Simpsonwood, thimerosal, and vaccines (II)
Robert F. Kennedy Jr. claims that the transcript of the CDC's Scientific Review of Vaccine Safety Datalink Information (hereafter, Simpsonwood) is definitive proof of a government/industry conspiracy to conceal the health effects of thimerosal in vaccines.
What follows is a discussion of the CDC/Vaccine Safety Datalink thimerosal study as presented at Simpsonwood retreat center in Norcross, Georgia in June of 2000. An expanded and updated analysis of the CDC/VSD data was published in 2003 in the journal Pediatrics.
The Vaccine Safety Datalink (VSD) was a collaboration between the CDC and major health maintenance organizations (HMO). The CDC acted as a clearinghouse for vaccine-related medical records from four major health maintenance organizations. The VSD received three major kinds of data from the HMOs: automated vaccination records, patient data (gender, birth date, enrollment), and outcome data (hospital discharge diagnoses, coded outpatient diagnoses).
The outcomes of interest to the thimerosal investigators were primarily outpatient diagnoses. Therefore study was restricted to Group Health (GH) and Northern California Kaiser Permanente (NCKP) because these were the only two VSD HMOs with adequate outpatient records.
The goal of the Simpsonwood studies was to see if thimerosal exposure was correlated with virtually any health outcome that has ever been linked to mercury poisoning in the medical literature. Thimerosal exposure was calculated by the number of thimerosal-containing vaccines (TCV) each patient received. The investigators calculated cumulative exposure to thimerosal at 1, 2, 3, and 6 months.
Phase I of the study used the patient records of about 110,000 children. The interval studied was 1992 to 1997. The original screening identified 200,000 infants who had received at least one thimerosal containing vaccine during their first year of life. However, in order to be considered for the analysis, a child had to have been born into the HMO and enrolled continuously for its entire first year of life. As anyone who has ever done a database analysis will tell you, this is not a trivial criterion to meet. Many HMOs have 20% annual turnover.
Subjects were also excluded if they suffered from any major congenital or perinatal disorder (fully 25% of the total potential sample) or prematurity (5%, analyzed separately). The investigators decided to exclude patients who received less than 2 polio vaccines during their first year of life on the grounds that a child who didn't even get those (non-TCV) shots probably just wasn't being taken to the doctor.
After all was said and done, a potential sample of approximately 200,000 had been winnowed to about 110,000.
The following thimerosal-containing vaccines (TCV) were available to the cohort: hepatitis B (HB), diphtheria/pertussis (DPT), acellular DPT, h influenzae B (HiB), and a DPT/HiB combo. The investigators were able to ascertain that all vaccines of interest contained thimerosal (except the aforementioned polio vaccines).
Unfortunately, the kids in the cohort received different vaccines in different combinations at different ages. If thimerosal has any effect, it is a probably a specific and subtle toxicological relationship. Any of the following variables might interact to modify the effects of thimerosal: the age of the patient at exposure, body weight, the specific vaccines administered, the spacing of the doses, etc. Unfortunately, the Simpsonwood experiment simply wasn't capable of disentangling such subtle relationships, if they existed.
The Simpsonwood study is often cited as evidence of a coverup of data supporting the autism/thimerosal hypothesis. In fact, no phase of the study ever found a statistically significant association between autism and vaccine exposure. (At least none of the data discussed at the retreat supported the autism hypothesis, I've heard unconfirmed allegations that the CDC has yet more embargoed analysis pointing to such a link.)
Anyway, here is the money shot from Simpsonwood:
From those risk analyses, excluding the dichotomized EPA, we have found statistically significant relationships between exposure and outcome for these different exposures and outcomes. First, for 2 months of age, an unspecified developmental delay which has its own ICD-9 code. Exposure at 3 months of age, tics. Exposure at 6 months of age, attention deficit disorder. Exposure at 1, 3, and 6 months of age, language and speech disorder, which are two separate ICD-9 codes. Exposure at 1, 3, and 6 months of age, the entire category of neurodevelopmental delays, which includes all of these, plus a number of other disorders.--Thomas Verstraeten, Simpsonwood transcript, p 40-41.
The transcript shows that the assembled experts agree that the data are interesting, and that the effects of thimerosal on neurological development deserve further study. However, they are also acutely aware of the severe limitations of the study, especially as a guide to policy making.
The Simpsonwood study could not have established a causal link between thimerosal and ill health. The study is sophisticated accounting of correlations. The original data show that infants who got more vaccines (or got their vaccines "on time") were also more likely to be diagnosed with certain neurological conditions later in life.
What might this mean? Maybe thimerosal in vaccines caused the problems. The most likely alternative explanation is that conscientious parents who vaccinate their kids on time are also more likely to seek treatment and obtain diagnoses for more subtle neurological symptoms like speech delay and ADHD.
Of course, there are confounds upon confounds. Care-seeking may be related to socioeconomic status, socioeconomic status to maternal mercury intake and breast feeding, and therefore to prenatal and postnatal mercury exposure, and so on. You can generate any number of alternative scenarios.
It is difficult to interpret the Simpsonwood data because the diagnoses of interest are so fuzzy. The study is trying to capture small differences in the relative risk for relatively rare conditions. As we learned from the recent JAMA obesity study, if you're studying a relatively rare phenomenon, like death at age forty (or infantile autism), small numerical differences between groups translate into huge increases in relative risk. Under these circumstances, it's especially important to be mindful of confounds that could account for small differences between the groups, e.g. local variation in diagnosis rates and diagnostic criteria in the case of relatively subtle and fluid diagnoses like "speech delay" or "attention deficit disorder."
