©1997 by Elissa A. Laitin and Elise M. Pelletier. All rights
reserved
In February of 1976, the Centers for Disease Control (CDC) investigated
and confirmed that an influenza outbreak at Fort Dix had been caused by the
swine-type influenza A virus. Subsequently, the Department of Health, Education
and Welfare, as well as numerous medical experts, became concerned that a major
flu epidemic was imminent for the coming fall. Fear of influenza deaths in
numbers similar to the 1918 flu epidemic led to a recommendation that the
federal government vaccinate all Americans. When insurance companies refused to
provide coverage to the vaccine manufacturers, the government agreed to accept
liability for claims of adverse events (Neustadt). This obstacle having been
cleared, the National Influenza Immunization Program (NIIP) officially started
in October of 1976. The number of vaccinations given each week increased rapidly
from less than one million in early October to more than four million in the
later weeks of the month, and reached a peak of more than six million doses a
week by the middle of November 1976 (Marks). The NIIP was unique in the annals
of epidemiology: an organized surveillance effort was in place from the very
beginning, and over forty million people were vaccinated during the short time
the NIIP was in effect. However, on December 16, 1976 the NIIP was suspended
following reports from more than ten states of Guillain-Barré syndrome
(GBS) in vaccinees. By January of 1977, more than 500 cases of GBS had been
reported, with 25 deaths (Langmuir, 1979). Millions of dollars in lawsuits and
many years later, we present in this paper a summary of the epidemiologic
evidence of the possible causal association between influenza A/New Jersey/76
vaccine and GBS.
GBS is a relatively uncommon neurologic disorder characterized by an acute or subacute onset of polyneuritis. The predominant symptom is weakness, mainly of the extremities, although in more severe cases respiratory muscles may be involved (Hogg). The acute flaccid paralysis of GBS is typically symmetrical and may progress for periods as long as ten days. Spontaneous remission is common, and recovery usually occurs within three months of the time of peak illness (Safranek). Case series show that males are more frequently affected than females, there may be an increase with age, and the disorder is often associated with a preceding infection such as a non-specific upper respiratory or gastrointestinal infection (Hogg). The exact cause of GBS remains unknown. Historically, GBS was proposed to be due to a primary viral infection. However, the more recent prevailing hypothesis is that GBS represents an immunopathological reaction triggered by recent exposure to an exogenous agent. Although physicians can readily diagnose typical cases, criteria for delimiting the syndrome’s diagnostic boundaries have been controversial (Schonberger, 1981). In fact, diagnostic criteria would prove to be a common criticism of the published studies, and the rarity of the illness led to arguments over the appropriate background rate to use in epidemiologic reports.
Before vaccine administration began, a nationwide surveillance system was established to evaluate illnesses that were temporally associated with influenza vaccination. No previous mass immunization campaign had included a similar prospective surveillance network. This passive reporting system was centrally coordinated by the CDC, and all state and territorial health departments were required to participate. A registration consent form had to be signed by all vaccinees. This form allowed health departments to verify the date, type, manufacturer and lot number of vaccines involved in reaction reports. A report form covering basic epidemiologic information was distributed, and any illness serious enough to cause hospitalization had to be telephoned in to the CDC. In addition, some states sent letters to non-neurologist practitioners, surveyed hospitals and conducted a repeat survey of neurologists after the middle of January 1977 (Schonberger, 1979).
Using data taken from the August 1977 public release of the nationally collected surveillance data, Alexander Langmuir presented a preliminary report in 1979 on the possible relationship between the swine flu vaccine and GBS. Concurrently with the vaccine-associated cases of GBS, more than 500 other cases among unvaccinated persons in the population were collected through a network of collaborating neurologists and practicing physicians, organized on an emergency basis throughout all fifty states. Langmuir calculated the average incidence rate in unvaccinated cases for an eleven-week period (October 9 to December 18). This rate, 0.185 cases per million person-weeks, was used to determine the expected weekly numbers.
