Some post-holiday antivaccine “science”

I hope that you and yours are having a fantastic holiday season thus far. Yesterday, we had a great family gathering, after which I settled down to watch the Doctor Who Christmas special; all in all, a most excellent Christmas Day. Unfortunately, towards the later part of the day, someone out there sent me an e-mail and, fool that I was, I actually read it. (Who is sending e-mails about bad science to random bloggers on Christmas evening, I ask?) So when I woke up this morning, fool again that I am, I actually read the danged thing.

Of course, I should have known that this was going to be truly, truly bad when I saw the names of the authors: Gary S. Goldman and Neil Z. Miller. You might remember them before, because they’ve published some spectacularly bad “science” in the past, papers so awful that I can’t help but wonder how even a bottom-feeding journal would let such tripe pass peer review. For example, do you remember a study that purported to show that the number of vaccines received correlates with infant mortality rates? It was a truly putrid attempt to slime vaccines that cherry-picked the data years chosen to analyze, as well as the countries chosen, all topped off with the typical antivaccine tactic of trying to artificially pump up the number of vaccines children receive by, for instance, counting multivalent vaccines multiple times even if they are only given in one dose, as has been explained before with respect to their previous atrocity against science. There are a number of examples that can be found by simply typing “Goldman Miller vaccines” or “Goldman vaccines” into the Search box of this blog, the most recent of which occurred less than a month ago and involved Dr. Goldman going solo to dumpster dive the Vaccine Adverse Events Reporting System (VAERS) database in search of a fantastical correlation between the H1N1 vaccine and miscarriages.

Come to think of it, much of what Goldman seems to do involves dumpster-diving the VAERS database, and Goldman and Miller’s latest antics are no exception, as you will see in a moment. The reasons should be obvious: VAERS data are publicly accessible and downloadable, and it’s a passive reporting database to which anyone can report a case as an “adverse event” related to vaccination, whether it actually is related to vaccination or not. For instance, there are famous examples of skeptics opposed to the antivaccine movement that claims that vaccines cause autism entering reports in which it was suggested that vaccines turned someone into the Incredible Hulk or Wonder Woman and had those reports accepted. True, someone from VAERS did contact the author who reported the Hulk reaction, but if he had insisted on leaving the report in the database it would still be there today. Meanwhile the VAERS database has been hopelessly distorted by vaccine litigation in which unscrupulous lawyers have encouraged parents to report autism as an adverse reaction to vaccines. Truly, it cannot be repeated often enough that VAERS is a perfect example of the computer programming maxim: Garbage in, garbage out (GIGO). VAERS was only ever intended to be an early warning system; it was never intended to give accurate estimates of prevalence or incidence that could be followed over time. After all, one can never be sure of the denominator to apply to its reports, and even honest, scrupulous scientists can be tripped up by this. That’s why VAERS, for all its usefulness as an early warning system, is the data source antivaccine activists pretenting to be researchers most love to dumpster dive.

So, with that background in mind (and, again, it’s hard to repeat that background too many times), let’s look at Goldman and Miller’s latest attempt to prove that vaccines are the root of all evil. The article, not surprisingly, has been published in a trash journal that has over the last couple of years seemingly become the go-to “peer-reviewed” repository of all things antivaccine, having published the infamous vaccine/miscarriage article and the vaccine/infant mortality article, Human & Experimental Toxicology. This time around, Goldman and Miller entitled their article Relative trends in hospitalizations and mortality among infants by the number of vaccine doses and age, based on the Vaccine Adverse Event Reporting System (VAERS), 1990–2010.

As per usual, let’s look at whether Miller and Goldman provide their conflicts of interest. This time around, they do a little better. Neil Z. Miller is listed as being affiliated with the ThinkTwice Vaccine Institute, and any readers familiar with that particular “institute” will know that it is a as wretched a hive of scum and antivaccine quackery as Generation Rescue and Age of Autism. On its website, you will find this particular study trumpeted, along with links to PDF of the study. The editors don’t do so well with Gary S. Goldman, however, who is described as a “computer scientist,” even though he is the president and founder of Medical Veritas, one of the most rabidly antivaccine groups out there. One also notes that the National Vaccine Information Center (NVIC) donated $2,500 for open access to the journal article (making it freely available to all researchers). The NVIC, as you recall, was founded by Barbara Loe Fisher and is one of the oldest and most influential anti-vaccine groups in the U.S., having recently teamed up with Joe Mercola to promote anti-vaccine views through ads on a JumboTron at Times Square.

Also, as per usual, I will point out that you can usually tell how good or bad a paper is going to be by its introduction. Although this paper’s introduction is not quite as bad as past Goldman and Miller “epics,” it does manage to cite a fair number of the “usual suspects” in the antivaccine literature, such as his own infant mortality rate paper and Gayle DeLong’s execrably bad paper correlating autism prevalence with vaccine uptake rates. Other references later in the paper include some Geier nonsense. Actually, looking over this paper, I was seriously tempted just to recycle what I had written before about the previous Goldman paper in which he tried to correlate vaccination rates with infant mortality, because it’s basically the same paper. Indeed, it even has the same graph, more or less, as you’ll see in a second. Basically, what Goldman and Miller tried to do was to correlate the number of vaccines received concurrently with hospitalization and mortality rates, and they came up with a graph like this:

Which very much looks like the graph from their previous paper:

Goldman and Miller also include a graph in which they examine hospitalizations as a function of age from 0 to 1 year old:

