Controlled Epidemiological Studies

Guinea-Bissau. Kristensen and colleagues (2000) studied the relationship between vaccination and childhood survival in a population of 8,752 children born to mothers participating in a longitudinal mortality study in Guinea-Bissau. Recommended childhood vaccines in Guinea-Bissau include BCG, OPV, DTP, and measles. Vaccination status was determined by inspection of immunization cards kept by the children's parents. Children were excluded if the card could not be examined, which was the case for more than a third of children.

Vaccine exposure for BCG, OPV, and DTP was assessed during a first visit when children were 0–6 months of age. Mortality was assessed at a subsequent visit approximately 6 months later. For surviving children, vaccination status, including measles vaccine exposure as well as that of BCG, OPV, and DTP, was updated. Mortality was assessed again at a third visit, approximately 6 months after the second visit.

A Cox proportional hazards model was used to calculate the mortality ratio for vaccinated and unvaccinated children. The overall mortality for any vaccine was nonsignificant (RR=0.74, 95% CI 0.53–1.03). Receipt of BCG vaccine was associated with lower mortality (adjusted RR=0.55, 95% CI 0.36–0.85), as was measles vaccine (adjusted RR=0.48, 95% CI 0.27–0.87). The mortality ratio for one dose of DTP vaccine versus none was 1.84 (95% CI 1.10–3.10), but the ratio for two to three doses was not significantly elevated (RR=1.38, 95% CI 0.73–2.61). The pattern was similar for OPV, with an elevated mortality ratio for one dose (RR=1.81, 95% CI 1.07–3.05), and nonsignificant ratio for two to three doses (RR=1.39, 95% CI 0.73–2.64).

The authors conclude that receipt of BCG and measles vaccines may have a protective effect against mortality, while receipt of a single dose of DTP and polio vaccines may carry a higher mortality risk compared with receiving no vaccinations. The results also suggest that DTP vaccine may negate the positive effects associated with BCG vaccine.

However, the interpretation of these findings warrants caution. The vaccination status of some children was unclear and more than a third of the children did not have records available. Many children may have been underimmunized, contributing to the increased mortality rates and reflecting limited access to health care. Vaccinated children were also more likely to receive health care than unvaccinated children, which may mean that getting vaccinated is associated with access to or use of other interventions that improve survival. Mothers of children vaccinated with DTP were younger than mothers of children vaccinated with BCG or measles vaccine, which means maternal age may have contributed to infant mortality risk. The adjustment procedure for potential confounders was also unclear. For the United States, the findings regarding BCG vaccine are not relevant since the vaccine is not routinely used in this country.

United States-Boston. In a case-control study, Burstein and Fleischer (1994) examined the relationship between vaccination and the risk of occult bacteremia. Cases and controls were patients treated in the emergency department at Children's Hospital in Boston between November 1987 and December 1990. Cases were 54 children age 3 to 36 months who participated in a multicenter antibiotic study. Pathogens isolated from these children included S. pneumoniae, E. coli, S. aureus, H. influenzae, or Salmonella spp. The 108 controls were matched to cases according to age. Each case had two controls. One control group included febrile nonbacteremic children. The other group included nonfebrile children who were treated for symptoms not related to infectious diseases. Vaccination history, including DTwP, was obtained from medical records.

The authors found no significant difference between cases and controls for time since last vaccination of any type, or for time since last DTwP vaccination. Limitations of this study included weak statistical power. It was also unclear which vaccines, other than DTwP, the children received.

United States-Tennessee. Griffin and others (1992) examined the association between DTwP immunization and the risk of invasive bacterial infection. The incidence of invasive bacterial diseases (H. influenzae, N meningitidis, Streptococcus pneumoniae, group B Streptococcus, or Listeria monocytogenes) was measured in a cohort of 64,591 children who received at least one dose of DTwP vaccine through any of the four largest Tennessee county health clinics from 1986 to 1987. Based on surveillance data, 158 children diagnosed with invasive bacterial infection after receiving DTwP vaccine were identified in this cohort. Using a Poisson regression model and controlling for age, the relative risk for infections during the early post-immunization periods (0–7, 8–14, 15–28 days) compared with the later period (29 or more days) was nonsignificant. The authors concluded that there was no increase in the risk for invasive bacterial infection following receipt of DTwP vaccine, especially during the early post-immunization period. Interpretation of the study is limited by the lack of an unvaccinated comparison group. In addition, the analysis was limited to cases of serious culture-confirmed infections.

United States-Kaiser Permanente Northern California. In a case-control study, Black and colleagues (1991) examined the relationship between vaccination and the risk of heterologous invasive bacterial disease. Cases and controls were identified from member records in the Kaiser Permanente Medical Care Program of Northern California. As cases, 223 children between 1 month and 2 years of age who were diagnosed with invasive bacterial disease (Pneumococcus, H. influenzae, E. coli, and Meningococcus) between 1986 and 1988. Invasive bacterial disease status was identified from a computerized microbiology laboratory database. The 446 controls were matched according to age, sex, zip code, and length of plan membership. For cases, all vaccines received within three months prior to disease onset were identified through medical chart review. For matched controls, the date of diagnosis for the corresponding case was the reference date used to obtain vaccination histories. Children had received one or more of the following vaccines: DTP, OPV, and MMR.

A conditional logistic regression model was used to estimate the effect of recent immunization on disease; odds ratios were calculated from the regression results for each vaccine. A separate analysis controlled for the effect of well care visits and day care attendance (information available for 72 percent of the subjects). Odds ratios were calculated for separate time intervals from date of vaccine receipt to date of disease diagnosis: 0–7 days, 8–30 days, 31–60 days, and 61–90 days. Receipt of individual vaccines was associated with a lower risk of disease in all time intervals, with significant effects for DTP at any interval after 7 days and for OPV at 8–30 days and 31–60 days. After adjustment for day-care attendance and well-care visits, however, no individual vaccine had a significant effect on risk of disease. But there was a significant protective effect in the adjusted analysis from the receipt of any vaccine within 30 days (OR=0.26, 95% CI 0.09–0.76) or 90 days (OR=0.31, 95% CI 0.13–0.73).

The authors conclude that vaccines do not increase the risk for invasive bacterial disease and that they may provide a protective effect against disease, especially within 3 months after vaccination. However, children who received well care visits were also less likely to develop invasive bacterial disease than those who did not receive them. A health care effect may account for the observed protection of vaccines against heterologous invasive bacterial disease.

United States-Alaska. In a two-part study, Davidson and colleagues (1991) examined risk of disease following receipt of DTwP vaccine among Alaskan Native children. They first conducted a case-control study to examine the risk of invasive bacterial disease. The 186 cases were children 2 to 24 months old who received at least one DTwP vaccine and were identified through surveillance reports for H. influenzae type b and S. pneumoniae diseases. There were 186 controls matched according to age, sex, residence, and number of DTwP vaccines received. The time interval between last DTwP vaccination and date of disease diagnosis (the reference date for controls) was obtained. There were no significant differences in DTwP vaccine intervals for Hib disease cases. For S. pneumoniae disease, significantly more cases had been immunized 31–60 days earlier (OR=3.3, 95% CI 1.1–10.0), but differences were not significant for shorter or longer intervals. The authors conclude that there was no clear association between the timing of DTwP vaccination and risk of invasive bacterial disease. The authors note that a possible explanation for the lack of difference observed in the study is overmatching. Overmatching is where cases and controls closely resemble each other on factors related to the exposure of interest. In this study, matching based on the number of DTP vaccines received may have resulted in the lack of difference in the timing of DTwP immunization between cases and controls. However, the authors observed that matching according to the number of DTP vaccines was necessary and reduced potential confounders such as those related to health status.

Subjects in the second part of the study included 104 cases and controls from the earlier part who had complete medical records available. Cases and controls were combined to compare the occurrence of any illnesses within 30 days before and after receipt of DTwP vaccine. Comparisons were made for the occurrence of any illness, any infectious disease, otitis media, other respiratory infections, temperature greater than 38°C, or hospitalization. There was a higher incidence of any illness during the 30 days prior to DTP vaccination than in the 30 days following vaccination (53% versus 43%, p=.004). There were no significant differences between the pre- and post-vaccination periods for the other indicators of illness. The authors again conclude that DTwP vaccination does not increase the risk of other illnesses. The authors note that the higher frequency of disease in the pre-DTP group compared to the post-DTP group may have resulted from a “well-child effect,” whereby immunization of children with illnesses was postponed until they were well.

Randomized Controlled Trial

Germany. Otto and colleagues (2000) examined differences between vaccinated and unvaccinated children in the risk of morbidity associated with infectious diseases. A total of 662 children, born between January 1995 and December 1996, were randomized to receive their first vaccine (against diphtheria, pertussis, tetanus, Haemophilus influenzae type B, and poliomyelitis) at either 60 days or 90 days after birth.⁵ Children were observed during the third month of life, beginning at the sixth day after vaccine receipt. Mothers kept a daily journal and recorded any occurrence of symptoms. Morbidity was assessed in terms of the incidence of coughing, signs of rhinitis, restlessness, vomiting, rash, pain, poor food or fluid intake, fever, and respiratory embarrassment.

A total of 166 children were excluded from the analysis, mostly because of missed vaccination. The authors observed a significant (p<0.01) increase in vomiting, cough, rhinitis, restlessness, rash, and pain in the unvaccinated group compared with the vaccinated group. Hospitalization was more frequent among unvaccinated children (n=4) than vaccinated children (n=1). The authors concluded that children who received vaccinations during the third month of life did not demonstrate an increased risk of infectious-disease symptoms and may experience some protective effect from vaccination. Study limitations included observer bias in that the mothers, who were responsible for monitoring morbidity, were not blinded as to vaccination status. The high exclusion and dropout rate (39% versus 11% in the unvaccinated group), especially in the vaccinated group, may also effect interpretation of the study results.

