Jon Mitchell brought to my attention some data I hadn’t seen before on DTP coverage from California. The data come from a post by JB Handley over at AoA where we find he’s still complaining about a Simpsonwood “cover-up” of sorts (decisively debunked back in 2005). The data, apparently obtained via FOIA, was being analyzed by the CDC back in 2001, at a time when an association between thimerosal and autism was still considered an open question.
What I think is interesting about the data is that I’ve found they convincingly show there’s no association between 4+ DTP vaccination coverage by age 2 (not just thimerosal) and autism. By extension, I believe this is fairly good evidence that there’s no association between vaccination in general and autism. As usual, my analysis can be easily replicated by readers. The only data outside of that provided by JB Handley that I will use is population data from California EPICenter. Since EPICenter doesn’t provide population data for all those born between 1980 and 1994, I’m using data on 11 year-olds from 1991 to 2005. This should be a good approximation.
Of course, I cannot prove that extremely rare cases of vaccine injury don’t manifest as autism. This is not demonstrable with ecological data of this nature. Obviously, JB Handley’s intention in the first place is to show that fluctuations in vaccine uptake have a very noticeable effect in autism prevalence by birth year. This is the type of claim I will scrutinize.
First, let’s look at what I call the “naive” correlation between DTP coverage and autism prevalence by birth year.
The figure above (click to enlarge) shows two upward trends, one for DTP coverage and one for the administrative prevalence of autism by birth year. Some people might argue that the coincidence of upward trends indicates there’s a potential association. That the administrative prevalence of autism was on an upward trend is not surprising. It has been on an upward trend pretty much since records have been kept in California. Is it surprising that vaccination uptake was on an upward trend as well during the years graphed? Let’s consider this question. How many things were not in an upward trend during the 1980s and 1990s? How many of those are associated with modernity and scientific progress?
It’s interesting that such naive and expected correlations are exploited to this day, for example, in the recent Young-Geier study.
After taking a closer look at the graph, an actual association is not clear at all, at least as far as I can tell. We could leave it at that. After all, these sorts of graphs are not evidence of an association at all. But I think we can do better and get to the bottom of this.
Since the graph is supposed to show that a change in vaccination uptake can significantly alter the administrative prevalence of autism, I think it’s fair to assume that small fluctuations in the uptake would tend to be associated with small fluctuations in prevalence, in roughly the same direction. But let’s be clear here. I’m not saying that every year there’s a DTP fluctuation, there should be a matching autism fluctuation. After all, some randomness is expected. What I’m saying is that we should expect to find a statistical trend in the association between the fluctuations, even if it’s a weak trend.
Let me describe the methodology I will use. I will have Excel produce order-2 polynomial models for both the DTP and autism trends. (These give excellent fits, with R2=0.94 for DTP and R2=0.99 for autism, but I think other types of models would produce basically the same findings.) Then I will obtain DTP and autism deltas by subtracting actual DTP and autism values minus expected values based on the corresponding model formulas. (A delta will be a small positive or negative number whose meaning can be thought of as “how different from expected the value is in a given year.”) Finally, I will have Excel produce a scatter graph of DTP deltas vs. autism deltas. Using this graph we will determine whether there’s a trend.
The polynomial models are the following.
Autism: y = 0.1013x2 – 0.2352x + 4.4526
DTP: y = 0.1598x2 – 0.3683x + 51.3715
The variable ‘x’ is the year of birth minus 1980. The variable ‘y’ is the expected prevalence or DTP coverage value. With the help of these formulas, we obtain the following deltas. The reader is encouraged to visually confirm the validity of these deltas using the figure above.
(Year, Autism Delta , DTP Delta)
1980, -0.48 , -0.47
1981, 0.06, 4.24
1982, 0.26, 0.83
1983, 0.22, – 4.0
1984, 0.16, – 3.56
1985, 0.40, 0.77
1986, 0.66, -0.81
1987, -0.68, -1.32
1988, -0.50, 2.25
1989, -0.73, 1.20
1990, -0.53, 2.23
1991, 0.10, 0.64
1992, 1.52, -0.16
1993, 0.14 , 0.01
1994, -0.58, -1.84
This is a completely random scatter of dots. In fact, if we try to fit a linear model to it, we get a slight downward trend. If DTP vaccination coverage were associated with autism, I would expect to see a more clear scatter and an upward trend.
I don’t know about my readers, but I personally find this data quite convincing, and I’m glad JB Handley brought it to our attention. Vaccines (at least the DTP vaccine) are not associated with autism at the population level.