JAMA, 2015;313;(15):1534-1540. doi:10.1001/jama.2015.3077
| AAP: The study looked at MMR vaccine status in children with and without autism spectrum disorder (ASD) with older siblings with and without ASD. Relative risk was calculated for a child receiving an ASD diagnosis at ages 2 years, 3 years or 4 years based on 0 doses or 1 dose of MMR vaccine and whether the child had a sibling with ASD or a sibling without ASD. Relative risk also was calculated any child who got an ASD diagnosis at age 5 years based on 1) whether they’d received 0 doses, 1 dose or 2 doses of MMR vaccine, and 2) whether the child had a sibling with ASD or a sibling without ASD. No harmful association was found between MMR vaccination and ASD risk. In addition, no causal association was found between receipt of 1 or 2 doses of MMR vaccine and having a higher risk of ASD for children with an older sibling with ASD. |
Abstract
Importance Despite research showing no link between the measles-mumps-rubella (MMR) vaccine and autism spectrum disorders (ASD), beliefs that the vaccine causes autism persist, leading to lower vaccination levels. Parents who already have a child with ASD may be especially wary of vaccinations.
Objective To report ASD occurrence by MMR vaccine status in a large sample of US children who have older siblings with and without ASD.
Design, Setting, and Participants A retrospective cohort study using an administrative claims database associated with a large commercial health plan. Participants included children continuously enrolled in the health plan from birth to at least 5 years of age during 2001-2012 who also had an older sibling continuously enrolled for at least 6 months between 1997 and 2012.
Exposures MMR vaccine receipt (0, 1, 2 doses) after 1 year of age.
Main Outcomes and Measures ASD status defined as 2 claims with a diagnosis code in any position for autistic disorder or other specified pervasive developmental disorder (PDD) including Asperger syndrome, or unspecified PDD (International Classification of Diseases, Ninth Revision, Clinical Modification 299.0x, 299.8x, 299.9x).
Results Of 95 727 children with older siblings, 994 (1.04%) were diagnosed with ASD and 1929 (2.02%) had an older sibling with ASD. Of those with older siblings with ASD, 134 (6.9%) had ASD, vs 860 (0.9%) children with unaffected siblings (P < .001). MMR vaccination rates (≥1 dose) were 84% (n = 78 549) at age 2 years and 92% (n = 86 063) at age 5 years for children with unaffected older siblings, vs 73% (n = 1409) at age 2 years and 86% (n = 1660) at age 5 years for children with affected siblings. MMR vaccine receipt was not associated with an increased risk of ASD at any age. For children with older siblings with ASD, at age 2, the adjusted relative risk (RR) of ASD for 1 dose of MMR vaccine vs no vaccine was 0.76 (95% CI, 0.48-1.22; P = .25), and at age 5, the RR of ASD for 2 doses compared with no vaccine was 0.56 (95% CI, 0.30-1.04; P = .07). For children whose older siblings did not have ASD, at age 2, the adjusted RR of ASD for 1 dose was 0.91 (95% CI, 0.68-1.20; P = .50) and at age 5, the RR of ASD for 2 doses was 1.09 (95% CI, 0.76-1.54; P = .65).
Conclusions and Relevance In this large sample of privately insured children with older siblings, receipt of the MMR vaccine was not associated with increased risk of ASD, regardless of whether older siblings had ASD. These findings indicate no harmful association between MMR vaccine receipt and ASD even among children already at higher risk for ASD.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Jain, Ms Marshall, and Mr Kelly report being employees of The Lewin Group. Ms Buikema and Dr Bancroft are employees of Optum. Optum is a wholly owned subsidiary of UnitedHealth Group and The Lewin Group is an Optum company. The Lewin Group operates with editorial independence. No other disclosures are reported.
