Volume 364, Issue 9438, p963-969, September 11, 2004
MMR vaccination and pervasive developmental disorders: a case-control study
Liam Smeeth, MRCGPa autism@lshtm.ac.uk ∙ Claire Cook, MScb ∙ Prof Eric Fombonne, FRCPsychc ∙ Lisa Heavey, PhDd ∙ Prof Laura C Rodrigues, PhDb ∙ Prof Peter G Smith, DScb ∙ Prof Andrew J Hall, FRCP
Summary
Background
Concern that measles-mumps-rubella (MMR) vaccination might cause autism has led to a fall in vaccine coverage. We investigated whether MMR vaccination is associated with an increased risk of autism or other pervasive developmental disorders.
Methods
We did a matched case-control study using the UK General Practice Research Database. Cases were people born in 1973 or later who had first recorded diagnosis of pervasive developmental disorder while registered with a contributing general practice between 1987 and 2001. Controls were matched on age, sex, and general practice.
Findings
1294 cases and 4469 controls were included. 1010 cases (78·1%) had MMR vaccination recorded before diagnosis, compared with 3671 controls (82·1%) before the age at which their matched case was diagnosed. After adjustment for age at joining the database, the odds ratio for association between MMR and pervasive developmental disorder was 0·86 (95% CI 0·68–1·09). Findings were similar when restricted to children with a diagnosis of autism, to those vaccinated with MMR before the third birthday, or to the period before media coverage of the hypothesis linking MMR with autism.
Interpretation
Our findings suggest that MMR vaccination is not associated with an increased risk of pervasive developmental disorders.
WHAT CRITICS SAID:
■ Problems in the study design operate against the probability of detecting an increase in risk.
■ There are significant changes from the methodology first proposed and subsequently cited in the present paper. Critics say it was crucial that case groups comprised only regressive, or late-onset, PDD, but Smeeth et al. confess (on p 967) that they were unable to do this.
■ The small sample size is an issue – In order to complete such a matched-pair study with an estimated control exposure rate of 80%, the appropriate sample size would be 7,145 cases – almost six times the number of cases used in Smeeth et al.
■ Two failings (below) in particular are “of such significance as to invalidate the conclusion that MMR vaccine is not associated with onset of autism in children.” (Wakefield)
■ The authors state that they were “not able to separately identify the subgroup of cases with regressive symptoms to investigate the hypothesis that only some children are vulnerable to MMR-induced disease and that this is always regressive. In this single statement they make it clear that they have not conducted an investigation of “what has been referred to as ‘the Wakefield hypothesis.’”
■ The paper “cannot be said to have concluded anything of relevance to the (Wakefield) hypothesis and has been grossly over-interpreted.
■ Despite the authors’ assurance that all diagnoses would be validated by a detailed review of hospital letters and information from parental questionnaires, only 25% of cases had their records examined and no questionnaire was used.
■ Other “substantial changes in the methodology” were also not explained in the paper, which therefore “meets neither the criteria for testing the original question nor those laid-down by the authors themselves.”
WHAT THE COCHRANE REPORT SAID:
■ Although the study “appeared to be carefully conducted and reported,” the database used “had no unexposed (to MMR) representative controls.” And though the 4% to 13% figure of unexposed controls was regarded by the authors as “representative,” such small numbers “may indicate some bias in the selection of controls.”
■ This underrepresented control “problem appeared to provide the rationale for the design of DeStefano 2004” (another study reviewed by Cochrane).
■ In this study, it was “impossible” to determine the “precise nature of controlled unexposed to MMR and its generalizability.”
■ This study suffered from a “moderate” risk of bias.
■ This study failed to report complete vaccine identification information, “including lot numbers, adjuvants, preservatives, strains, product and manufacturer.
■ This study failed to report any vaccine strains at all.
■ The authors provided “inadequate” explanations for missing information, even though there were “clearly missing unintended-event data” on as many as 20% of the participants.
SUMMARY:
The Smeeth et al. study goes a step further in highlighting the inadequacy of study designs that fail to isolate the correct case-group by specifically stating their intention to do so, (in a pre-study protocol discussion) and then clearly informing the reader that they failed to deliver on this intention. This represents a fundamental and fatal failure to address the right hypothesis. This means, in turn, that the study fails to add any data of scientific value regarding the vaccine-autism hypothesis whatever its other features might be.