I’m not going to try and prove a causal link between the two, however I am going to do a detailed commentary of this study.
Autism Occurrence by MMR Vaccine Status Among US Children With Older Siblings With and Without Autism JAMA. 2015;313(15):1534-1540. doi:10.1001/jama.2015.3077
“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.”
So the numbers are coming from the private files of an insurance company… Isn’t this a violation of privacy laws?
The restriction of an older sibling rather than any sibling who also received the MMR is interesting.
“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).”
So if there was only 1 claim code entered for a child they were not counted…
Now we are getting past the abstract into the actual privacy violating study.
“In a recent survey of 486 parents of children with ASD, nearly 20% had declined or delayed MMR immunization in the younger siblings of these children.”
Ah, that explains the age restriction.
“The New England Institutional Review Board waived the need for informed consent and deemed the study exempt based on its use of existing, deidentified data.”
Special privileges, cool.
“ASD status in index children and older siblings was determined using a claims-based algorithm with a positive predictive value of 87% that required 2 or more claims on separate dates of service with an International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis code in any position for autistic disorder, other specified pervasive developmental disorder (PDD) including Asperger syndrome, or unspecified PDD (299.0x, 299.8x, and 299.9x). Both index child and older sibling ASD status were determined using their entire enrollment time that fell within the study period. Index children had to have at least 1 older sibling with 2 claims with ASD diagnoses or all older siblings with no ASD diagnoses. Children with an older sibling with only 1 claim with an ASD diagnosis were excluded. Index children with only 1 claim with an ASD diagnosis were also excluded.”
87%? So that leaves a 13% margin of error, that is a huge margin or error. Also, only allowing for 2 claim codes leaves you wondering why you would have more than one diagnosis done unless they got it wrong the first time. Plus, how many got left out with only one claim?
“MMR vaccine receipt was defined as having a Current Procedural Terminology (CPT) or ICD-9-CM procedure code indicating receipt of each component (measles, mumps, and rubella) between birth and 5 years of age (eTable 1 in the Supplement). The date of administration of the trivalent MMR (or the last-administered component of monovalent vaccines) was used to determine age at administration for each dose (first or second).”
So they are taking it on total faith that no serious clerical errors have occurred and assuming age at time of administration as well. Judging by app’s made for keeping track of vaccinations that have been made by governing health bodies… Clerical errors seem to be rampant.
“Separate RRs were estimated for children with older siblings with and without ASD. Since no children were lost to follow-up before reaching age 5, unadjusted RRs were reported as cumulative incidence rate ratios by taking the ratio of the proportion of children who had an ASD diagnosis in an exposed group (either 1 MMR dose or 2 MMR doses) to the proportion of children who had an ASD diagnosis in the unvaccinated group at a given age.”
So they calculated the ratios separately but reported them as a cumulative total? How is this mathematically sound?
“Approximately 30% of the race/ethnicity data in this study were collected directly from public records (eg, driver’s license records), while the remaining data were imputed using commercial software (E-Tech by Ethnic Technologies) that uses algorithms developed with US Census data zip codes (zip + 4) and first and last names. This imputation method has been validated and demonstrates 97% specificity, 48% sensitivity, and 71% positive predictive value for estimating the race of black individuals. Individuals categorized as other/unknown for race/ethnicity were those whose race/ethnicity could not be assigned by the imputation algorithm or who were added to the data set after the imputation had been performed.”
So they assumed ethnic ratios with a mathematical technique rather than accurate data…
“A series of sensitivity analyses were conducted to explore the influence of potential MMR or ASD status measurement error on results. Quantitative bias analyses were implemented for both exposure and outcome misclassification following the approach described by Lash et al. More detail on bias analysis methods is provided in the online supplement (eAppendix and eTable 3 in the Supplement). In addition, associations between MMR receipt and ASD risk were also reestimated using a less-restrictive 1-claim criterion for ASD diagnosis in younger siblings. An additional sensitivity analysis was also performed rerunning final models on the subset of children with no missing data on any covariates.
Statistical significance testing of unadjusted rate ratios was conducted using the Yates χ2 test, and statistical significance testing of hazard ratios estimated by maximum likelihood were conducted using Wald χ2statistics. Likelihood ratio tests were used to test the statistical significance of Cox proportional hazards models with and without interaction terms. All statistical tests were 2-sided and the α level for all tests was .05. Analyses were performed using SAS statistical software, version 9.2.”
So they ran a bunch of different mathematical algorithms, including ones that would eliminate the previous errors. But are they reporting the better quality data or just the sloppy stuff that matches the desired outcome?
Either way it can be interesting to see what connections the authors have by looking them up on open payments and the company’s they have listed under author affiliations.
3 of the authors are affiliated with the Lewin group, 2 are with Optum, and the last one is with the A. J. Drexel Autism Institute.
Regardless of whether autism is associated with the MMR vaccine or not, this study is incredibly sloppy. This is a poor example of science and shows a disproportionate focus on only one vaccine in this question surrounding ASD prevalence. It reeks of a poorly executed attempt to refute the findings in the CDC whistle blower case without actually re-examining his findings.
If you are concerned that your data may have been used without your consent contact a lawyer that specializes in medical law to see what your options may be.