Unlike with the peak seen in the vaccinated population, the relative constancy of weekly incidence rates in the unvaccinated group throughout the eleven week period conformed to the general impression of most neurologists and epidemiologists that GBS was an endemic disease with little seasonal fluctuation. The incidence of unvaccinated cases declined after December 18, which Langmuir believed was unlikely to be due to under-reporting of cases. Instead, it was during December that the awareness of a possible association between the vaccine and GBS led to the end of the NIIP, and it would have seemed reasonable to expect an even prompter diagnosis than during the period before the relationship was suspected.
Based on the weekly numbers of vaccinations, a comparison of observed with expected cases showed that the relative risk of acquiring GBS during the six weeks after vaccination was about ten times the endemic expectation. No marked differences were observed in incidence by age or geographical distribution. Langmuir was careful to state that much of the available epidemiologic information about GBS was based on data from hospital records and therefore was difficult to interpret because of the lack of accurate estimates of the populations from which the cases were drawn. However, previous studies had shown the incidence rate of GBS to be about 1 case per 100,000 population per year, similar to Langmuir’s calculated baseline incidence.
Very little difference in clinical features was found between the unvaccinated and vaccinated cases. Visits to a neurologist, lower neuron signs and respiratory impairment were similar, and the case fatality rates were almost identical. However, the rate of acute illness in the month preceding GBS onset was much higher among the unvaccinated cases than the vaccinated cases (59% versus 30%, respectively). Langmuir concluded from these data that the swine influenza vaccine distributed in the U.S. during the fall of 1976 contained a "trigger element" which resulted in the development of clinically recognized GBS in 1 in 100,000 recipients of the vaccine.
Also in 1979, Schonberger and his collaborators at the CDC presented an additional analysis of the national surveillance data of cases with an onset between October 1, 1976 and January 31, 1977. In order to be accepted as a case by the CDC, GBS cases had to have been diagnosed by a physician and to have objective evidence of muscle involvement, and suspected cases reported directly to the CDC had to be validated by the state health department. State epidemiologists were sent a brief survey in April 1977 to help assess possible differences in case ascertainment after December 18 due to a sharp decrease nationally in reported cases after that week. A total of 1,098 cases were reported to the CDC during the four months under investigation.
Estimates of the civilian population by state and age (as of July 1, 1976) were obtained from the Census Bureau, and the number of vaccinations was provided by the CDC-NIIP surveillance center. For each state, both weekly and monthly reports of total vaccines administered by age group and vaccine type were gathered. Only data for cases reported on or before the end of January were analyzed, as this was when intensified surveillance for GBS ended. There were four different manufacturers of both monovalent and bivalent vaccine, and an estimated proportion of each vaccine type given in each state was calculated by an inventory of the amounts distributed and remaining at the end of the campaign.
Vaccinations increased from the beginning of the program in October, reaching a peak in mid-November, and declining thereafter (Appendix A, Figure 1A). The incidence of GBS in vaccinated adults (over age 17) rose rapidly through late October and November to reach a peak during the week of December 18. As vaccinations abruptly ended, there was a sudden drop among vaccinated individuals, from more than 70 cases to 22 (Appendix A, Figure 1B). The incidence in unvaccinated persons remained relatively constant, ranging from 28 to 45 cases each week from early October to the middle of December, then declined to half its previous level after the December 18 moratorium (Appendix A, Figure 1C). Using the population estimates, the expected attack rate for adults over 17 was calculated to be 0.22 cases per million person-weeks. It was found that 71% of vaccinated cases with known intervals became ill within four weeks after vaccination, 52% in the second and third weeks after vaccination (Appendix B). The relative risk in the adult population for the six weeks following vaccination was 7.6, although a statistically significant association was seen as long as ten weeks after vaccination. For these ten weeks, the attributable risk was just under 1 case per 100,000 vaccinations.