As one of the pediatric neurologists at Simpsonwood pointed out , most of the kids in the study were too young to be diagnosed as speech delayed or attention deficit disordered with any degree of rigor. There's just too much overlap between the "normal" and the "pathological" curves at that age. Of course, once someone gets diagnosed with something, it's pretty rare for them to get undiagnosed, especially in the eyes of an administrative database. So, it's unclear how many kids with early diagnoses actually had a pathological problem to begin with.
It is very suspicious that diagnosis rates varied so dramatically between the HMOs for many of the outcomes of interest. The discrepancy was so great the authors had to present the data from each one separately. For example, speech and language delays were diagnosed much more frequently at Group Health (which has a specialized speech and language pathology center) than at Northern California Kaiser. I haven't crunched the numbers, but I bet that enrollment in Group Health posed a larger relative risk of a speech and language delay diagnosis than full vaccination.
The toxicology experts at Simpsonwood were skeptical because the data didn't seem to show a classical dose/response curve. Many also objected that the diagnoses observed weren't plausible results of mercury poisoning. They pointed to epidemiological studies of children exposed to much higher doses of methyl mercury prenatally and in infancy. Critics countered that there is very little toxicological data on ethyl mercury (thimerosal) and that it was unfair to compare acute vaccine-related exposure to daily exposure to chronically high levels of mercury.
The bottom line is that all the Simpsonwood data showed was a correlation between number of vaccinations and subsequent language problems and ADD. These correlations may have been caused by confounding variables (like health-seeking behavior) rather than by any aspect of the vaccination. Even if we suppose that vaccination caused the problem, we still have no particular reason to assume that thimerosal is the harmful substance. As the experts noted, any number of alternative substances or mechanisms might account for the results: aluminum, the vaccine antigens, autoimmune reactions in susceptible patients, etc.
Robert F. Kennedy Jr. argues that the CDC should have immediately banned all thimerosal vaccines on the basis of the Simpsonwood data. That would have been an absurd overreaction. A year before Simpsonwood, federal public health authorities had already committed to removing thimerosal from the vaccine supply, which is worthwhile goal. Thimerosal is, after all, a preventable source of mercury exposure.
However, in 1999 there were no thimerosal-free hepatitis B vaccines suitable for administration at birth, there was a t-free vaccine on the market, but it was only indicated for infants aged 2 months. The CDC policy makers felt it was unacceptable to delay HB vaccination from birth to 2 months because of an unsupported theoretical concern about mercury. Since thousands of US infants are exposed to HB at birth, a delay in vaccination would have resulted in significantly more babies getting hepatitis B.
The anti-vaccine activists are busily mining quotes from the Simpsonwood transcripts. Towards the end of the meeting, Dr. John Clements says that the CDC should never have conducted this research project because the results created a legal and PR nightmare for the agency. Other participants note that these results would be a bonanza for "plaintiffs' attorneys" and "junk scientists" seeking damages. These fears are not entirely unjustified.
Obviously, it's disheartening to hear someone say that PR should dictate a research agenda. However, it is important to interpret this remark in context. Clements was pointing out that this study was an incredibly crude method to investigate a complex, emotionally and legally fraught issue. He is correct that the Simpsonwood study could never have settled the thimerosal issue one way or the other, or even shed much light on the matter. The countervailing argument, which was also repeated frequently at the meeting, was that the CDC had better do this research, not because it was an urgent scientific priority, but because they had to be on the record as having looked into thimerosal (again, for PR reasons).
The critics are making a big deal out of the fact that the information presented at the meeting was "embargoed." If you read the transcript, you see that participants were only sworn to secrecy until ACIP meeting the following week.
Whether the CDC acted properly in this case depends on what its usual level of transparency is for preliminary studies like the one presented at Simpsonwood. Maybe someone who is familiar with the workings of the CDC would like to comment on how the agency usually handles preliminary data of this kind, and whether there were departures from standard operating procedure for the thimerosal data?
RFK's article gave me the impression that Phase II of the study was an ad hoc modification to the original study protocol. However, the original proposal always included a provision for a Phase II study if Phase I generated any interesting results. (Cf. link: VSD Thimerosal Study Background.) Judging by the Pediatrics article, Phase II was conducted as described in the original proposal. The only controversial feature of Phase II was the decision to use an HMO based in Massachusetts as a cross-check on the Phase I results. According to the transcript, the rationale was that Harvard Pilgrim was the only organization that promised to deliver a preliminary analysis in time for the upcoming ACIP policy meeting.
Nothing said at Simpsonwood suggests an attempt to whitewash or cover up anything. The study got some interesting results, which later failed to withstand verification at the third HMO. Vaccine experts said some nasty things about anti-vaccine activists. That was about it.
As yet, a link between thimerosal and autism is just one more unsubstantiated etiological hypothesis. I'm prepared to accept that small doses of injected mercury might have deleterious effects in susceptible infants. Maybe some forms of autism are caused by inborn errors of heavy metal metabolism or autoimmune defects and maybe these conditions are exacerbated by vaccination or vaccine preservatives. However, it is absurd to conclude from the evidence at hand that mercury poisoning is a major cause of autism. If the effect were that big, we'd see decreases in autism corresponding to decreased thimerosal exposure in the developed world. Maybe it's too soon to see results, but more likely the effect (if there is any) is so small as to be obscured by much larger trends.
Erratum: I accidentally published the wrong draft of this post. The earlier version included an incorrect statistic about the percentage of US infants exposed to hepatitis B at birth. I wrote, incorrectly that 10%-15% percent American infants were exposed. The actual number of infants exposed at birth in the US is about 19,000.
The current draft of this post also includes additional commentary.