Whenever I see analyses like this, my first question is: Why does the author assume a linear relationship? There is no compelling scientific or biological reason to do so, which always makes me suspect that the author was just to lazy to do a proper statistical analysis and instead defaulted to what is easy, namely doing linear regressions. In fact, I find it rather amusing that Goldman used Prism to analyze his data. I use Prism a lotto analyze lab data and produce publication-quality graphs. It’s an excellent basic statistics program. However, the key word is “basic.” While much better than using the built-in tools for Microsoft Excel, as a statistics package, Prism is still relatively basic. True, it can handle a lot, including ANOVA and even Kaplan-Meier survival curves, but it isn’t really so hot for modeling non-linear relationships, and it’s really not so hot for trying to analyze the effects of potential confounders. When all you have is a hammer, I suppose, everything looks like a nail.

In any case, as usual, what Goldman and Miller look at are numbers without denominators. For instance, they don’t even try to normalize their data to time-dependent trends in the outcomes that they’re looking at. At the very least, they should have looked at their data by birth cohort in order to see whether the trends they claim to have found hold up when historical trends are controlled for. True, they’d be comparing to an outside dataset that might not be properly comparable, but at least such an analysis would give us a rough idea if the number of vaccines at a single sitting was actually correlated with a higher risk of hospitalization or mortality compared to age- and birth cohort-matched controls. As it is now, we can say nothing. Indeed, what Goldman and Miller have done is virtually meaningless because they’re looking at a set of patients selected because their parents or lawyers thought they had had a vaccine reaction, without verification of whether or not their reaction was actually due to vaccines. Add to that the usual rather arbitrary and questionable way in which Goldman and Miller count the number of vaccines in multivalent vaccines, and truly we have another GIGO epic.

I mentioned historical trends, and that is perhaps the worst failing of this paper. Basically, Goldman and Miller grouped together 20 years worth of data. For example, the child mortality rate in the U.S. has been steadily declining since 1960 and has declined significantly since 1990, as has the infant mortality rate. No attempt to control for this appears to have been made, nor has there been an attempt to control for the overall rates of hospitalization over time. Indeed, the authors themselves seem to admit (inadvertently, of course) that they haven’t controlled adequately for confounding variables:

A two-way ANOVA using the number of vaccine doses (2–8) and age, ranging from 0.1 to 0.9 years in 0.1 increments, was unproductive due to the too large an interaction between age and dose, particularly with those aged 0.6–0.9 years. When restricted to ages 0.1–0.5 years, the number of vaccine doses accounted for 85.3% of the total variation (F = 25.7, p < 0.001), the age factor was not significant at 1.4% (p = 0.64), and the residual was 13.3%.

Of course there is a huge interaction between the number of vaccines given at any one time and the age of a child. The vaccine schedule is set up so that children get different vaccines (or groups of vaccines) at different ages. Moreover that recommendation has changed over time. It is not at all surprising that that’s what Goldman and Miller found. Taking that consideration one step further, let’s go back to a particularly brain dead argument made by Vox Day, in which he correlated in VAERS the age at which the risk of mortality is the highest with the age of receiving multiple vaccines, namely a “death spike” at around three months. As I pointed out, the reason for that “spike” in the death rate between 2 and 4 months is because that’s age of peak incidence of sudden infant death syndrome (SIDS), an age that’s been known for decades and that hasn’t changed even with all the changes in the vaccine schedule over the last 30 years. As I also noted before, there are at least nine good studies showing no correlation between vaccination and SIDS. Interestingly, there is a hint of SIDS in Table 6, which lists mortality rates by age from 0.1 to 0.9 years of age, with a high death rate between 0.0 and 0.3 years of ages (0 and 3.6 months), which rapidly tails off after that. One also wonders about how the highest death rates occur before significant numbers of vaccines are even given. In any event, it all makes me wonder whether part of what produced Goldman and Miller’s result is the confusing of correlation between SIDS and the start of the vaccine schedule with causation. We can’t know because Goldman and Miller didn’t even try to control for some obvious potentially confounding variables, such as birth cohort.

None of this stops Goldman and Miller from concluding:

Studies have not been conducted to determine the safety (or efficacy) of administering multiple vaccine doses in a variety of combinations as recommended by CDC guidelines. Our findings show a positive correlation between the number of vaccine doses administered and the percentage of hospitalizations and deaths reported to VAERS. In addition, younger infants were significantly more likely than older infants to be hospitalized or die after receiving vaccines. Since vaccines are administered to millions of infants every year, it is imperative that health authorities have scientific data from synergistic toxicity studies on all combinations of vaccines that infants are likely to receive; universal vaccine recommendations must be supported by such studies.

Studies have not been conducted? Pretty much every epidemiological study done looks at adverse reactions examines vaccines administered according to the CDC-recommended schedule. This is just another antivaccine trope in which antivaccinationists try to claim that all vaccines must be specifically tested in every combination used in the manner that antivaccinationists think that they should be. As for the finding that younger infants are more likely to be hospitalized or die, well, that’s very likely because SIDS incidence peaks around 3 months, which means that by random chance alone infants in that age range are more likely to be hospitalized after vaccination.

So in the end what we are left with is yet another sad attempt by antivaccine activists who think themselves to be serious epidemiological researchers to demonize vaccines by dumpster diving the VAERS database in a risibly incompetent fashion. Same as it ever was.