Other Studies

The committee reviewed additional articles that reported adverse events after receipt of multiple immunizations. These studies helped inform the committee's assessment of risk for heterologous infection but did not contribute to the causality argument. Most of these articles reported on randomized controlled trials that primarily investigated the safety, efficacy, and/or immunogenicity of various vaccines. They did not specifically examine the incidence of heterologous infections in children post-immunization, compare the incidence of such infections between different exposure groups, or report statistical analyses from which to make inferences or extrapolations related to heterologous infections. The articles reviewed are briefly summarized below.

Shinefield and colleagues (1999) examined the safety and immunogenicity in infants of a heptavalent pneumococcal CRM1₉₇ conjugate vaccine administered concurrently with DTwP, Hib, and OPV vaccines. Hepatitis B vaccine was administered concurrently or at least two weeks earlier or later. Children received DTaP vaccine at a later stage of the study. The authors reported on adverse events observed following immunization. In the study, 302 infants age 2 months were randomized to receive the pneumococcal vaccine or meningococcal Group C conjugate vaccine. During the study, 12 emergency room visits occurred within 30 days of any vaccine dose. These visits were for croup, otitis, febrile illness, and urinary tract infection, but none were considered vaccine-related. Following the primary doses of vaccines, eight children were hospitalized, two within 30 days after vaccination. Following the booster doses, there were four emergency department cases (of viral illness, otitis media [two cases], and burn) and four hospitalizations (pneumonia, otitis media, elective surgery, asthma, and cough). The authors did not believe that these events were vaccine-related.

Olin and others (1997) compared the efficacy of three types of DTaP vaccine to the DTwP vaccine used in the United Kingdom. A sample of close to 83,000 infants age 2–3 months were randomized to different DTP vaccine exposures and also given Hib and IPV vaccines. The authors reported on adverse events following vaccine receipt. Those that may have involved heterologous infections included two deaths from pneumonia within 4–7 days of a trial vaccine dose, 20 cases of invasive bacterial infections, and one case of suspected encephalitis

Afari and colleagues (1996) examined the immunogenicity and reactogenicity of two types of DTaP vaccine (a freeze-dried, heat-stable product and a liquid product) and DTwP vaccine. Of the 403 infants who were studied, 136 were randomized to receive the freeze-dried DTaP, 130 to receive liquid DTaP product, and 137 to receive DTwP. The authors reported that three children who received the freeze-dried vaccine died of measles or malaria, two children who received liquid DTaP died of malaria or diarrheal disease, and two children who received DTwP died of multiple boils or malaria. Differences in mortality rates between either DTaP group and the DTwP group were not statistically significant.

Simondon and colleagues (1996) compared the safety and immunogenicity of DTaP and DTwP vaccines in a randomized clinical trial involving 286 Senegalese infants. Children also received BCG and IPV during the study. Six deaths occurred within 2 months after vaccination. On the basis of verbal autopsies, the four deaths in the DTwP group were attributed to diarrhea, malaria, and pneumonia. The two deaths in the DTaP group were attributed to meningitis and diarrhea. The authors suggested caution in drawing conclusions regarding the number and causes of deaths in the study. The infant mortality rate in the study region is high, and verbal autopsies are an imprecise means of determining cause of death.

Riordan and colleagues (1995) reported two cases of bacterial meningitis after receipt of MMR vaccine. Fever and rash occurred in two children, age 12 months and 13 months, within 4 days after MMR vaccine, and were initially attributed to the vaccine. After diagnostic tests, both children were found to have a high level of C-reactive protein and were diagnosed with a bacterial infection. Meningococcal infection was confirmed in one case and suspected in the other. The authors noted that the median age of children admitted to their hospital with meningococcal disease is 14 months.

Avendano and colleagues (1993) evaluated the safety and immunogenicity of a Hib vaccine made with purified polyribosylribitol phosphate conjugated to tetanus toxoid (PRP-T). The 287 infants in the study were randomized to receive either the Hib vaccine combined in a single injection with DTP, separate injections of Hib vaccine and DTP, or separate injections of DTP and a placebo. The one death during the study, resulting from pneumonia and sepsis, occurred 38 days after the second dose of combined DTP and PRP-T. Cultures obtained from the infant were positive for Streptococcus pneumoniae.

Chazono and colleagues (1991) described the side effects following use of DTaP vaccine in Japan and compared that experience with the side effects reported in a Phase III trial of the acellular pertussis vaccine component in Sweden in 1986. Pediatricians in four health centers in Japan collected information on the number of children diagnosed with any infectious diseases after receiving DTaP vaccine, as well as information on cases of pertussis or abnormal reactions such as febrile seizures. Information was obtained from medical charts or from parents or guardians contacted by telephone or mail. Of the 940 infants for whom information was available, three children had an infectious disease (one case each of measles, mumps, and varicella). The authors contrasted their findings with those from the Phase III acellular pertussis trial in Sweden, where three deaths from severe invasive bacterial infection (i.e., Hib, pneumococcal, and meningococcal infections) occurred among 1,385 children.

Ruuskanen and colleagues (1980) examined antibody responses and adverse reactions following receipt of inactivated polio vaccine. Children in the study received 2 doses of IPV as well as 4 doses of DTP between the ages of 3 months and 24 months. Information on reactions following receipt of IPV was obtained from questionnaires returned by parents of 225 of 380 children in the study. Fever (greater than or equal to 37.5°C) was reported for about 17 percent of children and was clinically significant (greater than or equal to 38.5°C) in 5 percent of the children. Fevers usually began the same day as vaccination and lasted up to 2 days. In a few children (exact number unspecified), fevers started 6 or more days following vaccination. The authors believed that these fevers were caused by infection, and not vaccination, although the basis for their belief was not stated.

Causality Argument

The committee reviewed several case-control or cohort studies (Black et al., 1991; Burstein and Fleisher, 1994; Davidson et al., 1991; Griffin et al., 1992; Kristensen et al., 2000) and a randomized controlled trial (Otto et al., 2000) (see Table 2). Vaccine exposure varied among the studies but fit the committee's definition of exposure to “multiple immunizations.” The studies examined the effects of the addition of one vaccine to an existing immunization schedule, of one vaccine consisting of antigens from more than one infectious agent or strain of virus (e.g., DTP, OPV, or MMR), or of several vaccines received at the same time. Outcome measures in the studies also varied, with the “disease” group including subjects who had a positive culture to invasive bacterial disease, who had symptoms related to infectious diseases, or who had died. Limitations of the studies included a potential health care utilization bias and high dropout rates. Despite these variations and limitations, the overall findings from the studies consistently demonstrated either no effect or a beneficial effect of multiple immunizations on heterologous disease. Therefore, the committee concludes that the epidemiological and clinical evidence favors rejection of a causal relationship between multiple immunizations and an increased risk of heterologous infections.

Uncontrolled Observational Studies

An ecological analysis by Hyoty and colleagues (1993) and an update by Hiltunen et al. (1999) examined the incidence of type 1 diabetes before and after introduction of MMR vaccine in Finland in 1982. Periodic mumps epidemics had been suggested as a contributing factor in the incidence of Type 1 diabetes, and the introduction of MMR vaccine resulted in almost complete disappearance of mumps. Data on the incidence of type 1 diabetes among children ages 0–14 years was obtained from a national registry for the years 1966–1996. The authors found a continuing increase in the incidence of diabetes over the period, especially among children ages 0–4 years and 5–9 years, but no cohort effect associated with the introduction of MMR vaccine was observed. The authors concluded that neither wild-type mumps nor MMR vaccine were related to the continuing increase in diabetes.

In a letter reporting on an ecological analysis, Classen (1996) examined IDDM incidence in children born before and after the introduction of a hepatitis B immunization program in New Zealand in 1988. At the time, children in New Zealand were also routinely immunized with DTP, MMR, and OPV vaccines. Exposure was based on birth year, and diabetes cases were identified through the diabetes registry in Christchurch. The author reported an increase in average annual incidence of diabetes after introduction of hepatitis B vaccine, although no control was made for the general secular trend of increasing diabetes incidence rates. The incidence rate for 1982–1987 was 11.2 cases per 100,000 per year, and the rate increased to 18.2 cases per 100,000 per year for 1989–1991. The authors proposed a possible link between the hepatitis B vaccine, and the timing of its administration, and the rising incidence of type 1 diabetes, but the ecological nature of the study limits the ability to make inferences about causation.

Controlled Epidemiological Studies

Vaccine Safety Datalink. In a case-control study, DeStefano and colleagues (2001) examined the association between childhood vaccines and the risk of developing type 1 diabetes. Data for both cases and controls were obtained from the four health maintenance organizations (HMOs) that participate in CDC's Vaccine Safety Datalink project. The cases were 252 children diagnosed with type 1 diabetes and the 768 controls were matched to individual cases on HMO, sex, date of birth (within 7 days), and length of health plan enrollment. For each case, vaccination history prior to the date of diabetes diagnosis was gathered from a medical chart review. The same reference date was used to obtain vaccination histories for the matched controls.

The vaccines evaluated included DTaP, DTwP, MMR, varicella, Hib, and hepatitis B. Oral polio vaccine was not included in the analysis because so few children had not received it (one case and three controls). Also tested was the effect of the timing of the first hepatitis B vaccine (never vaccinated; birth to 14 days, 15 to 55 days, or 56 or more days). In addition, the effect of differences in the schedule of Hib immunization (one dose at 21 to 27 months of age versus 3 doses in the first 8 months plus a fourth dose at 12 to 18 months) was examined.

On the basis of a conditional logistic regression model stratified by the matching variables, the odds ratio for each of the vaccines was nonsignificant. However, after adjusting for race/ethnicity and family history of type 1 diabetes, the odds ratio for whole-cell pertussis was 0.23 (95% CI 0.06–0.93). The highest adjusted odds ratio was for MMR (1.43, 95% CI 0.71–2.86) but was not statistically significant. Variations in the timing of hepatitis B vaccine produced no significant differences in the risk of type 1 diabetes. Similarly, there were no significant differences among various Hib immunization schedules. However, the ability to compare the different Hib schedules was limited, since only a few children received either no Hib vaccine or received one dose at 21 to 27 months. The authors concluded that neither the receipt of routine childhood vaccines nor the timing of certain vaccines was associated with an increased risk of type 1 diabetes.

