcal infection (PANDAS). Behavioral manifestations of PANDAS include obsessive-compulsive disorder and tics (Swedo et al., 1998). Children with PANDAS have an increased incidence of some of the immunological and neurochemical laboratory abnormalities (e.g., evidence of antibrain autoantibodies) compared with children who have likely had the same streptococcal infection but did not develop neuropsychiatric disease (Pavone et al., 2004). Injection of an inactivated GABHS in mice leads to abnormal behavior and immunoglobulin staining in the brain (Hornig, 2004).

A similar rodent model using ethylmercury exposure is under development by the same laboratory (Hornig, 2004).⁴⁰ Investigators dosed mice in a manner intended to be analogous in time and dose to the U.S. childhood immunization schedule. Strains with differing sensitivity to autoimmune reactions were used. The SJL strain of mice reacted to the thimerosal administration differently than did four less-reactogenic mouse strains. The mice were said to display differences in weight gain, to exhibit mutilation behaviors, and to have destruction of the CA3 region of the hippocampus, increased cell density in other parts of the hippocampus and dentate gyrus, and enlarged brains. Although this model uses thimerosal and is possibly more relevant to the discussion at hand, it assumes that autism is caused by an autoimmune reaction. A previous section discussed the lack of evidence of autoimmune-mediated CNS damage in the brains of patients with autism.

The relevance of rodent models, including the three described above, is difficult to assess because the rodent “clinical” endpoints may not reflect the human ones, because there is limited understanding of the etiology of autism, and because the methods used to cause changes in the animals may bear no relationship to pathogenesis of the human disease. The committee accepts that under certain conditions infections and heavy metals, including thimerosal, can injure the nervous system. These rodent models are useful for understanding some of the processes by which these exogenous agents may exert their damage. However, the connection between these models and autism is theoretical.

Genetic Susceptibility

Despite legitimate concern about vaccine safety, few dispute the fact that adverse reactions to vaccines are rare, leading some to claim that infants and young children will not experience adverse reactions unless they possess some unusual characteristic that increases susceptibility—in particular, a genetic predisposition. This suggestion has currency now because of the discovery of genetic variants in humans that do indeed change the way individuals react to certain medications, a field of research known as pharmacogenetics. For example, Factor V Leiden deficiency, a genetic variant present in up to five percent of the population, increases susceptibility to blood clotting. Some have suggested that individuals be tested for Factor V Leiden deficiency before using hormones, oral contraceptives, or other drugs that are known to increase blood clotting. Another example is an enzyme, CYP2C9, that metabolizes warfarin, a medication commonly used for blood thinning. One percent of the U.S. population is deficient in this enzyme, and thus requires only a fraction of the normal dose of warfarin. Some have argued that patients should be tested for this enzyme before treatment with warfarin in order to avoid the risk of overdose with the drug.

These and other examples can be used to hypothesize that something similar might be operating in infants and young children exposed to certain vaccines or vaccine components. This hypothesis cannot be excluded by epidemiological data from large population groups that do not show an association between a vaccine and an adverse outcome. Depending upon the frequency of the genetic defect, a rare event caused by genetic susceptibility could be missed even in large study samples.

Some researchers working in pharmacogenetics believe that this area of research will eventually explain many of the variations in outcomes currently observed in medical practice. Others are skeptical. The committee recognizes this line of reasoning as a theoretical explanation for the data presented in this report but has found no corroborating data in the laboratory, in animals, or in humans linking vaccines or vaccine components to autism based on genetic susceptibility.

Biological Mechanisms Argument

Autism is a heterogeneous disorder with a broad range of behavioral symptoms and severity. As yet, a biological marker specific for autism has not been defined. It is thus possible that autism encompasses a spectrum of disease subtypes that have different etiologies. This could explain the wide range of immunological abnormalities that have been found in the serum of patients with autism, with some studies reporting evidence of decreased cell-mediated immunity (CMI), and others reporting increased/overactive CMI. The evidence presented for an association of autoimmunity with autism is the report that autoimmune enteropathy was seen in a large proportion of subjects with “regressive autism.” This effect is purported by some to be a clinical subtype of autism, making it difficult to extrapolate this finding to the majority of patients with autism. Other support for an association of autism with immune dysfunction includes reports of increased frequency of an extended MHC haplotype in autism, increased autoantibodies to brain antigens, and an increased incidence of autoimmune diseases noted in a retrospective study of relatives of people with autism.