Schonberger, et al. indicated that one of the weaknesses of the study was the poorly defined clinical boundaries of GBS from the national surveillance data and the lack of medical record review for validation. The authors concluded, however, that their results of markedly increased attack rates in vaccine recipients showed strong support for an etiologic link. The average incidence rate of the unvaccinated (0.79 cases per million people per month) fell within the range of previously reported rates in U.S. cities and other countries (0.5 to 1.6 cases per million per month). In addition, the nonrandom distribution of intervals between vaccine and GBS onset and the markedly lower proportion of acute illness preceding GBS onset in vaccinated cases provided further evidence of an etiologic link. Normally, over 50% of GBS cases are preceded by a respiratory or gastrointestinal illness; in this study, only 33% of vaccinated cases reported having an acute illness within four weeks of onset.
In the following years, lawsuits stemming from the NIIP led to a 1981 court order that the CDC release the data used by Langmuir and Schonberger, et al. in the above studies. A computerized summary was submitted for each reported case. At this time, Langmuir and his colleagues formed a panel to re-evaluate the data. Their intent was to answer the question of how long after vaccination any possible causal effect had continued as well as to re-examine the overall association, due to criticism of methodological issues like the clinical classification of cases.
The panel first examined information on cases for completeness of data. Of the 1,098 cases originally used by Schonberger, et al., only 154 were excluded due to insufficient data or being under age 18. The cases were reviewed for extent of motor involvement and clinical characteristics, then grouped according to severity. Information regarding the vaccination status was deliberately omitted in the process of classifying the cases. The panel immediately found problems with misinterpretations by those who reported the data to the CDC, as well as errors of recording or coding and failures to complete forms. The panel initially considered the adoption of more stringent diagnostic criteria, which would have excluded more than 25% of cases otherwise having clinical features believed to be characteristic. However, the panel only had access to information released by the CDC and from public records. The basic files remained impounded at the CDC, so the investigators were unable to verify records, check inconsistencies or investigate apparent typographical errors or possible miscodings. Thus, the panel chose to accept the original classifications of all 1,098 cases in recognition of the limited data available and the fact that more information was probably available to the CDC reviewers than had been released by the court order.
The panel had access to two distinct data sources which contained the numbers of vaccinations performed. The first was the weekly and monthly reports collected routinely by the NIIP, comprised of the total counts accumulated in the NIIP centers in each state. This was the data used by Schonberger, et al. and Langmuir in 1979. However, some lag or under-reporting was unavoidable in early stages of the program, while there was compensatory over-reporting during the six months after the program ended, as delayed reports came in. Because no adjustments had been made to account for these problems, the panel chose to use data from the Health Interview Survey (HIS) conducted by the National Center for Health Statistics. The HIS data was presumably relatively free of the reporting lags because interviewers were collecting current information each week.
Establishing the baseline incidence rate proved to be difficult. The authors were wary of previous reports, most of which consisted of a collection of cases observed by one or a group of neurologists in which the population from which the cases had been drawn was unknown. The few population-based studies to date had reported rates which were not corrected for adults 18 years of age and older or classified by severity of motor involvement. The 1979 studies by Schonberger, et al. and Langmuir cited these reports because at the time they were considered the best available. However, the panel had greater confidence in the diagnostic criteria used by the Mayo Clinic and cited unpublished data from Kurland and Beghi in Olmsted County, Minnesota as one possible source for baseline rates (the Olmsted County study in its published format will be discussed below). Another population-based study which the panel felt was applicable to their evaluation was that reported by Breman and Hayner out of Michigan. The Michigan study was published simultaneously with this panel’s findings and will also be discussed below.
Langmuir’s panel graphed vaccinated cases by numbers of days since vaccination. The incidence curve for cases with extensive paralysis was skewed to the right. This pattern, characteristic of point source epidemics, was identical to that seen in the previous studies, including Schonberger, et al. (see Appendix B). The lognormal curve suggested a causal relationship between disease and vaccine. Cases with limited motor involvement showed no such pattern, implying that this group included a substantial proportion of cases which were unrelated to the vaccine. The panel decided to use two separate baseline rates of GBS to compare observed with expected cases. Data from the Michigan and Minnesota studies were combined: 0.275 cases per million person-weeks was used as a "high" background estimate. A "low" estimate of 0.14 cases per million person-weeks was calculated from the mean rate in unvaccinated cases. Looking at the first six weeks following vaccination, the relative risk for severe cases was 39.6 using the higher baseline estimate and 7.75 when using the lower baseline estimate. Using the lower baseline rate also showed that the risk extended to eight weeks after vaccination, slightly less than Schonberger, et al.’s estimate of an elevated risk up to ten weeks.
