The title for this article should have a question mark, “A Positive Association found between Autism Prevalence and Childhood Vaccination uptake across the U.S. Population?”. The paper’s faults have already been discussed, but I was unable to sleep earlier this week and I decided to graph some of the data. For some reason, even this didn’t help me to sleep.
here is the abstract:
The reason for the rapid rise of autism in the United States that began in the 1990s is a mystery. Although individuals probably have a genetic predisposition to develop autism, researchers suspect that one or more environmental triggers are also needed. One of those triggers might be the battery of vaccinations that young children receive. Using regression analysis and controlling for family income and ethnicity, the relationship between the proportion of children who received the recommended vaccines by age 2 years and the prevalence of autism (AUT) or speech or language impairment (SLI) in each U.S. state from 2001 and 2007 was determined. A positive and statistically significant relationship was found: The higher the proportion of children receiving recommended vaccinations, the higher was the prevalence of AUT or SLI. A 1% increase in vaccination was associated with an additional 680 children having AUT or SLI. Neither parental behavior nor access to care affected the results, since vaccination proportions were not significantly related (statistically) to any other disability or to the number of pediatricians in a U.S. state. The results suggest that although mercury has been removed from many vaccines, other culprits may link vaccines to autism. Further study into the relationship between vaccines and autism is warranted
The author made a number of strange decisions in this paper, as already discussed (and here, here, and here). First, she chose “autism prevalence” for her title when what she discussed was a combination of autism and speech or language impairment. So, I will put quotes around “autism” in “autism” prevalence, as this isn’t a real autism prevalence. Second, she chose a vaccination rate that is based on 100% completion of the 1995 vaccine schedule. This rate was changing notably, as she starts the study period when the schedule was introduced. So, as states and pediatricians and parents adopted the schedule, the “vaccination rate” as defined by the author increases notably. Again, I will use quotes around “vaccination rate” as this is an odd definition of the term.
Here is the main result of the paper:
The results suggest that if a given U.S. state has a 1% higher vaccination rate than another U.S. state, then the state with the higher vaccination rate might have, on average, a 1.7% higher prevalence of autism or speech disorder
With more than 4 × 10^6 babies born in the United States each year, this finding translates into an additional 680 children (= number of children [4 × 10^6] × coefficient [0.017] × 1% [0.01]) exhibiting autism or speech disorders for every 1% rise in children receiving the 4:3:1:3:3 series of vaccinations by age 2 years.
To put all this simply, the author is claiming that if there is some baseline prevalence of “autism” if the “vaccination rate” is 0, say 5%, then the prevalence rate of “autism” would be 5+1.7=6.7% if the “vaccination rate” were 100%.
One would expect that as “vaccination rates” go up or down, the “autism rate” would go up or down with this proportionality factor. It doesn’t happen that way, though.
The author used household income and ethnicity (%Hispanic, % African America, %Other) as variables in the model. Let’s assume that those numbers don’t change significantly during the time period considered for each state (the author appears to make this assumption, so let’s go with it.)
I took a look at the first 4 states in the table (listed alphabetically): Alabama, Alaska, Arizona and Arkansas. If anyone has a particular interest in any given state, I’ll graph them up (or you can do it yourself).
Here is the “vaccination rate” as a function of study year:
As Prometheus has noted, this rate shows the biggest change in the first two years. Given the result of the study, we should see the biggest changes in “autism rate” in these two years. But we don’t. I took the data for the “vaccine rates” as a function of time and applied the 1.7% increase in “autism” prevalence the author states as a result. Let’s look at these states and what the model predicts and what the actual data showed (click any graph to make bigger):
The data not follow the predicted trends. Not even close. Not only that, but for two states, the predicted values are higher than the reported values (red curves higher than black) while for the other two states the opposite is seen.
This isn’t a case of “I don’t know how the analysis came to the conclusion but I don’t think it is right” type of paper. This is a case of “how did this get past an editor and referee” type of paper. It is just that clearly wrong.