EURODIAB. In a case-control study, a group examined infections and vaccinations as risk factors for type 1 diabetes (EURODIAB, 2000). Cases and controls were identified through seven European centers, each of which operates a population-based diabetes registry. As cases, there were 900 children with diabetes onset before age 15. The 2,302 controls were matched to the cases by age distribution. Vaccination history was obtained from parents through interviews or questionnaires and validated by official sources or entries in child health records held by the parent. Vaccines received included BCG, polio, tetanus, diphtheria, pertussis, rubella, measles, mumps, and Hib.

The odds ratios for all nine vaccines were nonsignificant using either a Mantel Haenszel analysis stratified by center, or a logistic regression analysis with adjustment for center, age group, breast feeding, birth weight, maternal age, jaundice at birth, asthma, and vitamin D supplementation. There was no evidence, the authors concluded, that vaccinations increase the risk of type 1 diabetes. The study may have been compromised by ascertainment bias. About 75% of responders had validated vaccination records available. Validation was based on either a review of official records or on parental recall of exact vaccination dates, even if the investigator did not see a record. The latter may have contributed to imprecision in assigning vaccine status.

Finland. Karvonen and colleagues (1999a) studied the relationship between multiple vaccines and type 1 diabetes by examining the effect of adding Hib vaccine to the routine childhood immunization schedule. Incidence of type 1 diabetes was compared in cohorts of Finnish children born before or after a Hib vaccine efficacy trial and followed for 10 years. One cohort of 128,936 children was born between October 1983 and September 1985, prior to the Hib vaccine trial and thus was not exposed to the vaccine. Children born between October 1985 and August 1987 participated in the Hib vaccine efficacy trial. These children were divided into two cohorts: 59,238 children who were born on odd days were vaccinated with Hib at 3, 4, 6, and 14 to 18 months; 57,114 children born on even days were vaccinated at 24 months only. All children were assumed to have received BCG, diphtheria-tetanus-pertussis, polio, and measles-mumpsrubella vaccines. Newly diagnosed cases of diabetes among all three cohorts were ascertained from a national hospital discharge registry (1983–1986) or a nationwide prospective childhood diabetes registry (1987–1997).

There was no significant difference in the risk of diabetes by age 10 between the children who did not receive the Hib vaccine and children who were vaccinated at 24 months of age. Similarly, no difference in risk was found between the children first vaccinated at 3 months of age and those vaccinated at 24 months. For each of the comparisons, the relative risk was near 1.0. The authors concluded that neither the addition of Hib vaccine to the immunization schedule nor the timing of Hib vaccine increased the risk of type 1 diabetes in children. Estimates of both vaccine exposure and diabetes cases were based on aggregate data from three cohorts and from the population as a whole. Thus, interpreting the results at the level of the individual is difficult.

Sweden. Heijbel and colleagues (1997) examined the effect of pertussis vaccination in infancy on the risk of developing type 1 diabetes. Cumulative incidence of type 1 diabetes at ages 0 to 12 years was compared in cohorts of children born before or after the pertussis vaccine was removed from the routine immunization schedule in Sweden. Specifically, cohorts of children born in 1977 (96,057 children) and in 1978 (93,248 children) received pertussis vaccine through DTP vaccinations. Children born in 1980 (97,064 children) and in 1981 (94,065 children) received DT vaccine.

Rates of exposure to DT or DTP vaccine were assessed from yearly vaccination reports of child health care centers. Estimates of pertussis vaccine exposure among children born in 1980 or 1981 were based on reports of vaccine doses released. Data from the national diabetes registry were used to determine the cumulative incidence of type 1 diabetes in each birth cohort. No difference was found in the cumulative type 1 diabetes incidence for any of the birth cohorts. The population-level immunization data limits ability to make extrapolations at the individual level.

In another study in Sweden, Blom and colleagues (1991) used a case-control approach to examine infections and vaccinations as possible risk factors for type 1 diabetes. A total of 339 children age 0 to 14 years and newly diagnosed with diabetes from September 1985 through August 1986 were identified from the Swedish childhood diabetes register. The study included 528 controls matched on age, sex, and county, who were identified from the Swedish population register. Vaccination histories in the two groups were collected from local child health care centers and school health units. The vaccines received by these children included BCG, smallpox, DTP, DT, tetanus, polio, MMR, measles, mumps, and rubella. Similar proportions of children in both the case and control groups had received each of these vaccines.

The odds ratios for risk of type 1 diabetes for individual vaccines (or combination products) were nonsignificant. The odds ratio for measles vaccine alone was 0.74 (95% CI 0.55–1.00), but when the measles-mumps-rubella and/or measles vaccine was considered, the odds ratio suggested a significant reduction in risk (OR=0.69, 95% CI 0.48–0.98). The authors concluded that the evidence did not support an increased risk of type 1 diabetes after vaccination and that the measles vaccine may have a protective effect against type 1 diabetes. The study was well conducted, but the number of cases of diabetes was small, so that the study may have been under-powered to detect significant differences in the 13 categories of vaccination examined. Power calculations for Type II errors were not provided. The isolated finding of a possible protective effect (just at the 0.05 level of statistical significance) of the measles vaccine is difficult to interpret given the multiple comparisons made in the analysis.

Other Studies

The committee reviewed additional studies examining the relationship between diabetes and BCG or smallpox vaccination (Classen and Classen, 1996, 1997, 1999; Dahlquist and Gothefors, 1995; Parent et al., 1997). But because these vaccines are not routinely administered to children in the United States, the studies were not considered directly relevant to the committee's task.

Causality Argument

The committee found five controlled studies (Blom et al., 1991; DeStefano et al., 2001; EURODIAB, 2000; Heijbel et al., 1997; Karvonen et al., 1999a) (Table 3) and three ecological studies (Classen, 1996; Hiltunen et al., 1999; Hyoty et al., 1993) that studied this relationship. The studies looked at the effects of the addition of one vaccine to an existing immunization schedule, of one vaccine consisting of antigens from more than one infectious agent or strain of virus (e.g., DTP, OPV, or MMR), or of several vaccines received at the same time. Despite these variations, the overall findings from the studies consistently demonstrated no effect of multiple immunizations on the incidence of type 1 diabetes. Therefore, the committee concludes that the epidemiological and clinical evidence favors rejection of a causal relationship between multiple immunizations and an increased risk of type 1 diabetes.

Uncontrolled Observational (Ecological) Study

Anderson and colleagues (2001) examined the relationship between immunization and allergic disease by comparing trends in immunization rates with the prevalence of allergic disease symptoms. Allergy data, specifically for asthma, allergic rhinoconjunctivitis, and atopic eczema, were obtained for children ages 6 to 7 years and 13 to 14 years from centers participating in the International Study of Asthma and Allergies in Childhood (ISAAC). Those data were compared with national and local immunization rates for BCG, DTP, and measles.

For children age 6 to 7 years, allergy data were obtained from 91 centers, with a median of 2,996 children per center (range 1,104–6,533). Local immunization rates were available for 57 centers. Allergy data for children age 13 to 14 years were available from 154 centers with a median of 2,064 children (range= 1,046–11,400); 92 centers had local immunization data.

In the 13- to 14-year-old age group, the authors observed a significant negative association (rank correlations, p<.05, adjusted for socioeconomic factors) between local DTP rates and wheezing (−0.53, 95% CI, −1.49, 0.43), rhinoconjunctivitis (−0.60, −1.02, −0.19), and atopic eczema (−0.27, −0.76, 0.21). For measles vaccine, significant negative correlations were found for rhinoconjunctivitis (−0.47, −.98, 0.04) and atopic eczema (−0.42, −0.98, 0.13). No associations were observed for children age 6 to 7 years. The authors concluded that DTP vaccines are not risk factors for allergic disease at the population level. However, the authors noted that because immunization data were available only at the population level, associations at the individual level could not be excluded.

Controlled Observational Studies

Wellington, New Zealand. Wickens and colleagues (2001) studied potential risk factors for asthma in a case-control study of children age 7 to 9 years. Cases were 233 children selected from the Wellington, New Zealand, arm of ISAAC who currently had asthma. Controls were 241 children who never had asthma or asthma-like symptoms. Information on demographics and risk factors for asthma, such as parental smoking history, was collected through interviews with parents. Vaccination histories were gathered from the medical records of study participants. Vaccines to which case and control children were exposed included DTP, DT, hepatitis B, polio, MMR, measles, and BCG.

Odds ratios were individually calculated for the following vaccine categories: DPT, DT or DPT, hepatitis B, polio, measles or MMR, MMR, and BCG. Only a few children were unvaccinated (no DTP=38; no polio=25; no MMR =333), and children in the unvaccinated group may have been misclassified, as some of their vaccination records were missing. In an univariate analysis, the asthma odds ratios for most vaccine exposures were non-significant. For MMR, however, the odds ratio was 1.62 (95% CI 1.06–2.47), though for measles or MMR combined the risk for asthma was not significantly increased (OR=1.52, 95% CI 0.89–2.58). In a multivariate analysis, the association between asthma and MMR vaccine was nonsignificant (OR=1.43, 95% CI 0.85–2.41). The authors suggested that the significant univariate association observed between MMR vaccination and asthma was confounded by measles infection and polio vaccination and concluded that there was no association between vaccination and asthma, commenting that the (non-significant) elevated odds ratios could be due to selection bias, exposure misclassification or multiple comparisons.

United States. Hurwitz and Morgenstern (2000) examined the association between vaccination and allergies (as well as allergy-related respiratory symptoms) among children and adolescents, using data collected between 1988 and 1994 for the third National Health and Nutrition Examination Survey. For 13,944 infants, children, and adolescents ages 2 months through 16 years, information was gathered regarding DTP or tetanus vaccination, lifetime history of physician-diagnosed allergy (asthma or hay fever), and allergy symptoms in the past 12 months. Children's parents or guardians answered the survey.