…A failure to control for healthy user bias. As has been seen in other studies, when parents of a child with autism decide not to vaccinate a younger sibling, or delay vaccination due to early warning signs, it creates the illusion that vaccinated children are at lower risk. Hviid et al. were aware of this bias—having cited Jain et al. (2015), which identified it clearly—but dismissed its relevance without conducting adequate controls. In doing so, they systematically distorted their own subgroup comparisons. Their conclusion that MMR was not associated with autism risk in girls, for example, rests on observed lower rates of autism among vaccinated girls. But this could just as easily reflect the fact that parents of at-risk girls avoided the vaccine, not that the vaccine reduced risk. – James Lyons-Weiler, PhD
Healthy User Bias in MMR-Autism Studies?
Yes. An example of HUB at work is the Jain et al 2015 MMR-autism study. In the Jain study children with pre-existing autism diagnosis are about 60% less likely to receive the MMR vaccine. Sick and already-vaccine-injured children don’t receive MMR, and therefore are concentrated in the control group. The Jain study is reviewed here: http://vaccinepapers.org/review-of-jain-et-al-jama-2015-and-comments-on-mmr-autism/
There is also evidence in the form of unusually low RRs in MMR-autism studies. Jain et al report RRs for almost all age groups less than 1. Fine and Chen assert that consistent and implausibly low RRs can be a signature of HUB. Jain in fact agrees that HUB is likely the reason for the low RRs, but then inexplicably and irrationally Jain makes no effort whatsoever to control for it. Jain et al acknowledge that this bias is happening and then they ignore it. Jain et al does not even consider the possibility that HUB could be so strong as to conceal a positive MMR-autism association. What would the RR be without HUB? Jain et al makes no effort to find out.
Above: The low RRs in Jain et al are almost certainly the result of healthy user bias (HUB). Parents that notice neurological damage in their second child avoid the MMR. Families that already have one autistic child are the most cautious. This is why the RRs are lowest when there is an older sibling with autism (circled results). From Jain et al 2015.
The results below are from a widely-cited meta-analysis of MMR-autism studies by Taylor. The odds ratio (OR) is 0.84, which is surprisingly low and almost statistically significant (with p=0.07; p=0.05 or less would indicate statistical significance). Compare this to the pooled odds ratios for Hg and thimerosal exposure, which are both 1.00. Its impossible to know for sure, but it seems likely the surprisingly low odds ratio for MMR (0.84) indicates these MMR-autism studies suffer from HUB, and conceal the true risk of the MMR vaccine. Healthy user bias has reduced the OR almost significantly below 1.0.
By comparison, we would not expect to see HUB in the Hg/thimerosal studies, because these exposures occur from the earliest vaccines, when neurodevelopmental problems have not yet appeared. It takes time for damage to manifest, and for parents and doctors to notice the damage caused by vaccines. MMR is given at an age when this damage is starting to become observable. So the MMR studies are affected by HUB, and the Hg/thimerosal studies are not. Remember, HUB occurs when: 1) poor health is noticed and 2) vaccines are not given to those with poor health.
Note that the odds ratio (OR) is different from the RR. Risk is A/(A+B), where A+B is the total sample size. Odds is given by A/B. Risk is intuitive, but odds can be confusing. If a disease prevalence is less than about 5% (as is the case with autism), then the OR is a good approximation of the RR. For uncommon diseases (with <1% prevalence) the OR and RR are almost exactly the same.
Above: Results of a widely-cited meta-analysis of MMR and autism by Taylor et al. Taylor pooled the results of the best MMR-autism studies and calculated an aggregate odds ratio (OR). All of the MMR-autism studies are non-experimental, and consequently susceptible to healthy user bias (HUB). None of the included MMR-autism studies made any effort to control or estimate HUB. The OR (0.84) was almost significantly below 1 (p=0.07), which suggests that HUB is present in the MMR-autism studies. Notice that ORs for Hg/thimerosal studies are 1.00. Hg/thimerosal exposure occurs from the earliest vaccines (before neurodevelopmental problems are apparent), and this implies that HUB will be smaller in studies of Hg/thimerosal. From Taylor et al, 2014.