However, despite evidence of immune dysregulation in the serum of people with autism, there is as yet no evidence that the immune system plays a role in the neuropathogenesis of autism. Unlike neuroimmunological diseases such as multiple sclerosis, there is no evidence of immune activation or inflammatory lesions in the brains or CSF of people with autism. Furthermore, wild-type measles infection is associated with a variety of very serious CNS damage, but it has never been associated with autism. The lack of such evidence also argues against a causal link with MMR vaccination.

It is clear from twin and family studies that there is a strong genetic basis for autism. The recent discovery of the genetic basis of Rett's syndrome, a phenotypically similar NDD with similarly described laboratory immunological abnormalities, may shed some light on the pathogenesis of autism. Similar epigenetic mechanisms in autism may lead to combined abnormal development of the CNS and immunologic irregularities.

The hypotheses reviewed by the committee were that vaccine-induced autism represents the end result of a combination of susceptibilities (possibly genetic) to immune dysfunction or to abnormal mercury metabolism. Posited intermediate steps include enzymatic abnormalities that might be related to apoptosis or cellular signaling, leading to an array of behavioral, cognitive, sensory, and motor disturbances. Other environmental exposures could have similar effects. Demonstrating an adverse effect of Hg in vitro does not readily translate into a physiologic argument. These processes are widespread in the body, so that the argument why these adverse effects cause autism must also include how the effects occur in specific tissues at specific times in development to cause the symptoms of autism, and do not cause more widespread dysfunction. No theory explains why vaccination would be the only environmental cause of these wide-ranging effects. If MMR vaccine or thimerosal is one of many environmental causes of these effects (or if thimerosal is one source of many of mercury exposure), without a specific “fingerprint” or “signature” that suggests vaccine or mercury as the causative factor, knowing if MMR or thimerosal is responsible for the adverse effects in an individual case is impossible.

Rodent models suggest that reactions to some infectious agents (e.g., bornavirus and group A streptococcus) lead to somewhat specific neuronal cell death and evidence of autoimmune reactions in the developing and adult brains of rodents. The animals also exhibit abnormal behaviors. These immunological and behavioral findings are similar to those seen in some humans after infection: the behavior in children with PANDAS or in the animal models resembles the behavior constellations in children with autism. A similar set of comparisons can be made with mercury exposures (Bernard et al., 2001), although autism has never been documented as a consequence of high-dose mercury exposure, including acrodynia. While analogies are useful for hypothesis generation, they do not substitute for direct evidence.

The committee notes several factors that limit acceptance at this time of the hypothesis that vaccines cause autism. The evidence offered for the hypothesis includes data from in vitro experimental systems, analogies between rodent behavior, and human behavior and clinical observations that are at least as well explained as being comorbid disease expressions than as causal factors. That is, it is possible that some people with autism, perhaps even a subgroup that could eventually be identified by genetic markers, have abnormal immune reactions and abnormal mercury metabolism, but that vaccination of these individuals does not cause these abnormalities or autism itself. However, the experiments showing effects of thimerosal on biochemical pathways in cell culture systems and showing abnormalities in the immune system or metal metabolism in people with autism are provocative; the autism research community should consider the appropriate composition of the autism research portfolio with some of these new findings in mind. However, these experiments do not provide evidence of a relationship between vaccines or thimerosal and autism.

In the absence of experimental or human evidence that vaccination (either the MMR vaccine or the preservative thimerosal) affects metabolic, developmental, immune, or other physiological or molecular mechanisms that are causally related to the development of autism, the committee concludes that the hypotheses generated to date are theoretical only.