Of the 13,944 subjects, 284 children reported no receipt of DTP or tetanus immunization and were classified as unvaccinated. Subjects with missing data (n =332) were excluded from the analysis. Adjusted odds ratios were calculated for nine different definitions of allergy related symptoms in the vaccinated and unvaccinated groups. Three closely related outcomes were statistically significant: allergy-related symptoms in the past 12 months (adjusted OR=1.63, 95% CI 1.05–2.4), any lifetime allergy history or 12-month allergy symptoms (OR= 1.7, 95% CI 1.1–2.6), and nose and eye symptoms (OR=2.2, 95% CI 1.3–3.7). Odds ratios for asthma, severe allergic reactions, sinus problems, and wheezing ranged from 1.2 to 2.0 and were not statistically significant. The prevalence of atopy as determined by skin testing with 10 environmental allergens was similar in the two groups (49% and 51%). The authors concluded that DTP and tetanus vaccination increases the risk of clinical allergy but not atopy in children and adolescents. The authors noted several limitations. Subjects excluded from the analysis were more likely to have risk factors related to being unvaccinated than subjects included in the analysis. Excluded subjects were also more likely to have allergies than unvaccinated subjects. These limitations may have biased the estimated effects towards a positive effect. The authors' reanalysis of the data, assuming that the excluded subjects were all unvaccinated, lowered the estimated effects. The limitations reported by the authors included the cross-sectional design, self-reported vaccination status and allergy symptoms, use of proxy respondents, missing data on 2.4% of subjects, small number of unvaccinated subjects (2% of the population), lack of clinical information about the clinical nature of symptoms, recall bias, selection bias for care-seeking behavior, and unmeasured confounding. To this could be added the very limited ability to control for confounders due to the small numbers of subjects in the unvaccinated groups, particularly when analyses were restricted to children over 5 years of age, when the number of unvaccinated children for different outcomes ranged between 1 and 8, and the inclusion of children under two years of age, in whom allergic symptoms are difficult to separate from infectious ones.

United Kingdom. Farooqi and Hopkin (1998) examined the relationship between childhood infections and subsequent allergic disorders, along with other potential risk factors, such as immunizations. Of the 3,062 children born between 1975 and 1984 and seen at a general medical practice in the United Kingdom, 1,934 children (63% of original cohort) who remained registered at the practice and who had complete immunization records were included in the study. Information was obtained from medical records and a regional child health database on vaccines received (i.e., diphtheria-tetanus-whole-cell pertussis/diphtheria-tetanus, polio, and measles) and diagnoses of allergic diseases such as hay fever and asthma.

A higher proportion of children with allergic disease received pertussis vaccine, which showed a statistically significant association with allergy, asthma, and hay fever. The unadjusted odds ratio for DTP immunization (either complete or incomplete course) and allergy was 1.57 (95% CI 1.28–1.96); for asthma, the odds ratio was 1.44 (95% CI 1.17–1.85); and for hay fever, the odds ratio was 1.56 (95% CI 1.21–2.02). A multiple logistic regression analysis showed pertussis immunization statistically associated with allergic disease (OR =1.76, 95% CI 1.39–2.23), along with maternal atopy (OR=2.0, 95% CI 1.5– 2.7) and antibiotic treatment in the first two years of life (OR=2.1, 95% CI 1.6– 2.6). The authors concluded that whole-cell pertussis vaccination significantly increased the odds of developing asthma, and they noted as well that the association persisted when children with a history of allergic disease prior to immunization were excluded from the analysis. The authors also noted the possibility of health care-seeking bias in their results, but they found no difference between vaccinated and unvaccinated children in the number of health care visits. However, the removal of 36 percent of potential subjects who were no longer registered at the practice, and their exclusion from the study suggests a potential health care seeking bias in the sampled population. Because of this bias, interpretation of the study's findings warrants caution.

Christchurch, New Zealand. Kemp and others (1997) examined the relationship between immunization and allergic disease in infants. Data on vaccines received, asthma, and allergic diseases were collected for 1,265 children born in 1977 and participating in a New Zealand health study. Children in this health study were expected to have received DTP and polio vaccines at ages 3 and 5 months and measles vaccine at 12 to 15 months. Information on vaccines and the occurrence of asthma and/or allergic diseases (including rhinitis) at ages up to 5, 10, and 16 years was obtained from parents' medical diaries and physicians' records, plus additional inquiries directed to parents. Of the children in the study, 23 did not receive either DTP or polio vaccine at the scheduled age and were classified as “nonvaccinated.”

For children age 0 to 10 years, a greater proportion in the immunized group reported asthma episodes, asthma consults, and allergy consults, compared with children in the nonvaccinated group, for whom there were no reports of asthma or consultations for asthma or allergy. Because no events occurred in the nonvaccinated group, the relative risk was infinite (95% CI 1.03-infinity). For children age 11 to 16 years, among whom 5 allergic and asthma consultations/episodes occurred in the nonvaccinated group, the relative risks for vaccinated children (2.7 to 5.6) were not significant. No association was seen between measles vaccine and risk of asthma episodes in children age 0 to 10 years (OR 1.0, 95% CI 0.9–1.1). Minimal effect of potential confounders was seen when each were considered singly in multivariate models, and the unvaccinated children differed from vaccinated children in ways that should have put them at higher risk for asthma-related symptoms (e.g., lower SES, more parental smoking, more family history of allergy, more pet ownership). The authors concluded that infant immunization for DTP and polio may increase the risk of developing asthma in childhood. But limitations of the study, including the small number of unvaccinated children, the marginal significance of the results (with high sensitivity to a change of even one subject in the unvaccinated group), the potential for a health care utilization bias, and difficulty in adjusting for confounders (due to the small number of unvaccinated children), warrant caution in interpreting these findings.

Controlled Clinical Trials

Sweden. Nilsson and others (1998) studied the development of allergic disease during a randomized controlled trial of pertussis vaccines in Sweden (Gustafsson et al., 1996). A total of 669 children were randomized to one of four exposure groups: 2-component DTaP, 5-component DTaP, DTwP, or DT (control). As part of the recommended immunization schedule in Sweden, children were also given inactivated polio and Hib vaccines two or more weeks later or at the same time. Parent questionnaires, clinical findings, and medical record information provided data on diagnoses of bronchial asthma, atopic dermatitis, allergic rhinoconjunctivitis, urticaria, and food allergy. Children were followed from ages 2 months to 2.5 years.

Cumulative incidence of allergic disease at age 2.5 years in the three pertussis vaccine groups was similar to that in the DT group after adjusting for family history of allergy. Among children whose parents had no history of allergic disease, those who received DTwP were estimated to have an 8% relative reduction in risk (one-sided 95% CI=28% increase) of allergic disease compared with the control group, which had a predicted risk of 22.5%. The combined DTaP group had an estimated 10% relative increase in risk (one-sided 95% CI=41% increase). When both parents had a history of allergy, DTwP was associated with a 6% relative reduction in risk (one-sided 95% CI=18% increase) and DTaP with a 8% relative increase in risk (one-sided 95% CI=27% increase). With the authors' one-tailed test for increased risk, they calculated that with the size of the sample, they had approximately 80% power to detect a 50% increase in risk, roughly the magnitude reported in some other studies. Differences between the DTwP and DTaP groups were not significant. Interestingly, subsequent asthma was elevated over two-fold (p=0.03) by the development of clinical pertussis, which occurred in 14.5% of unimmunized and 4.4% of immunized children, although the authors raised the possibility that this could partly be due to transient bronchial hyperreactivity after whooping cough.

This study had a nonstandard technique of calculating confidence limits, so the upper limits on the relative risk increase may not be directly compared to those of other studies. Although they used a logistic regression model for analysis, which would ordinarily generate odds ratios (and attendant Is) for the risk of asthma due to various immunizations, they appeared (not described) to use the regression model to estimate absolute risks, and calculate relative risk estimates based on these fitted absolute risks. It is also not clear how upper confidence limits on these relative risks were calculated. In addition, their use of one-sided upper limits for the relative risk would be lower than the more conventional two-sided limits by roughly 4–8 percentage points (e.g., 34% [two-sided] vs. 28% [one sided]). The combination of these two factors makes the upper confidence limits reported in this study somewhat difficult to interpret and to compare to other studies, which used odds ratios. For example, the one-sided upper limit of a 28% relative risk increase reported by these authors for whole cell vaccine (over a baseline risk of 22.5%) corresponds approximately to a two-sided upper limit of a 50% increase in the odds of asthma.

The authors concluded that there was no evidence supporting an increase in allergic disease after pertussis immunization of the magnitude reported in some other studies, and if there was any increased risk, they thought it was most likely to be associated with the acellular pertussis vaccines. However, limitations included wide confidence intervals restriction to symptom development before 2.5 years of age, the unclear and nonstandard statistical approach, and the focus only on the pertussis component of the immunization schedule. Strengths included the randomized, prospective design and small (<5%) loss to follow-up.

Causality Argument

The committee reviewed five studies that utilized controls (Farooqi and Hopkin, 1998; Hurwitz and Morgenstern, 2000; Kemp et al., 1997; Wickens et al., 2001), including a randomized controlled trial (Nilsson et al., 1998) (see Table 4) and one ecological study (Anderson et al., 2001). Outcomes assessed included allergic symptoms (wheezing) and allergic disorders (hay fever and asthma). All the studies examined exposure to DTaP or DTwP, and other vaccines given concurrently, such as MMR and polio vaccines, but no two studies examined exactly the same exposure.

While many of these studies reported elevated odds ratios linking immunizations to some allergic outcome, some of which were statistically significant, methodological weaknesses within individual studies, as well as the pattern of results across studies diminish the confidence that the observed associations reflect causal relationships. In the two studies that reported a significant positive effect of DTP or tetanus immunization or the pertussis component of DTwP (Farooqi and Hopkin, 1998; Hurwitz and Morgenstern, 2000), potential sampling bias, caused by substantial losses to follow-up or restriction to subjects with regular medical care, could have distorted the relationship between immunization and allergies.

A problem in most of the studies was that the number of unvaccinated children was small, limiting the ability to control for potentially confounding factors, which are numerous and strong for the outcomes of asthma and atopy, and particularly complex when considering risk over an entire childhood. Adequate control of confounding is a serious issue for observational designs, particularly in this domain, as nonimmunized children typically differ on baseline characteristics from immunized children in ways that are not always measurable. Control can be compromised by imperfect and possibly biased confounder and outcome measurement introduced by retrospective, unblinded review of records or parental report of outcomes or exposures that occurred in the past.