Soon after the publication of this study, Nathan Mantel released strong criticisms of both Schonberger, et al.’s original study and the panel’s conclusions. Mantel felt that the analyses of Schonberger, et al. were done at the expense of bringing out effects somewhat late after vaccination. Although the authors found that the period of increased risk lasted for approximately nine or ten weeks after vaccination, they did not look for later increases in risk in any effective manner. The study only examined GBS cases with onsets up to January 31, 1977. But for thirteen-week delays, for example, the vaccination would have had to have been in the month of October 1976, a time during which much of the vaccination program was still at a low level. Also, a case with an onset in mid-January was unlikely to be diagnosed and reported until February. Therefore, because data for establishing long delay effects were limited, Mantel felt that some people may have been unfairly denied compensation if their onset of GBS occurred after ten weeks post-vaccination.
Mantel also criticized Langmuir and his panel’s study, as they had trouble establishing reliable baseline rates for GBS. Both estimates were subject to scrutiny, as the "high" Michigan-Minnesota rate was based on cases gleaned from active and thorough surveillance and so was undoubtedly too high to be reasonably compared with the rate found in vaccinated cases. The "low" estimate from the unvaccinated cases was based on data which was collected at the same time and by the same surveillance program, and which was approximately half as large as the high estimate. Although under-reporting of unvaccinated cases was certainly probable as Mantel states, the finding of large relative risks using both baseline rates seems to lend more credibility to the panel’s findings.
As noted above, one of the baseline rates cited by the panel came as a result of a study by Breman, et al. in Michigan (1984). As part of the national surveillance program of 1976 and 1977, a system was established in Michigan to detect GBS cases with an onset from July 1, 1976 to June 30, 1977. The purpose of the investigation was to determine GBS frequency over the one-year period, to describe its epidemiology in terms of demographic variables and seasonality, and to determine any association with the A/NJ vaccine. Cases had to be diagnosed by a physician, meet a strict list of criteria, and receive confirmation from the primary care physician and/or consulting neurologist. Classifiers were not blinded, but were told not to consider vaccine status when confirming cases. Between December 20, 1976 and June 30, 1977, over 300 possible cases were screened, but over half were excluded because they did not meet diagnostic criteria, leaving a total of 132 confirmed cases for study.
The incidence of GBS in unvaccinated individuals was found to be about twice the rate for the nation. This finding points to the fact that the Michigan case reporting system was very aggressive and probably detected most, if not all, cases in the state. Using this baseline rate, Breman, et al. found that the incidence rates of GBS among vaccinees in the weeks before vaccination and in the seven or more weeks after vaccination were much lower than during the first six weeks post-vaccination. The risk attributable to the vaccine for adults 18 or older with an onset within six weeks of vaccination was 11.7 cases per million vaccinees. No differences in attack rates were found by sex, season or vaccine manufacturer. Breman, et al. concluded that immunization clearly led to an increased risk of GBS, but the risk period was only for six weeks post-vaccination, similar to a number of the earlier studies. The authors did not posit a biological mechanism for this occurrence, but did note that the unvaccinated population had a slightly increased incidence with age. They thought this might indicate that a prior sensitization was needed to trigger an autoimmune response, but did not extend any theories as to how the vaccine might be responsible.
In addition to Breman’s Michigan study, Marks, et al. (1980) performed a study in Ohio using statewide instead of national data. To ascertain cases, neurologists were called by telephone and asked to check their records for all cases of GBS with an onset since October 1, 1976, and to continue to notify the health department of cases with an onset before January 31, 1977. The authors defined a case as a person with a diagnosis of GBS and with physical evidence of bilateral, but not necessarily symmetrical, lower-motor neuron weakness with acute onset.
