Intended for healthcare professionals

Rapid response to:


Study finds no association between autism and vaccination

BMJ 2013; 346 doi: (Published 03 April 2013) Cite this as: BMJ 2013;346:f2095

Rapid Response:

Re: Study finds no association between autism and vaccination

Of all of the papers I have reviewed over my 26-year career as a research scientist, the Destefano et al. 2013 J. Peds. article describes perhaps the most flawed and disingenuous study I have encountered.
The basis for the study is essentially a rehash of the data that was used to generate the Price et al. 2010 Pediatrics study (Price et al. 2010 “Prenatal and Infant Exposure to Thimerosal From Vaccines and Immunoglobulins and Risk of Autism” Pediatrics 126:656) that was supposed to be the CDC’s “final word” stating that thimerosal, the mercury-containing preservative in some vaccines, was in no way causally linked to autism. Not only was this original study fatally flawed due to a statistical error called “overmatching” (which I’ll discuss further below) but also the study authors hid data regarding the only valid part of the study (i.e., prenatal thimerosal exposure) which showed that children exposed to just 16 microgram mercury in thimerosal in utero were up to 8 times more likely to receive a diagnosis of regressive autism (Price C, et al. Thimerosal and Autism. Technical report. Vol I. Bethesda, MD: Abt Associates Inc; 2009).
The statistical model within the Price et al. 2009 report was run in 6 different ways. The most robust models (showing an interaction parameter between prenatal and postnatal thimerosal exposure) were run in three of the six iterations. These yielded odds ratios for autism incidence within the exposure group of 5.77 (p=0.057), 7.70 (p=0.022) and 8.73 (p=0.009) as compared to “zero exposure” controls. The study authors instead falsely reported no risk of autism associated with prenatal thimerosal exposure. Pregnant women in the U.S. are routinely given thimerosal containing flu shots recommended in any trimester, for a prenatal exposure of 25 micrograms mercury.
Within the Destefano study released last week, CDC researchers merely added up the number of vaccine antigens that the case (autism) and control (neurotypical) children were exposed to through the infant vaccination schedule. The theory that they were trying to refute essentially was “children exposed to a greater total number of antigens had a greater risk of autism.”
The first series of flaws with this study pertained to the case and control groups. As quoted from the resulting paper:
“Of the remaining 752 controls included in the analysis, 186 had an SCQ (Social Communication Questionaire) score <16 but had indications of speech delay or language delay, learning disability, attention deficit hyperactivity disorder or attention deficit disorder, or tics, or had an individual education plan.”
This clearly shows that the 186 aforementioned controls (25% of the control group) were not controls at all but instead had some underlying developmental deficit (all of which are features of autism or autism spectrum disorder). Unlike the study design described (i.e., where autism cases were matched to neurotypical controls), autism cases were matched with “cases” of other, similar neurodevelopmental maladies. Thus, you would expect to see no difference between the two groups.
The high participant refusal rate in this study is also problematic. Out of 668 cases and 2444 controls originally selected for the study, only 321 cases (48.1%) and 774 controls (31.7%) chose to participate in the research. In other words, 65% of the individuals contacted as potential participants refused to participate in the study. This indeed could produce selection bias in that the 35% of individuals that did participate were less likely to believe that vaccines were responsible for neurodevelopmental sequelae including autism.
Next, the basis of the study was to confirm or deny a correlation between the “number of antigens received” and the incidence of autism. The possible number of antigens per given vaccine was reported in Table 1 of the study. However, the term “number of antigens” is a complete oversimplification of what is actually in vaccines, their concentrations and their relative strengths in terms of inflammatory response, and is not an accurate predictor of how the body will respond to specific antigens. For example, “antigens” for the five antigen DTaP vaccines (e.g., Infanrix, which in this formulation was admittedly not used during the 1994-1999 study period, but is used as an example) include diphtheria toxoid, tetanus toxoid, pertussis toxoid, filamentous hemagglutinin and pertactin. The number “5” assigned in this category is merely the number of different antigens and doesn’t account for each antigen’s amount or relative strength. Neither does this account for the fact that Infanrix also contains aluminum (an adjuvant – designed to elicit a non-specific immune response), formaldehyde and polysorbate 80, all which could also elicit some form of inflammatory reaction. This analysis also precludes thimerosal exposure which must be considered given that the data were from children born between 1994 and 1999. Thus, the main “independent” variable of “number of antigens” within the Destefano et al. 2013 study is essentially completely meaningless.
Also, the analysis is plagued with a statistical error called “overmatching.” For a comprehensive analysis of the previous CDC study completed on the same data set (Price et al. 2010 Pediatrics), regarding thimerosal exposure rather than the number of vaccine antigens, please see Chapter 6, “Vaccine Safety Study as an Interesting Case of ‘Over-Matching’” by M. Catherine DeSoto and Robert Hitlan ( in the book "Recent Advances in Autism Spectrum Disorders - Volume I", edited by Michael Fitzgerald, ISBN 978-953-51-1021-7. The point made by Dr. DeSoto and Dr. Hitlan is that the cases and the controls in this study are too closely matched to each other. Cases were matched with controls of the same age, sex, within the same HMO and essentially the same vaccination schedule using the same vaccine manufacturers. This may be seen in Figures 1 and 2 of the Destefano et al. 2013 paper which indicated that there are almost no differences between the exposure to antigens between the case (autism) and control groups in every exposure group tested. This holds for cumulative antigen levels (Figure 1) as well as single day antigen exposure levels (Figure 2). This type of error of course precludes “finding a difference” between cases and controls because all differences were matched out case-by-case.
Finally, this type of study misses the point entirely that children with autism are physiologically different than neurotypical children. Numerous studies have shown genetic (e.g., James et al. 2006), morphological (e.g., Herbert et al. 2005) and biochemical differences (e.g., Waly et al. 2004) between these two populations. To perform a case-control study such as that presented in the Destefano et al. 2013 paper assumes a genetically, morphologically and physiologically homogeneous population, which is simply not the case. No one is claiming that children with autism or ASD got higher doses of vaccine antigens, thimerosal, MMR or whatever. What we know instead is that when these children received the same vaccines within the ACIP recommended schedule, they reacted differently. The scientists at the CDC are trained in managing infectious diseases with progressions that may be predicted with reasonable certainty. However, these neurological sequelae to vaccines are chronic, multifactorial conditions that cannot be put into the same tiny box as the common cold, influenza or chicken pox.
Given these fatal flaws, the Destefano et al. 2013 paper should not be considered as a scientific piece but instead as public relations to fulfill part of the CDC’s mission which is to maximize vaccination rates in the United States.

Competing interests: The author has a 15 year old son with neurological sequelae due to infant vaccines.

07 April 2013
Brian Hooker
Associate Professor of Biology
Simpson University
2211 College View Drive, Redding, CA 96003, USA