Full paper (Taylor et al): Vaccines are not associated with autism: An evidence-based meta-analysis of case-control and cohort studies
In my opinion, HUB in the MMR-autism studies is caused by neurological damage from aluminum-containing vaccines given prior to MMR. Most vaccines given at 0, 2, 4, and 6 months contain aluminum adjuvant that causes neurological damage and brain inflammation. The infants injured by the 0, 2, 4, and 6-month vaccines are less likely to receive the MMR than unaffected infants (parents noticing adverse effects skip the MMR). Hence, infants injured by Al-containing vaccines are concentrated in the MMR-unvaccinated control group of the MMR-autism studies. This conceals the adverse effects of the MMR vaccine. The vaccine-injured subjects have the two characteristics required for producing HUB: 1) Lower MMR-vaccination rate, and 2) Higher risk of adverse outcome.
Above: Parents noticing damage from 0, 2, 4, and 6 month vaccines refuse or delay the MMR. The injured babies therefore are concentrated in the “control” group of the MMR-autism studies. Consequently, the control and experimental groups in these studies are not matched, and the results are wrong. This is healthy user bias (HUB) and it conceals the damage caused by the MMR vaccine.
Final Word
HUB explains why vaccine safety studies often fail to observe harmful effects. Healthy user bias almost certainly affects MMR-autism studies, because 1) MMR is given after other vaccines have caused harm, and 2) MMR is given at an age when neurodevelopmental problems are increasingly noticeable.
The CDC researchers Fine and Chen concluded that vaccine safety studies are not useful if they do not control for HUB, such as by using randomization. Very few are designed this way, so very few should be believed. There are no studies of vaccines and autism that make efforts to avoid healthy user bias.
Jain, 2015: Autism occurrence by MMR vaccine status among US children with older siblings with and without autism
There have been a lot of studies on the hypothesis that the MMR vaccine causes autism. In view of the well-proven fact that immune activation is a cause of autism, this is a reasonable and biologically plausible hypothesis. There are many credible accounts of children suffering neurodevelopmental regression and acquiring autistic behaviors immediately after receipt of the MMR vaccine. In view of this circumstantial evidence and the case reports, it is surprising that an MMR-autism association is almost completely absent in the scientific literature.
The studies on the MMR-autism hypothesis generally do not find evidence for it. Some of these studies are garbage, like the well-known Madsen 2002 study (see our review here). Another well known study by scientists at the CDC (DeStefano, Pediatrics, 2004) reported misleading/fraudulent results by not following the predetermined study protocol. This study deliberately omitted a positive association between MMR receipt and autism in boys (especially in black boys). This positive association was explored by Dr Hooker using the CDC data in this paper, which was retracted. Dr Hooker’s results are almost exactly the same as what the CDC scientists (DeStefano et al.) found and deliberately removed from their paper. I consider the retraction of Hooker’s paper to be completely unjustified and obviously political. So I accept Hooker’s results. Here are the reasons why Dr Hooker’s results should be accepted. And here is a presentation by Dr Hooker explaining this story (analysis starts at about 10:00).
The deliberate removal of the positive MMR-autism association by CDC scientists is important because it proves the CDC is motivated to conceal evidence of vaccine injury. The CDC is not impartial. The CDC will lie to promote vaccines.
Several MMR-autism papers funded by CDC were performed by Dr Poul Thorsen as lead investigator. Today Dr Thorsen is a fugitive criminal wanted for embezzlement of about $2 million in research funds. This does not necessarily mean that his results are fraudulent; but in science, reputation is important. The CDC has an obvious strong desire to conceal vaccine dangers. A dishonest, greedy scientist like Dr Thorsen is likely to fabricate results to advance the CDC’s political agenda. So I am not accepting of studies led by Dr Thorsen. Criminality or dishonesty in combination with motivation to lie is fatal for a scientific study, in my opinion (especially in highly controversial subjects). The story of Dr Thorsen is explained here in the Huffington Post.