Finally, the findings of the studies, taken as a whole, did not show a consistency of findings that would outweigh the concerns about individual studies. While some studies pointed to the pertussis vaccine as a risk factor for allergic syndromes with no effect of MMR, another found that MMR vaccine was the strongest risk factor. The ecological study indicated a protective DPT effect, and the only randomized study indicated minimal or no effect of pertussis vaccines, with a non-significant reduction in risk from the whole-cell vaccine.

Given the design weaknesses in the observational studies, with effect sizes and modest degrees of statistical significance that are not robust to probable biases, and a randomized trial study that does not support the risk factor most frequently implicated in the observational studies, the committee concludes that the epidemiological and clinical evidence is inadequate to accept or reject a causal relationship between multiple immunization and an increased risk of allergic disease, particularly asthma.

Proposed Mechanisms for Induction of Autoimmunity Through Infection

Infection can induce immune-mediated tissue injury. In most cases, this injury is short-lived and resolves as the immune system eliminates active infection. The injury is a consequence of the immune response to the foreign invader, and when the invader is eliminated, the damaging immune process ceases. In some diseases, however, infection appears to induce an injurious immune response directed, at least in part, against self antigens. Nevertheless, true autoimmune injury must be distinguished from immune-mediated injury resulting from persistent but undetected infection. If the infectious agent was not detected, ongoing immune-mediated responses to that agent and the resulting injury of host tissues could be interpreted as autoimmunity, when in fact the immune response was directed against the foreign microbe and not against self.

Two major mechanisms—molecular mimicry and bystander activation—are proposed to account for the activation of self-reactive T and B cells and the induction of autoimmunity by infection (Albert and Inman, 1999; Bach and Chatenoud, 2001; Benoist and Mathis, 2001; Davidson and Diamond, 2001; Marrack et al., 2001; Regner and Lambert, 2001; Rose, 2001; Singh, 2000; Wucherpfennig, 2001; Zinkernagel, 2001).

Also described below is the possible link between infection and autoimmunity through the mechanism of nonspecific or polyclonal T cell activation. The evidence regarding the possibility that these mechanisms actually contribute to autoimmune diseases is discussed in a later section.

Molecular mimicry. An antigenic epitope from a microbe that is structurally similar to (mimics) an epitope of a self-molecule has the potential to trigger the activation of self-reactive, naïve T or B lymphocytes. Once activated, self-reactive T cells could expand in number and mature into effector (memory) T cells that have a lower threshold for activation by self antigens. These cells would also gain the ability to migrate to specific tissues, produce additional mediators/cytokines, and to mediate injury on contact with cross-reacting self antigens. In addition, they would gain the potential to help B cells that are responding either to the same antigen as the T cells or to other self-antigens that are physically linked to it.

Bystander activation. Bystander activation results when an infection creates environmental conditions that allow the activation of self-reactive T and B cells that are normally held in check. It does not require that antigens of the infectious agent be structurally similar to self-antigens. Bystander activation may be mediated in part by infection-induced death of host cells, which results in the release of greater amounts of self peptides or in the generation of novel self-peptides (i.e., novel or cryptic epitopes not normally found in the absence of the infection). As part of this process, molecules derived from the microbes (and perhaps also from the necrotic host cells—e.g., heat-shock proteins) would stimulate other components of the immune system.

In particular, antigen presenting cells (i.e., dendritic cells) gain an increased abundance of self peptide/MHC complexes and costimulatory molecules (e.g., B7 molecules) and produce cytokines, including interleukin 12 (IL-12) and tumor necrosis factor (TNF). Collectively, these changes can allow the activation of naïve self-reactive T cells that are induced to differentiate into Th1 effector T cells, which play a central role in most forms of human autoimmunity (reviewed in Bach and Chatenoud, 2001; Marrack et al., 2001; Martin et al., 2001; Parham, 2000; Robles and Eisenbarth, 2001; Singh, 2000; Wills-Karp et al., 2001). In turn, these self-reactive Th1 T cells would be available to provide help to self-reactive B cells, responding either to the same antigen as the T cells or to other antigens that are physically linked to them.

Nonspecific or polyclonal T cell activation. It is also possible that infections could activate a variety of T or B lymphocytes that would otherwise respond only to certain antigens. This non-antigen-specific response is referred to as a polyclonal or oligoclonal response. For example, toxins produced by Streptococcus pyogenes and Staphylococcus aureus can act as superantigens, binding both to T cell receptors and to MHC molecules and activating a substantial fraction (greater than 5 percent) of the total T cells of an individual. This overactivation of the immune system leads to the acute “toxic-shock” syndromes associated with the toxin-producing strains of these bacteria. Another example is polyclonal B cell activation following Epstein-Barr virus (EBV) infection and capable of enhancing autoimmune reactions. These nonspecific immune responses are usually self-limited, however, and resolve as the infection is cleared.

If self-reactive T cells are activated by a nonspecific immune response, they can induce autoimmunity. Superantigen-induced activation normally ends in the programmed death (apoptosis) of the activated cells, terminating the response. However, other microbial products, like endotoxin, that trigger the innate immune response can enhance the survival of superantigen-activated T cells (Vella et al., 1995, 1997), and could—in the context of genetic differences in mechanisms controlling the death of activated T cells—prolong the survival of self-reactive T cells, in theory allowing them to mediate self-injury.

Examples of Molecular Mimicry and Bystander Activation in Autoimmunity

The processes of molecular mimicry and bystander activation are not mutually exclusive. These mechanisms may act in synergy or at different stages in the initiation and progression of autoimmunity. Mimicry can be thought of as an initiator, because it triggers the initial activation of cross-reactive T cells, which then replicate and differentiate into effector T cells with a lower threshold for activation by self-antigens. The bystander activation of antigen presenting cells facilitates further replication of these T cells and their differentiation into Th1 T cells and also facilitates their recruitment to the target tissue.

Bystander effects are almost certainly essential for molecular mimicry to succeed in overcoming normal regulatory mechanisms (see Figure 3). Through a process referred to as epitope spreading, the initial activation by an infection of cross-reactive T cells that react to a single self-antigen may play a critical role in facilitating, in a bystander-like manner, the activation of T cells that recognize other self antigens. Thus, even after the infection has resolved and cross-reactive T cells are no longer stimulated by foreign antigens, the autoimmune process would be perpetuated and amplified by self-reactive T cells. These mechanisms have been delineated in animal models, which provide illustrative examples:

Herpes simplex-induced keratitis. In some strains of mice, acute herpes simplex virus infection of the cornea is followed by the development of autoimmune corneal injury (keratitis) (Benoist and Mathis, 2001; Panoutsakaopoulou et al., 2001; Zhao et al., 1998). Genetic factors, including MHC type, account for the differences in susceptibility between strains of mice. In susceptible mice, the development of keratitis depends on the presence of a specific viral antigen, which triggers the generation of T cells that are cross-reactive with a self-protein. These activated, cross-reactive T cells migrate via the blood back to the cornea, where they induce the keratitis.

Infection at sites other than the cornea does not induce the disease, even though cross-reactive T cells are generated. This most likely reflects the need for multiple non-antigen-specific bystander effects of the viral infection. For example, injury to the cornea releases cross-reactive self-peptides that help amplify the number of self-reactive T cells that mature into injurious Th1 effector T cells. The injury to the cornea also triggers the production of inflammatory cytokines that recruit antigen presenting cells and T cells to the infected eye, where the T cells are activated and produce keratitis. (In humans, a similar bystander process likely accounts for trauma-induced sympathetic ophthalmia, in which injury of one eye leads to an autoimmune attack against both the injured and uninjured eye (Chan and Mochizuki, 1999).

Consistent with an important role for non-antigen-specific or bystander effects in herpes simplex-induced keratitis is the ability to induce the disease with mutant strains of herpes simplex virus that lack the cross-reactive epitope. But keratitis can be induced by these mutant strains only if the numbers of potentially self-reactive T cells have been artificially increased. Simply increasing the number of T cells is not sufficient, however; local infection or injury to the eye is necessary to initiate the disease process.

Experimental allergic encephalomyelitis. Experimental allergic encephalomyelitis (EAE) is a demyelinating disorder induced in rodents by immunization with myelin extracts, proteins, or peptides found in myelin (Goverman et al., 1997; Zamvil and Steinman, 1990).⁶ Although EAE is one of two commonly used models of MS, it may more accurately model post-infectious acute disseminated encephalomyelitis (ADEM). As with most autoimmune disorders, only some strains of mice are susceptible to EAE, and susceptibility is determined in part by the MHC type. Molecular mimicry is not involved in EAE, since the auto-antigen is the inducing agent.

To induce this disorder, mice are immunized with myelin protein or peptides in a potent adjuvant. The adjuvant must not only increase the persistence of the antigen, but must stimulate antigen-presenting cells to more effectively present these peptides and to produce cytokines, including IL-12, that induce the formation of Th1 T cells that in turn can migrate to the central nervous system and produce injury.

The most commonly used adjuvant in animal studies is complete Freund's adjuvant, which contains dead mycobacteria in a water-in-oil emulsion. Heat-killed whole Bordetella pertussis or pertussis toxin can also be use as an adjuvant in this model (Blankenhorn et al., 2000; Goverman et al., 1997; Jee and Matsumoto, 2001; Zamvil and Steinman, 1990). It is the presence of dead mycobacteria that makes Freund's adjuvant effective. Multiple components of the mycobacteria (including those of their membranes and DNA) trigger the innate immune response via Toll-like receptors (Akira et al., 2001; Medzhitov and Janeway, 2000), causing the maturation of antigen presenting cells and the production of IL-12 and other potent cytokines. The inclusion in the initial immunization not only of complete Freund's adjuvant but of biologically active pertussis toxin is common and increases the frequency of clinical disease. The pertussis toxin probably acts as adjuvant and enhances the ability of the T cells to migrate into the central nervous system (Goverman et al., 1997).