Marks, et al. found 54 reported cases of GBS, with 32 giving a history of vaccination against influenza. The cases in vaccine recipients tended to cluster in the two months from late October to late December, while no such clustering was seen in unvaccinated cases. In the vaccinated cases, the length of time between receiving the vaccine and the onset of GBS showed a pronounced peak two to three weeks after vaccination, a finding in line with previous studies. The authors calculated an attributable risk of 80.5%. As in the Michigan study, this Ohio investigation used both active surveillance to find cases and clear diagnostic criteria. Although the authors failed to report the response rate of the contacted health professionals, this study supported the CDC’s initial findings.
Soon after these studies were performed, Beghi, et al. (with Kurland, 1985) published their findings of the epidemiologic and clinical features of GBS in Olmsted County, Minnesota from 1935 to 1980. The authors felt that the validity of prior studies was questionable due to the general confusion over diagnostic criteria and the lack of rigorous case ascertainment. The aim of their study, therefore, was to present an updated evaluation of GBS using more rigid diagnostic criteria, to assess incidences and trends of the vaccine, and to estimate the expected number of GBS cases at time of vaccination. The health care of Olmsted County was documented in a centralized index at the Mayo Clinic. Population rates were computed using time-weighted averages for each of the census years. Complete ascertainment of suspected and diagnosed cases was accomplished by having neurologists review the medical records of all residents under strict criteria.
Beghi, et al. determined that the mean annual incidence was 1.7 per 100,000 population. An acute illness prior to onset was present in 65% of cases. Age-adjusted rates were higher for males (2.3) than for females (1.2), and the rate increased with age, from 0.81 for under 18 years of age to 3.2 for those aged 60 years and older. The overall incidence rate for individuals aged 18 and older was 2.1 per 100,000 population, or 0.4 per million person-weeks. This rate was almost twice that reported by Schonberger, et al. from the national data. Because of this difference in rates and the fact that ascertainment and reporting of unvaccinated cases may have been less intense and complete than among vaccinated, the authors speculated there might not be a true association between the vaccine and GBS. They cited the CDC’s inadequate diagnostic criteria as a problem as well as the lack of any finding of an increased occurrence of GBS during earlier vaccine programs. They also did not uncover any cases of GBS in their county during the swine flu vaccine program. Using the Olmsted County baseline rate, they did show that nationally, there were more cases than expected during the six weeks after vaccination, but did not feel that any argument had been made for a causative association.
Ellsworth Alvord was quick to criticize Beghi, et al.’s assertions. They had compared rates in vaccinated cases taken from national surveillance data in 1976-1977 to background rates taken from a Minnesota county with high quality medical records during a 45-year period. Alvord felt that this comparison was inappropriate, and likened it to comparing "apples and oranges." Beghi, et al. did not find any cases of GBS in their county during the swine flu vaccine program, but Alvord pointed out that they also did not state in their report that there were only approximately 60,000 people in the county. Even if everyone had been vaccinated, the incidence of GBS due to vaccination had been previously calculated at only 1 per 100,000 million people and therefore, only 0.6 excess cases would have been found. The authors should not have been surprised at their lack of reported cases.
Beghi and Kurland immediately replied, stating their comparison was not a case of apples and oranges, but of how many "rotten apples" there were in the cases accepted by the CDC. They agreed that because only 39,000 people in the county received the vaccine, it was not surprising that there were no reported GBS cases during that time. But they argued that Olmsted County rates of rare and uncommon diseases were often higher than that for other areas that did not have such high quality records-linkage, thereby suggesting that they should have found at least one case. This argument, however, certainly offered no additional evidence to refute Alvord’s criticism.
Several of the studies examined possible differences between the four different manufacturers, the 47 lots of vaccine and the type of vaccine itself. Langmuir, Schonberger, Marks and Breman’s studies concluded there were no differences in the outcome whether the vaccine was monovalent or bivalent or whole versus split-virus. No single manufacturer’s vaccine had a significantly higher rate of GBS when compared with the other three manufacturers combined, and no study could show any explainable difference in lot-specific attack rates. Although two studies found that the rate in recipients of one lot was somewhat higher than that for recipients of the other lots of vaccine, it was high in only one of the many states where that lot had been distributed. This finding was most likely a chance association related to random variation.