But some of the MMR-autism studies are apparently good quality and free of fraud, conflicts of interest, or criminality, such as Smeeth 2004 Lancet. This was a large case-control study from the UK and it found no association between MMR receipt and autism. I looked at this study closely, corresponded with the author, and find it to be free of obvious flaws. I accept the results of Smeeth 2004, which are in apparent contradiction with Hooker’s results.
This was the state of the MMR-autism controversy at the time Jain’s paper was published in April 2015.
Jain et al. JAMA 2015
Full paper (Jain et al): Autism occurrence by MMR vaccine status among US children with older siblings with and without autism.
The Jain et al study received a lot of media attention when it was published April 21, 2015. It was promoted as the “final nail in the coffin” for the MMR-autism hypothesis. I disagree with this assessment.
The Jain study looked at autism and exposure to the MMR vaccine by age (at 2, 3, 4, and 5 years). Separate analyses were performed for children with and without older siblings diagnosed with autism. The idea here being that children with an autistic older sibling may be more sensitive to MMR. This is a good hypothesis.
A substantial problem with the Jain study is that parents observing developmental delays or autism symptoms will avoid MMR vaccination. Consequently, sick children (e.g. vaccine-injured children) will be concentrated in the MMR-unvaccinated control group, even if the problems are caused by vaccines (i.e. non-MMR vaccines). This of course will conceal an association between MMR and autism. If parental MMR-avoidance is large enough, it will create a negative association (i.e. the incidence of autism will be higher in the MMR-unvaccinated group). Jain et al report this negative association (since all but one odds ratio is <1.0).
In other words, there are two possible causal links between MMR vaccination and autism, and they interfere with one another:
Causal Link #1: MMR vaccination causes autism (the hypothesis under investigation)
Causal Link #2: Autism (or vaccine injury) causes MMR avoidance (due to parental concern that vaccines may cause autism. Mere indicators suggestive of autism, not enough for autism diagnosis, will also impact parental vaccination decisions)
For link #2 I prefer the term “healthy user bias”, or HUB. Its also known as “healthy vaccinee effect”. It occurs because healthy people get vaccines and sick people don’t. Its a terrible confounder for many vaccine studies, including studies of the flu vaccine. See this article on healthy user bias: http://vaccinepapers.org/healthy-user-bias-why-most-vaccine-safety-studies-should-not-be-trusted/
Above: Vaccine injury from first-year vaccines cause healthy user bias (HUB) in MMR-autism studies. Children injured by first-year vaccines are less likely to receive the MMR, and hence are used as “unvaccinated controls” in MMR-autism studies. These so-called “unvaccinated” subjects receive lots of vaccines, just not MMR. Of course, this is extremely misleading and conceals the damage caused by MMR.
To measure only link #1, there must be a way to avoid HUB, or eliminate it from the data. In a population partially accepting the possibility that vaccines may cause autism, HUB is probably impossible to avoid. There is great awareness of the vaccine-autism controversy. A 2014 survey by the National Consumers League found that about 33% of parents believe vaccines can cause autism. These parents will avoid vaccinating a child showing signs of autism or experiencing neurodevelopmental problems.
Also, its expected that parents with one autistic child may be more cautious regarding vaccination of their second child. Hence, we should expect HUB to be stronger in children that have an autistic older sibling.
Jain et al. acknowledge HUB , but they make no effort to control for it. Jain states:
“It is possible, for example, that this pattern is driven by selective parental decision making around MMR immunization, i.e., parents who notice social or communication delays in their children decide to forestall vaccination. Because as a group children with recognized delays are likely to be at higher risk of ASD, such selectivity could result in a tendency for some higher-risk children to be unexposed. To be consistent with observed data, this would need to happen more often at younger ages. This seems feasible because by the time the child is older, developmental concerns are more likely to have been confirmed or ruled out and parents may then be less worried about a new exposure, such as a vaccination, influencing a child’s developmental trajectory. Estimates at older ages would thus be less susceptible to bias related to selective parental decision making, which also aligns with the pattern observed here.” (Emphasis Added)
Jain et al, 2015
I agree with the above description of the problem. Jain says that HUB will be stronger at younger ages, and I agree.
