A population-based study of measles, mumps, and rubella vaccination and autism
N Engl J Med. 2002 Nov 7;347(19):1477-82.
doi: 10.1056/NEJMoa021134.
Kreesten Meldgaard Madsen 1 , Anders Hviid, Mogens Vestergaard, Diana Schendel, Jan Wohlfahrt, Poul Thorsen, Jørn Olsen, Mads Melbye
| AAP: No link was found between autism and age at receipt of MMR vaccination, time since MMR vaccination, or date of MMR vaccination in a population-based study of 537,303 children born between 1991 and 1998. This study was conducted in Denmark. |
Abstract
Background: It has been suggested that vaccination against measles, mumps, and rubella (MMR) is a cause of autism.
Methods: We conducted a retrospective cohort study of all children born in Denmark from January 1991 through December 1998. The cohort was selected on the basis of data from the Danish Civil Registration System, which assigns a unique identification number to every live-born infant and new resident in Denmark. MMR-vaccination status was obtained from the Danish National Board of Health. Information on the children’s autism status was obtained from the Danish Psychiatric Central Register, which contains information on all diagnoses received by patients in psychiatric hospitals and outpatient clinics in Denmark. We obtained information on potential confounders from the Danish Medical Birth Registry, the National Hospital Registry, and Statistics Denmark.
Results: Of the 537,303 children in the cohort (representing 2,129,864 person-years), 440,655 (82.0 percent) had received the MMR vaccine. We identified 316 children with a diagnosis of autistic disorder and 422 with a diagnosis of other autistic-spectrum disorders. After adjustment for potential confounders, the relative risk of autistic disorder in the group of vaccinated children, as compared with the unvaccinated group, was 0.92 (95 percent confidence interval, 0.68 to 1.24), and the relative risk of another autistic-spectrum disorder was 0.83 (95 percent confidence interval, 0.65 to 1.07). There was no association between the age at the time of vaccination, the time since vaccination, or the date of vaccination and the development of autistic disorder.
Conclusions: This study provides strong evidence against the hypothesis that MMR vaccination causes autism.
Walter Spitzer, Professor Emeritus of Epidemiology, McGill University et al., in a letter published in the March, 2003 issue of the NEJM, noted that there were still some methodological problems outstanding with regard to the Danish study.[4]
Spitzer charged that researchers did a clinical record review of just 40 cases (13%), which he claimed was inadequate, especially if the purpose was only to validate an existing diagnosis. Spitzer claimed that ‘…without a multidisciplinary review of original lifetime records as well as double verification in a large descriptive single cohort, important errors would have been unavoidable, both in classification and numbers for the numerators.” Spitzer et al. also raised the question of whether pediatric clinical psychologists, pediatric neurologists and speech therapists were involved in the review and whether the reviewers were blind as to exposure status.
Though the power of the published study was high, it was “misleading,” Spitzer et al. claimed. In elaborating this point Spitzer et al. explained that if, for example, one assumed a vulnerability to MMR-induced disease in 10% of the regressive ASD cases, with 95% of this group being vaccinated, and if 80% of the non-regressive ASD cases were also assumed to be vaccinated, then “the odds ratio for MMR as a risk factor for regressive autism would be 4.17.”
However, if children with autism, regardless of sub-types, were combined and compared against non-affected controls, the odds ratio would plummet to just 0.97. “Thus a small non-statistically significant reduction in uptake of MMR in the 90% of non-regressive autistic children would mask a strong causal association in a small subgroup,” Spitzer et al. Whilst the sub-group might be small, they claim, ‘…conservatively the 10% would represent 50,000 children in the U.S. alone with a financial burden of disease to parents and government of at least $1.25 billion per year.”
Goldman and Yazbak, in a letter published in the Journal of American Physicians and Surgeons, pointed out the “substantial under-representation of autism diagnoses and vaccination status for children born in the later study years.[5]” Children with ASD in Denmark are diagnosed at about 5 years old; many were simply too young to receive an ASD diagnosis by the end of the study period. This would apply to all children under the age of 36 months and, in a practical sense, to many of the 3-5 year olds. Among children born in 1997 and 1998, who made up a substantial proportion (39%) of the total years of observation time, many had yet to even receive an MMR vaccine all.