Once the demyelinating process has been initiated, exacerbations can be induced by bystander mechanisms alone with the administration of IL-12 or microbial agonists (endotoxin, dsRNA [double-stranded RNA], and bacterial DNA) that stimulate the production of IL-12 and activate other aspects of the innate immune response (Constantinescu et al., 1998; Goverman et al., 1997; Segal et al., 1997).

Human Inflammatory and Autoimmune Conditions Induced by Infection

Various specific infections are associated with the development of self-injury. For many of these infections, there is considerable evidence that the injury is caused, at least in part, by autoimmunity and is not simply a consequence of the immune response to the infectious agent. For example, group A streptococcal infection in humans is clearly linked to the induction of rheumatic fever. Inflammation also occurs in the skin and joints, but the major disability results from injury to the heart or, in the case of Syndenham's chorea and perhaps pediatric autoimmune neuropsychiatric disorders (PANDAS), to the brain (Marrack et al., 2001; Perlmutter et al., 1998; Wucherpfennig, 2001). Human (and mouse) antibodies to certain streptococcal M proteins cross-react with cardiac myosin and can injure cardiac muscle cells in culture, suggesting that molecular mimicry is a factor. However, it is not certain that these antibodies are the cause of the clinical damage observed following infection, nor is there clear evidence that infection leads to the generation of antibodies or T cells that are truly self-reactive rather than cross-reactive. That is, there is no clear evidence of epitope spreading and autoimmunity that persists after infection is eradicated.

Other examples in humans of infection-induced acute inflammatory injury that appears to be autoimmune in nature include Guillain-Barré syndrome, ADEM, reactive arthritis, and herpes simplex keratitis (IOM, 1991, 1994; Marrack et al., 2001; Stuve and Zamvil, 1999; Wucherpfennig, 2001). Guillain-Barré syndrome is most strongly associated with Campylobacter jejuni infection, which may account for up to 30 percent of cases (McCarthy and Giesecke, 2001). It has also been associated with a variety of other, mostly viral, infections. ADEM has been associated with measles and mumps virus infections (IOM, 1994). In these conditions, the injurious immune and inflammatory response resolves over time once the infection is eradicated. The damage produced during the period of active infection may, however, result in death or persistent disability, and injury may be re-induced by recurrent infection with the inciting agent. Nonetheless, the resolution of active immune-mediated injury suggests that the loss of self-tolerance is not permanent, and the precise contribution of the various mechanisms cited remains uncertain.

Arguably the strongest evidence for true persistent autoimmunity induced by a specific infectious agent in humans is chronic Lyme disease arthritis. In this disease, there is clear evidence of genetic predisposition (linked to HLA type). A putative molecular mimic has been identified, with bacterial OspA (outer surface protein A) as a mimic of human LFA-1 (lymphocyte function-associated antigen). The disease persists in the absence of any evidence of persistent infection, and cross-reactive T cells are present in increased numbers in the joints of these patients (reviewed in Benoist and Mathis, 2001; Wucherpfennig, 2001).

As noted above, there is also strong evidence for a causal association between congenital rubella infection and type 1a diabetes: approximately 20 percent of individuals with congenital rubella develop diabetes by adulthood (Robles and Eisenbarth, 2001). The pathogenesis of diabetes in these cases shares many features with typical type 1a diabetes. The similar preponderance of HLA class II alleles and the presence of T cells reactive to GAD65 (glutamic acid decarboxylase) peptides (Ou et al., 1999; Robles and Eisenbarth, 2001) suggests that the disease results, at least in part, from congenital rubella-induced autoimmunity. By contrast, there is no evidence for a similar link between acquired rubella infection after birth and type 1 diabetes.

With the exception of chronic Lyme arthritis and congenital rubella-induced diabetes, the role of infection and of specific infectious agents in the induction or exacerbation of the major chronic autoimmune diseases in humans— including type 1a diabetes, MS, systemic lupus erythematosus, and rheumatoid arthritis—is uncertain. A number of associations have been made (see Table 5), but findings across studies have not been consistent and experimental evidence to support a causal link is currently lacking (Benoist and Mathis, 2001; Marrack et al., 2001; Wucherpfennig, 2001). It is worth noting, however, that of the possible mechanisms by which infection might contribute to autoimmunity, only molecular mimicry requires a specific association between an infectious agent and autoimmunity. The evidence that more than one infectious agent can trigger the onset of reactive arthritis and demyelinating neurological diseases is consistent with the notion that bystander effects may be a common feature of infection-induced autoimmune injury. Also, the evidence that a given T cell may be able to recognize multiple peptide-MHC complexes (Marrack et al. 2001; Wucherpfennig, 2001) means it is clearly possible that the proliferation of T cells that are cross-reactive with self-antigens might be induced by more than one infectious agent.

TABLE 5

Putative Examples of Molecular Mimicry in Human Autoimmune Disorders with Proposed Cross-Reactive Autoantigens and Infection-Derived Antigens.

Vaccine-Preventable Diseases and the Possibility of Vaccine-Induced Autoimmunity

The discussion above provided examples and a mechanistic framework from which to consider infection-induced immunological injury and autoimmunity. If the vaccine acts as a surrogate for the infectious disease it is designed to prevent, it might trigger the same type of immune-mediated injury as the infection itself. In previous reviews by the IOM and others (Chen, RT et al., 2001; IOM, 1991, 1994), the plausibility of this notion, and the evidence for or against causation in response to individual vaccines, were reviewed. A causal relationship between a vaccine and an autoimmune disorder was found for MMR and thrombocytopenia, OPV and Guillain-Barré Syndrome (GBS), and tetanus-containing vaccines and GBS. In addition, some types of influenza vaccine are associated with GBS and rubella with arthritis. Vaccinia is associated with acute disseminated encephalomyelitis. None of these findings, however, assessed the risk of autoimmune disease secondary to a “skewing” of immune response caused by multiple immunizations early in life. These studies do provide epidemiological evidence for a causal link between certain vaccines and immune-mediated injury and disease. However, in several cases in which the infection itself is known to trigger immune-mediated diseases (examples are measles, mumps, rubella, Borrelia burgdorferi infections), a causal link between immunization against the infection and the disease has not been established.

Assessment of Biological Mechanisms by Which Multiple Immunizations Might Contribute to Autoimmunity

As discussed previously, the total number of different vaccines and vaccine doses administered to children in the first years of life has increased substantially over the past 20 years in the United States. However, the number of individual proteins or polysaccharide antigens administered has declined from a peak of more than 3,000 in the early 1990s to about 130 with the current immunization schedule. By comparison, a single bacterial infection (e.g., H. influenzae or group A streptococcus) results in exposure to many more antigens, of which 50 or more induce B or T cell responses (Cunningham, 2000; Halsey, 2001). Thus, the number of antigens in current vaccines resembles the number to which the immune system responds in the physiological context of an infection. Nonetheless, the overall content and context of vaccine administration has changed because of changes in the number of vaccines administered, their timing, the nature and formulation of the vaccines, and the main vaccination route.

The current vaccine formulations and immunization schedule have been shown to induce protection or concentrations of antibody responses sufficient to mediate protection (Blackwelder, 1995; Halsey, 2001). In some cases, achieving protective efficacy has required modification of the amounts of individual components in multivalent or combination vaccines, such as the trivalent polio virus vaccines and the MMR vaccine. The need for such modifications is consistent with the notion that prior or simultaneous infection, or immunization with multiple vaccines or antigens, can influence the magnitude and quality of the immune response to individual antigens (Chen, HD et al., 2001; Gomez et al., 1997; Insel, 1995; IOM, 2001a; Selin et al., 1998, 1999; Vekemans et al., 2001).

That factors related to infection and immunization can affect immune responses raises the theoretical possibility that changes to vaccine formulations or the vaccine schedule might also alter the potential for vaccines to induce, facilitate, or amplify immune-mediated injury or autoimmunity. Theoretically, concomitant or sequential administration of multiple vaccines could have several possible effects, alone or in combination. Molecular mimicry might be enhanced through an additive or synergistic mechanism in which more than one vaccine induces cross-reactive T cells. Another possibility might be the generation of cryptic or novel epitopes that are not found when vaccines are given separately. In addition, the nature or magnitude of bystander effects might be altered.

Molecular mimicry. Molecular mimicry appears to be a mechanism by which Lyme disease causes rheumatoid arthritis. This constitutes evidence for a biological mechanism by which the OspA Lyme vaccine (which is not recommended for routine use, seen in Figure 2) could possibly induce immune-mediated injury. It remains theoretically possible that molecular mimicry occurs in response to other vaccines but that the extent of the immune response induced has been insufficient to induce clinically evident disease. If this were the case, the addition of new vaccines that also induced cross-reactive immune responses might initiate injury and epitope spreading sufficient to result in clinically manifest autoimmunity.

The following example illustrates this hypothetical possibility. In postinfectious encephalitis, a rare but known complication of measles virus infection, the injury appears to be mediated by infection-induced expansion and entry into the nervous system of T cells directed against myelin basic protein (also a target antigen in MS) (Johnson, 1987; Liebert, 1997). Whether this reflects a cross-reactive immune response induced by measles virus or a bystander effect, or both, is uncertain. If measles vaccine (or another vaccine) weakly stimulated T cells cross-reactive with myelin basic protein and were co-inoculated with a second vaccine that also weakly stimulated such cross-reactive T cells, clinically apparent autoimmunity might result from their combined effects, even though it might not result from use of either vaccine alone.

A related theoretical mechanism is altered molecular mimicry. Ingestion of high amounts of inorganic mercury or methylmercury can induce an autoimmune disease in genetically susceptible strains of mice and rats. This effect appears to reflect the modification by mercury of self-proteins—laminin in the rat and fibrallarin in the mouse (Hultman and Nielsen, 2001; Pollard et al., 2000). These modified self-molecules are perceived as foreign by the host, inducing a T and B cell response that may spread to include a response to the native (unmodified) self-protein. At very high concentrations of mercury a polyclonal B cell response is induced, but the basis for this is not known. Mercury-induced immune injury (hypersensitivity) has also been described in some humans, but it is infrequent and the mechanisms are less clearly defined than in the rodent models (Enestrom and Hultman, 1995; Griem and Gleichman, 1995; Pollard and Hultman, 1997).