However, I want to go further. Is it possible to estimate the strength of HUB? If so, we will be able to determine if HUB is large enough to influence the study outcome. If HUB is strong enough, it could completely reverse the study outcome.
How Strong is HUB (link #2)?
The data provided by Jain (see Table 2 in the Jain paper) enables a calculation of an estimate of the strength of HUB. Jain correctly assert that “estimates at older ages would thus be less suceptible to bias related to parental decision making…”. In other words, HUB will be strongest at younger ages, and weakest at ages 4-5. Accordingly, in order to obtain a conservative estimate of HUB, we will use the data for the oldest ages (4-5) in the Jain study.
So, lets determine the strength of HUB. We do this by calculating the use of the MMR vaccine between age 4 and age 5, and comparing vaccine usage for autistic and normal children. Specifically, this is done by looking at the number of unvaccinated at age 4 and age 5.
Having sibling with ASD:
Age 4 unvaccinated: 25/387 (25 with ASD, and 387 without ASD)
Age 5 unvaccinated: 23/269 (23 with ASD, and 269 without ASD)
The change in these numbers indicates the number that received the MMR vaccine between age 4 and 5.*
2/25 autistics were vaccinated, a rate of 8%.
118/387 non-autistics were vaccinated, a rate of 30.5%
So, non-autistics were vaccinated at a rate 30.5/8= 3.8 times greater than autistics. Parents with one ASD child are strongly avoiding the MMR vaccine in a second child with autism. Obviously this will bias the study results by putting autistics in the so-called “unvaccinated” group.
Since only 2/25 autistics were vaccinated, this calculation has substantial statistical uncertainty. The 3.8 ratio will have a wide confidence interval. This issue is ignored in this simple analysis.
Lets repeat the calculation for the non-ASD sibling group.
Having sibling without ASD:
Age 4 unvaccinated: 65/11957
Age 5 unvaccinated: 56/7735
9/65 autistics were vaccinated, a rate of 13.8%
4222/11957 non-autistics were vaccinated, a rate of 35.3%
So, non-autistics were vaccinated at a rate of 35.3/13.8=2.6 times greater than autistics. Parents without an autistic older child are also avoiding the MMR in a second child with autism.
As expected, parents that already have one autistic child (the older sibling) are more strongly avoiding the MMR vaccine in their second child (since 8% < 13.8% and 30.5% < 35.3%). This is the case whether the second child has autism or not. This is a clear demonstration that autism in the family leads parents to avoid MMR vaccination.
Rate ratios of 2.6 and 3.8 are rather strong effects. We will choose the more conservative rate ratio of 2.6 for further analyzing the results. And remember that Jain asserted that parental vaccine avoidance will be weakest at age 4-5, compared to younger ages. So at younger ages, the rate ratio is likely to be even greater than 2.6, according to Jain et al.
The rate ratio gives us the information we need to remove the effect of parental vaccine avoidance due to autism. It allows us to approximately remove the effect of HUB, and thereby get a more accurate estimate of link #1.
Removing the Effect of HUB
The question now is: what effect will an avoidance rate ratio of 2.6 have upon data at the other ages? Will this be enough to change the outcome of the study?
To answer this question, we calculate odds ratios of autism for the other age groups, corrected by a factor of 2.6.
We combine all the data for each age (non-ASD sibling and ASD sibling). This makes the calculation more conservative, since the parental avoidance is stronger in the ASD older sibling group.
Autistics are 1/2.6 (38.5%) as likely to be vaccinated as non-autistics, due to autism-motivated parental behavior (i.e. HUB). Removing this effect requires increasing the number of vaccinated, autistic subjects by a factor of 2.6. Alternatively, the same correction can be obtained by reducing the number of unvaccinated autistics by a factor of 2.6 (i.e. removing the autistics that were unvaccinated due to parental observation of autism). Either choice has the same effect on the odds ratio calculation.