In fact, ASD prevalence among children aged 5-9 years increased from a mean of 8.38/100,000 in the pre-licensure era (1980-1986) to 71.43/100,000 in 2000, making the adjusted prevalence rate-ratio 4.7 for the post-licensure period compared with the pre-licensure period. This suggested a temporal association between the introduction of MMR vaccination in Denmark and an almost five-fold increase in autism cases.
Mark Blaxill, SafeMinds director, in an unpublished critique written for SafeMinds , criticized the use of person years rather than prevalence by birth group as the choice of outcome measure. He pointed out that although person-years is common incidence measure in epidemiological studies, it is an odd choice in the study of a chronic disease like autism. He argued that “there is no really good reason (and the authors offer none) to consider duration of the disorder as opposed to its presence. Autism is generally considered a lifelong disorder, so the effect is the same among two year olds as it is among eight year olds.”
Analyzing the Denmark data using case prevalence measures reveals importance problems with Madsen et al. according to Blaxill. Most notably, he points out that the most straightforward analysis of the data provided by the study authors directly contradicts their conclusion (see table below). The actual prevalence of autism in the 440,655 children who received MMR vaccinations in Denmark was 6.1 per 10,000 as compared to the rate of 4.9 per 10,000 in the 96,648 unvaccinated children. At the population level, the risk of autism was therefore 26% higher in the group vaccinated with MMR, a calculation the authors never reported. Blaxill highlighted two biases:
| Unadjusted Relative Risk of Autism in MMR-Vaccinated Danish Children | |||
| Total | Vaccinated | Not vaccinated | |
| Population | 537,303 | 440,655 | 96,948 |
| Cases | |||
| Autistic | 316 | 269 | 47 |
| Other ASD | 422 | 352 | 70 |
| Total ASD | 738 | 621 | 117 |
| Rates per 10K | |||
| Autistic disorder | 5.88 | 6.11 | 4.86 |
| Other ASD | 7.85 | 7.99 | 7.24 |
| Total ASD | 13.74 | 14.09 | 12.11 |
| Relative risk | |||
| Autistic disorder | 1.26 | ||
| Other ASD | 1.10 | ||
| Total ASD | 1.16 | ||
1. Biased exposure adjustment. Madsen et al. introduce an adjustment for the timing of diagnosis relative to the timing of MMR vaccination. The authors determined that six of the children diagnosed with autism and seven of those diagnosed with other autistic spectrum disorders had such an early onset of the symptoms that the disorder was diagnosed before the MMR vaccine was administered. They decided that this reversed sequence of events argued against a causal role for MMR in autism, so they placed these vaccinated children in the group they called “unvaccinated” even though they had clearly received MMR vaccine. In moving autistic children into the unvaccinated group, the authors increased the pool of unvaccinated children by 13% and reduced the pool of vaccinated children by 2%. This adjustment substantially reduced the relative risk of autism among the vaccinated group, from 1.26 in the table above to 1.09 at the population level.
The MMR hypothesis argues more specifically, however, that vaccination of an otherwise normal child will contribute to an autistic regression. To test this hypothesis, the most relevant population would exclude all cases with early onset autism, both from the vaccinated group (as the study authors chose to do) and from the unvaccinated group (which they chose not to do). The method chosen by the authors artificially raised the incidence rate in the control group. If, instead of moving early onset (but clearly vaccinated) cases into the “unvaccinated” group, the authors had removed all early onset cases from both groups, they would have increased the relative risk of autism in the vaccinated group to 1.28, instead of reducing it to 1.09.