It is theoretically possible that administration of ethyl mercury-containing vaccines along with vaccines not previously given at the same time (and in the same extremity/site) could result in a mercury-induced modification of a constituent of the coadministered vaccine, creating an antigenic epitope capable of cross-reaction with self epitopes that might activate T cells cross-reactive to self proteins. No experimental evidence is available to support this effect, however. Moreover, the concentrations of mercury required to induce autoimmunity in rodents (blood levels of mercury commonly >100 μg/ml) (Hultman and Nielsen, 2001) are markedly greater than those theoretically achievable through infant immunization, even before the mercury-containing preservative thimerosal was removed from routine childhood vaccines (Ball, 2001).

In the absence of experimental or human evidence regarding molecular mimicry or mercury-induced modification of any vaccine component to create an antigenic epitope capable of cross-reaction with self epitopes as a mechanism by which multiple immunizations under the U.S. infant immunization schedule could possibly influence an individual's risk of autoimmunity, the committee concludes that these mechanisms are only theoretical.

Bystander effects. Pathogenic microbes causing vaccine-preventable human diseases are acquired at mucosal surfaces, where they sufficiently evade the innate immunity of local mucosa so that they can replicate, locally or at other sites, before antigen-specific immune responses are induced. Most vaccines, however, are administered by injection. Vaccines are also complex biological products that may contain, in addition to the immunogenic microbial components, alum, stabilizers, small amounts of components carried forward from the preparation of the microbial antigens (e.g., bovine serum, egg or human cellular constituents), and, until recently, the preservative thimerosal. Theoretically, the mechanism of acquisition of vaccine antigens—by injection rather than the mucosal route—or the presence of alum or microbial constituents with adjuvant properties, could alter the quality or magnitude of the immune response compared with a wild-type infection and thereby favor the development of an autoimmune response.

Injection versus mucosal exposure. In principle, bypassing the mucosal surfaces through injection eliminates the mucosal immune response and induces solely systemic immunity. The most obvious consequence is that mucosal antibody responses leading to the production of secretory IgA are absent or less robust following vaccine injection than with wild-type infection (reviewed in Nagler-Anderson, 2001; Neutra et al., 2001). Also in principle, the circulation pattern of the memory T and B lymphocytes induced by an injected vaccine may also differ, as they would be unlikely to home to mucosal-associated lymphoid tissues. This effect does not appear to be critical to protection from disease for which vaccines are currently in use. It may, however, limit the extent to which mucosal colonization and transmission of pathogens are reduced, particularly for polio virus, which is acquired via the intestinal mucosa.

Shifting the primary site of antigen delivery from a mucosal surface to an extremity could, in theory, bypass immunoregulatory mechanisms, such as the production of immunoregulatory cytokines (transforming growth factor-β [TGF-β] or IL-10) and local regulatory T cells, that might modulate or attenuate the immune response and thereby impede the activation of potentially self-reactive cells capable of inducing autoimmunity (Maloy and Powrie, 2001; Neutra et al., 2001; Wills-Karp et al., 2001). There is, however, no experimental evidence that this occurs for current vaccines. Vaccines, in contrast to the infections they are designed to prevent, are instead a less potent stimulus of the immune response. (The immune response to tetanus and diphtheria toxins presented by their vaccines is an exception. Tetanus and diphtheria are diseases of toxin production rather than invasive infection; such minute amounts of toxin are produced during infection that the immune response is not effectively stimulated.) Thus the bystander effects of vaccines on the systemic immune responses, which in theory might induce autoimmunity, are less robust than those resulting from wild-type infection.

Coadministration of vaccines. Another theoretical possibility is that coadministration of multiple vaccines (particularly if given in the same extremity so that they would drain to the same regional lymph nodes) might produce an additive bystander effect, augmenting the magnitude or altering the quality of the response to the individual vaccine components. Such an effect is consistent with the findings by Ota et al. that BCG, a live mycobacterial vaccine, augmented the magnitude of the T cell response to other coadministered vaccines (Ota et al., 2002). This effect could also potentially increase the risk for activation of self-reactive T cells (Marchant et al., 1999; Vekemans et al., 2001). However, although BCG induced a potent Th1-type response to mycobacterial antigens, it promoted the production of both Th1- and Th2-type cytokines in response to unrelated vaccines. Thus, BCG is likely to impact immune response to unrelated antigens in early life through its influence on the maturation of dendritic cells, and not by shifting Th2 responses towards Th1 (Ota et al., 2002).

A reciprocal inhibitory effect of BCG vaccine on Th2 responses and allergic disease has been proposed on the basis of animal studies and ecological studies in humans (Aaby et al., 2000; Erb et al., 1998; Herz et al., 1998; Shirakawa et al., 1997), but the data regarding this effect in humans are conflicting (Alm et al., 1997; Gruber et al., 2001a; Strachan, 2000; Wills-Karp et al., 2001). In human infants in The Gambia, BCG induced a strong Th1 T cell response to mycobacterial antigens (Marchant et al., 1999; Vekemans et al., 2001) and enhanced the magnitude of the response to coadministered vaccine antigens (e.g., hepatitis B, tetanus toxoid), but the Th1-Th2 balance of the T cell response to these coadministered vaccines was not affected (Ota et al., 2002). Similarly, prior BCG immunization showed no effect on the development of IgE antibodies (an index of Th2 dependent responses) to tetanus and diphtheria toxoids following subsequent immunization with DT or DTP (Gruber et al., 2001b). Other vaccines that induce IFN-γ responses in human infants to the homologous disease antigens (i.e., live viral vaccines) (Arvin et al., 2001; Gans et al., 2001) are theoretically capable of enhancing a bystander Th1 response to heterologous antigens. The potency of these vaccines is almost certainly less than that of BCG, suggesting that if such an effect were to occur, it would likely be minor.

Th1 versus Th2 response. Autoimmunity is generally associated with Th1 responses, whereas allergy is associated with Th2 responses. DTwP vaccine includes whole formalin-inactivated B. pertussis cells, which contain endotoxin and bacterial cell membrane structures that activate the innate immune system and favor a Th1 response and production of IgG1 antibodies (but not IgG4 antibodies). The vaccine also contains inactivated pertussis toxin and alum, which are adjuvants that favor a Th2 response and production of IgE and IgG4 antibodies (reviewed in Gruber et al., 2001). Some human studies have been interpreted as showing that DTwP, and pertussis infection itself, induced the development of IgE antibodies to vaccine antigens and might have a similar effect on responses to environmental antigens, thereby predisposing to allergy and presumably impeding risk of autoimmunity (Farooqi and Hopkin, 1998; Nilsson et al., 1998; Odelram et al., 1994; Odent et al., 1994; Pershagen, 2000).

More recent studies suggest that the antibody responses to tetanus and diphtheria antigens in children given DTwP vaccine are similar for IgG but significantly lower for IgE and IgG4 (a correlate in humans of Th2 responses) when compared with responses in children given DT vaccine (Gruber et al., 2001b). There was no significant effect on IgE antibody response to environmental antigens, suggesting that DTwP may shift the Th1-Th2 balance modestly in the Th1 direction, but only for coadministered antigens (Gruber et al., 2001a). The relative contribution of alum to the Th1-Th2 balance in response to immunization with DT seems to be minor, as the amounts of IgE induced in those given DT with or without alum were similar in one study (Mark et al., 1995). It is important to note, such results do not necessarily mean this would apply to all of the other vaccines in which alum is employed as an adjuvant (e.g., hepatitis B, various conjugate vaccines).

These studies, in which the outcome measure was IgE antibodies, parallel the findings of T-cell responses in infants with pertussis infection or in those who have received DTwP: the T cell response is dominated by the Th1 cytokine interferon-γ, with very weak production of Th2 cytokines, such as IL-5. Conversely, DTaP, which contains acellular pertussis antigens, including inactivated pertussis toxin, but not the Th1-inducing components of B. pertussis whole cells, induces a mixed Th1-Th2 response. The Th2 response is more prominent and persistent in children with a family history of allergy (Ausiello et al., 1997; Rowe et al., 2001; Ryan et al., 1997a, 1997b, 1998).

Thus there is some evidence of a bystander effect associated with vaccines, but this effect is relatively modest, most evident with coadministered vaccine antigens rather than other environmental antigens or infections, and inconsistently shown. Current vaccines have, on balance, weak or no Th1-inducing activities. BCG appears to demonstrate the principle for co-administered antigens. However, BCG is not used in the United States, so the relevance for this mechanism in the effects of the U.S. recommended schedule is not demonstrated. Viral vaccines carry some potential for bystander activation, but likely would have a small effect, if it occurs at all. The data on DTaP vaccine indicates that Th1 dominance is not prominent. There is also no evidence in humans that vaccine antigens lead to the pathophysiological disease state. The limited evidence from humans that does exist regards surrogates of the disease process, that is, just some components of the events that would need to take place for the appearance of clinically relevant pathophysiology. Given the dominant Th1 nature of type 1 diabetes, MS, and most other autoimmune diseases, the prediction is that even if multiple immunizations had a cumulative bystander effect on potentially autoreactive T cells, the current vaccine program would be biased against the generation of Th1-dominated responses. And it is noted that this bias would be stronger now than at any time in the past.

The committee concludes that there is weak evidence for bystander activation, alone or in concert with molecular mimicry, as a mechanism by which multiple immunizations under the U.S. infant immunization schedule could possibly influence an individual's risk of autoimmunity. Moreover, the current routine childhood immunization schedule in the United States appears even less likely to act as an initiator or facilitator of autoimmunity than in the past.

Multiple Immunizations and Infectious Diseases That Protect from Autoimmunity

The mechanisms described above posit that immunizations play a direct role in the initiation or amplification of autoimmune processes. An alternative hypothesis is that immunizations increase the risk of autoimmunity by preventing infectious diseases that have protective effects. The theoretical deleterious effect might be specific to the infection prevented by the vaccine (a homologous effect), or result from a non-antigen-specific effect on the overall nature of the immune response (a heterologous effect). In either case, the notion is that a heretofore protective effect of infection has been lost by immunization.