One methodological criticism of a number of the studies was that the CDC did not establish clear diagnostic criteria for GBS. Records were abstracted by health workers with varying qualifications, no review of clinical records and no systematic follow-up of cases occurred, and no standardized neurologic assessment of the medical records was ever performed. Because there was no method of validating the cases, it was not possible to determine if any bias existed due to differential or non-differential misclassification. As several of the authors used their own diagnostic criteria in their investigations, comparisons of studies is more complicated. The Ohio and Michigan studies were performed with very clear diagnostic criteria, and the fact that their outcomes supported the earlier studies helps to discount fears that this flaw would cause a major bias in the findings.
When trying to establish that a positive association is also a causal relationship, it is important to examine clinical differences between exposed and unexposed cases. Both Schonberger, et al. and Langmuir (1979) found one statistically significant difference between vaccinated and unvaccinated cases, which was the higher proportion in the latter group who had experienced an acute illness during the four weeks preceding onset of GBS. Schonberger, et al. found that prior acute illness had occurred in 61.8% of unvaccinated individuals versus 32.8% of those who were vaccinated, while Langmuir’s data was similar, with rates of 59.3% versus 30.2%, respectively. This difference between vaccinated and unvaccinated cases indicates that the cause of GBS in those vaccinated was different than what was normally seen, and thus a new influence was most likely responsible for the increased incidence of GBS.
Establishing a plausible biological mechanism is also important when assessing causation. But for GBS, the evidence for specific etiologic factors is scanty and conflicting. The syndrome is thought to be an autoimmune reaction, but there is not enough known about the disease to be able to pinpoint the exact cause. Experimental allergic neuritis has been induced by intradermal injections in rabbits, and in humans there is an association between allergic encephalomyelitis and polyneuritis and killed rabies vaccine, with similar latent periods between vaccination and GBS onset as reported here (Langmuir, others). However, no clear connection has been made between these experimental results and the swine flu vaccine-GBS relationship. In the epidemic of GBS, the "trigger element" could have been a component of the Fort Dix strain of swine influenza virus used for making the vaccine, possibly the residual myelin protein of chick embryo origin which might have been retained in the vaccine though all its stages of manufacture and purification. Previous and future non-swine flu vaccine campaigns did not lead to increased cases of GBS, giving credence to this hypothesis. It is not clear whether this question will ever be answered.
Despite the lack of a definitive biological explanation for the association between the swine flu vaccine and GBS, there is strong evidence for a causal relationship. In the multiple studies performed immediately following the discontinuation of the NIIP as well as in those done almost a decade later, and using at least three distinct sources of data, there were consistent findings of increased numbers of GBS cases during the six weeks following vaccination. A peak onset at two to three weeks post-vaccination was clearly shown. Study after study reported an excess risk of GBS of approximately 1 in 100,000 vaccinees. An additional finding consistent in most of the studies was the much lower rate of prior acute illness in vaccinated cases. Although the 1976 NIIP was short-lived, it was a unique time in our nation’s public health history. It included several elements that allowed for the discovery and analysis of the association between the swine flu vaccine and GBS: (1) massive amounts of vaccine were administered in a short time, with a great deal of attendant publicity surrounding the deaths and other negative reactions; (2) written informed consent was required for administration of the vaccine in the public sector, which allowed for accurate reporting of total doses administered; and (3) the passive surveillance system succeeded as an early warning system, such that a large-scale active investigation began when only seven GBS cases had been reported nationwide. Taken together, the most important results from the investigations and the NIIP may be not only that more research is needed to determine the etiology of autoimmune reactions, thereby enabling a stronger argument for a causal relationship to be made, but also that a well-organized surveillance effort is an invaluable tool in the advancement of epidemiology.
Return to the DDIL Homepage
This
document has been provided through the Drugs and Devices Information Line,
http://www.hsph.harvard.edu/Organizations/DDIL/ddil.html