2. Age adjustment bias effect. The use of person-years also had a more direct effect on the published risk of MMR on autism, by introducing a skew in the sample by age group. In reporting a relative risk 0.92 (a level that suggests a protective effect of MMR in autism) rather than the case-based relative risk of 1.09, the authors weighted certain portions of their sample (the older children with more person-years) more heavily than others. Given the wide variation of risk by age, extra weight was actually given to portions of the sample with relative risks well below 0.92. Blaxill made a rough calculation of the relative risks of autism comparing children born in 1997-98 (children were one and two years of age when the data was collected) to those born between1991-96. This calculation shows an even higher protective effect (relative risks of 0.87 and 0.77 for children with autism and autism spectrum disorders, respectively).
The distribution of relative risks is highly variable across birth years, more than would be expected under the null hypothesis of no vaccination effect. This raises questions about the quality of the vaccination data records among the older children. The apparent high rate of autism among unvaccinated older children could reflect lost vaccination records or other data integrity problems.
Carol Stott, Mark Blaxill, and Dr. Andrew Wakefield, claimed in the Journal of American Physicians and Surgeons, that Madsen et al. appeared to have adjusted inappropriately for age.[6] That being the case, Stott et al. argued, the findings need to be reinterpreted,” Stott et al. went on to state that in the absence of such adjustment, there is a statistically significant 45% excess risk of autism in recipients of the MMR vaccine and therefore, an apparent association between MMR and autism in this Danish population.
In addition, Stott et al. argued that a proper trend analysis would compare autism rates not by age at diagnosis but rather by date of birth. They obtained data from the same registry used by the authors that showed a clear upward trend in autism rates in birth cohorts born after the introduction of MMR in Denmark (see below). ASD Prevalence in Denmark by year of birth, 1982-1992. Annual growth rate before MMR was -0.5%, but rose to 14.8% after MMR introduction in 1986.
Moreover, Stott et al. claimed that the authors of the Danish study had selected a particular adjustment to their population groupings that removed a total of 13 ASD cases from the vaccinated group and placed them in the unvaccinated group. This single adjustment reduced the relative risk of autism associated with MMR vaccination at the population level by 17%, from 1.26 to 1.09, Stott et al. claimed that if the authors had removed all cases diagnosed before two years of age from their risk analysis, the relative risk at the population level would have risen from 1.26 to 1.28.
And Blaxill added another commentary of his own: “I conclude that the authors’ conclusion is not warranted. In my opinion, the Madsen article is useful in many ways but it definitely does not rule out MMR as a cause of autism, particularly not in a subgroup of the affected children.”
WHAT THE COCHRANE REVIEW SAID:
■ Follow up on medical records terminated just one year after the last day of admission to the cohort. “Because of the length of time from birth to diagnosis, the Cochrane reviewers felt it became ’… increasingly unlikely that those born later in the cohort could have a diagnosis.”
■ The study was judged to have a “moderate” probability of bias.
■ Interpretation of the study was “made difficult by the unequal length of follow up for younger cohort members” and the “use of date of diagnosis rather than onset of symptoms for autism.”
■ The study failed to report complete vaccine identification information, “including lot numbers, adjuvants, preservatives, strains, product and manufacturer.”
■ There was inadequate description of exposures, such as vaccine content and schedules.
■ The study suffered from “clearly missing unintended-event data” and many participants were missing for adverse event monitoring. Adverse event data were missing in up to 1-in-5 participants (20%).
■ The study failed to provide descriptions of all outcomes monitored.
SUMMARY:
Madsen et al. argue no effect of MMR vaccination on autism in Danish children and even suggest there might be a protective effect to MMR exposure. Unfortunately, their study is plagued with questionable methodological choices, unexplained data anomalies and biased adjustments. In any study that asks a fundamental question about relative proportions of exposure in affected vs. unaffected groups, accurate definitions and classifications of (a) exposure and (b) affected status are crucial to the validity of any conclusions drawn from the data. Numerous criticisms of Madsen et al. highlight a source of error in one or another of these classifications. Methodology questions aside, more straightforward approaches to the population data they report suggest an increased risk of autism in Danish children based on MMR exposure, especially when adopting a case-based approach rather than relying on person-years. A simple comparison of autism rates by birth year shows a clear increase in autism rates after the introduction of MMR in Denmark. These analyses demonstrate that frequent references made based on Madsen et al. regarding the safety of MMR are incorrect.