Results of the corrected calculation are shown in the table below.
Above: The odds ratios are corrected by a factor of 2.6 to remove the effect of HUB. The resulting corrected odds ratios are now greater than 1, suggesting that MMR vaccination is indeed associated with autism.
The important thing to notice is that the corrected odds ratio is greater than 1. This means that HUB is strong enough to account for the finding of no association by Jain et al. In fact, HUB appears to be so strong as to conceal a positive association between MMR and autism. In other words, when the effect of HUB is removed, the Jain et al study data shows that MMR is associated with autism.
However, I am not arguing that Jain et al. provides evidence for a positive MMR-autism association. I think there are too many assumptions for such a claim. Rather, I assert that HUB is strong enough to make the Jain et al study unreliable. The Jain et al results are essentially meaningless. The reason why Jain et al. find no MMR-autism association is simply because of the effect of autism on parental vaccination choices, not the other way around. When parents observe autism or autism traits, they avoid vaccination, and the result is that autistic children wind up in the “unvaccinated” group. Obviously, this is a highly misleading effect that conceals an autism-causing effect of MMR.
HUB is why MMR-autism studies are wrong
Many vaccines, including MMR, cause autism. Vaccines cause autism if they stimulate interleukin-6 or interleukin-17a production in the brain. The autism effect is not observed in Jain et al due to its failure to control for HUB.
There are profound difficulties in studying health outcomes in a population that is aware of the hypothesis under investigation. Many people are aware of the vaccine-autism controversy and believe in the link. When they adjust their behavior accordingly, the direction of causation is reversed and the phenomenon becomes almost impossible to observe.
HUB biases every MMR-autism study. Unfortunately, the other MMR-autism studies do not provide the data necessary to calculate the strength of HUB. Jain et al. is the only study I am aware of that provides data sufficient to estimate the strength of HUB.
HUB is a good reason to be skeptical of all human vaccine-autism studies, and instead look at controlled animal studies. Animal studies always have better controls and do not suffer from biases due to behavior of study subjects. And many animal studies prove beyond any doubt that activation of the immune system during brain development causes autism, schizophrenia, epilepsy, seizure disorders and other brain injury. Animal studies also prove beyond doubt that aluminum adjuvant in vaccines can cause brain injury, at dosages routinely given to human infants.
Researchers at the CDC published an important paper in 1992 on HUB, showing that it can be strong enough to completely reverse the outcome of vaccine safety studies. See this article for a detailed explanation: http://vaccinepapers.org/healthy-user-bias-why-most-vaccine-safety-studies-should-not-be-trusted/
Final Word
The Jain study only looked at MMR. Media reports about this study have falsely and deceptively asserted that the Jain study shows that “vaccines” in general do not cause autism. In reality, the Jain study says nothing about other vaccines. The MMR vaccine is the only vaccine that has been much studied in relation to autism, and all of the MMR-autism studies suffer from HUB. The other likely more dangerous aluminum-containing vaccines, given at younger ages, have hardly been studied at all. It is a blatant lie to claim that the science shows “vaccines” in general do not cause autism.
The science actually shows the opposite. Controlled animal experiments overwhelmingly prove that immune activation (i.e., interleukin-6) in the developing brain causes autism. Animal experiments also prove that aluminum adjuvant causes brain damage, at dosages human infants routinely receive from vaccines. There is also evidence that aluminum and the Hepatitis B vaccine stimulate interleukin-6 production in the brain. Hep B vaccine evidence: http://vaccinepapers.org/two-vaccines-opposite-effects-brain/
NOTES:
* I have made two assumptions here: that the 269 are all derived from the 387, and the 23 were all derived from the 25. However, this is not necessarily true (some of the 387 could have become part of the 23 autistic at age 5). I believe any resulting error is small. Also, I believe this error will not likely affect the ratio calculation, because it likely affects both autistics and non-autistics similarly (we are calculating a ratio, after all). I would need more detailed data from Jain to know for sure (Jain has not responded to this request).