Homologous effects. The nature of a homologous effect can be illustrated by changes in the epidemiology of poliomyelitis, although this example does not involve autoimmune disease or vaccine-induced effects. The poliomyelitis epidemics that commenced in developed countries near the end of the 19^th century are thought to have followed improvements in hygiene that postponed exposure to the virus beyond infancy (Plotkin and Orenstein, 1999; Wilson and Marcuse, 2001). Previously, wild-type poliovirus exposure most commonly occurred in infancy, at a time when passive maternally derived antibody provided partial protection. Thus, a child was first exposed under conditions in which acute paralytic disease was blocked but infection sufficient to immunize against disease on subsequent encounters resulted. Improved hygiene is believed to have reduced circulation of wild-type poliovirus sufficiently to delay exposure in many children until after passively acquired antibody was lost and they were fully susceptible. Further, reduced circulation of wild-type poliovirus would be predicted to result in lower levels of maternal antibody because of less frequent boosting, so that the magnitude of passive antibody transferred and duration of passive protection would be shorter (Zinkernagel, 2001). An immunization program might also be expected to leave infants more susceptible if vaccine-induced immunity did not induce protective levels of passive maternal antibody and herd immunity sufficient to reduce wild-type virus spread. This potential problem was overcome by actively immunizing infants.

A similar situation may apply to other vaccine-preventable diseases. In the case of measles, mumps, and rubella, the amount of passively-acquired antibody in infants born to mothers whose immunity is due to vaccination is less than and falls more quickly than in infants born to mothers whose immunity is due to wild-type virus infection. As a result, the age at which immunization induces protective antibody responses to measles is younger now than before the widespread use of MMR vaccine (Gans et al., 2001; Redd et al., 1999).

For loss of homologous protection that results from multiple immunizations to be a factor in autoimmunity, a vaccine-preventable infection must cause autoimmune disease. The relative risk will also be affected by the extent to which wild-type infection still occurs in the community; herd immunity will reduce the risk of infection and therefore increase the risk of autoimmune disease. Of the vaccine-preventable diseases, only congenital rubella, which has been noted as inducing type 1 diabetes in about 20 percent of affected individuals, has been causally linked with a chronic autoimmune disorder. Mumps virus infection has been linked to type 1 diabetes in rare cases (IOM, 1994), but causality has not been established. Even though waning of immunity following immunization may delay the age of onset of vaccine-preventable infections, and thus may be a factor in the chronic rubella viral arthritis in adolescent females, there is no evidence that waning immunity and delayed infection increases the potential for induction or acceleration of type 1 diabetes by wild-type rubella or mumps virus. Furthermore, the incidence of type 1 diabetes has increased most in children in the youngest age group (age 0–4 years) (Karvonen et al., 1999b; Podar et al., 2001), arguing against waning immunity as the basis for the increased incidence. The same theoretical considerations apply to the vaccine-preventable diseases that are capable of inducing acute neurological autoimmune injury, including ADEM and Guillain-Barré syndrome.

In the absence of experimental or human evidence regarding loss of protection against a homologous infection as a mechanism by which multiple immunizations under the U.S. infant immunization schedule could possibly influence an individual's risk of autoimmunity, the committee concludes that this mechanism is only theoretical.

Heterologous effects and the hygiene hypothesis. Possible heterologous effects of vaccination on autoimmunity can be considered in the context of the larger theoretical construct of the hygiene hypothesis, introduced earlier in the report. As noted previously, this hypothesis was first developed as a model to explain the rising increase in asthma and allergic diseases in the developed world (Strachan, 2000), and has been broadened recently to address the apparently parallel increase in certain autoimmune diseases, including type 1 diabetes and MS (Rook, 2000; Rook and Stanford, 1998; reviewed in Wills-Karp et al., 2001).

The hygiene hypothesis. The hygiene hypothesis is based on the notion that the human immune system (and that of other mammals) evolved in concert with constant exposure to a diverse and changing array of nonpathogenic environmental and commensal microbes, as well as under the persistent threat of lethal infectious diseases caused by microbial pathogens. This microbial exposure is thought to have conditioned the human immune system to respond vigorously to pathogenic microbes but not to harmless environmental antigens (allergens), normal microbial flora and environmental commensals, or self-antigens. Because birth is associated with a rapid switch from a sterile environment to a microbe-rich environment, to which the immune system must learn to respond properly, a component of the hypothesis is that the immune system may be particularly dependent on receiving appropriate conditioning through microbial exposure in early childhood.

During the past century, the developed world has seen improvements in hygiene, the development of effective immunizations, and the advent of antibiotics. These changes have altered the relationships between humans and microbes and, by inference, the challenges that the immune system must meet to provide protection from infection. The magnitude of the change in these relationships over the past century is arguably greater than the total change over all previous millennia of human existence.

The original hygiene hypothesis, developed as a model to explain the rising incidence of asthma and allergy, relied on the then recently elucidated Th1-Th2 paradigm that describes the ability of T cells to differentiate into cells with divergent effector functions. Th1 T cells produce interferon-γ and mediate or regulate cellular immunity and protection against viruses, bacteria, and invasive protozoans. Th2 T cells produce IL-4, IL-5, and IL-13, induce B cells to secrete IgE antibodies, and promote the development of eosinophils, which together mediate allergy and immunity to worms or helminthic parasites. As noted above, a number of microbial components stimulate cells of the innate immune system to produce IL-12 and other cytokines that cause T cells to differentiate into Th1 T cells and inhibit the development of Th2 T cells.

The logic of the hygiene hypothesis is that constant microbial exposure has a bystander effect that impedes the development of Th2 T cells and allergic responses to harmless environmental antigens such as foods and pollens. Therefore the reduced microbial exposure that infants now experience is associated with reduced constraints on the development of Th2 T cells. The hygiene hypothesis also fits well with the notion (Adkins, 2000; Prescott et al., 1998; Rowe et al., 2001; Siegrist, 2001), not yet firmly established as fact (Delespesse et al., 1998; Hassan and Reen, 2000; Lewis and Wilson, 2001; Marchant et al., 1999), that T cell responses of the human fetus and young infant are Th2-biased and gradually switch to a more balanced pattern over the first year or two of life in nonallergic children.

Although these mechanisms were consistent with observed increases in allergic diseases, they ran counter to two other observations. First, the incidence of autoimmune diseases, particularly type 1 diabetes and MS, appears to have increased in the same or overlapping populations. These diseases, however, are characterized primarily by Th1 T cell responses. Second, although worms or helminthic parasites induce robust Th2 and IgE responses, children in the developing world with large worm burdens have a lower incidence of asthma or allergic diseases (van den Biggelaar et al., 2000). Moreover, in certain mouse models intestinal worms can protect from allergic diseases (Wang et al., 2001), and conversely Th1 promoters can exacerbate allergic disease, depending on the context in which they are administered (Bryan et al., 2000; Hansen et al., 1999).

To address these apparent inconsistencies, modifications of the hygiene hypothesis posit that microbial exposure primarily acts not by deviating the immune response from Th2 to Th1, but by inducing the production of immunoregulatory cytokines (including IL-10 and TGF-β) and T cells that dampen the immune response broadly, including both Th1 and Th2 responses (Rook et al., 2000; Wills-Karp et al., 2001). Many microbes, including worms or helminthic parasites, bacteria, and viruses, induce IL-10 and/or TGF-β production (Letterio and Roberts, 1998; Moore et al., 2001; Rook et al., 2000; Wills-Karp et al., 2001). For example, increased IL-10 production in response to chronic parasitic infection with Schistosoma haematobium has been correlated with reduced evidence of allergic sensitization (van den Biggelaar et al., 2000). IL-10 and TGF-β can impede antigen-specific T cell responses directly, by impairing antigen presenting cell function, or by the induction of anergic or regulatory T cells. When stimulated via the T-cell receptor, regulatory T cells suppress the responses of other T cells in a nonspecific manner by contact-dependent mechanisms and by production of IL-10 or TGF-β. Mice that lack these regulatory cytokines or regulatory T cells develop inflammatory bowel disease or inflammatory or autoimmune disease in other tissues (Ermann and Fathman, 2001; Maloy and Powrie, 2001; Roncarolo and Levings, 2000; Rook et al., 2000; Shevach, 2000; Singh, 2000; Wills-Karp et al., 2001; Zhang et al., 2001).

Various findings suggest that that the human neonate can produce regulatory cytokines and T cells, supporting the notion that the neonate's immune system has the requisite immunoregulatory potential if the proper environmental signals are provided. Regulatory T cells are present and inducible in the blood of neonates (Roncarolo and Levings, 2000). Although the conditions required for their induction may differ somewhat from those in adults, it is not known at present if these differences are reproducible or biologically important. In addition, peripheral blood mononuclear cells and monocytes from the blood of human neonates can produce IL-10 and TGF-β in response to microbes or their components (reviewed in Lewis and Wilson, 2001). Although the magnitude of the response may be reduced compared with that in adults, the production of proinflammatory cytokines is also reduced in neonates, suggesting no imbalance in the production of pro- and anti-inflammatory cytokines.

Thus the extent, nature, and timing of contact with microbes are proposed to play an important role in establishing a proper balance in the immune response in early childhood and in maintaining this balance thereafter. A balanced immune response fosters the development of protective immune responses against pathogenic microbes, while preventing both a deleterious Th1 response to self antigens or harmless commensal microbes and a Th2-mediated allergic response to harmless environmental antigens. The type of microbial exposure that is important in establishing this balance is not currently known. Proposed candidates include various gut microbial commensals, chronic infections with intestinal worms or helminthic parasites or Mycobacterium tuberculosis, frequent exposure to environmental mycobacteria or other soil organisms, and the timing, number, and nature of acute infections. Factors such as breast feeding, number of siblings and birth order, day care attendance, contact with animals, antibiotic use, and the timing and nature of immunizations have been proposed to affect risk for autoimmune (or allergic) diseases through their effects on the extent and nature of microbial contact (Rook, 2001; Rook and Stanford, 1998; Singh, 2000; Strachan, 2000; Wills-Karp et al., 2001). Whether the sum of all microbial exposure, some specific combination of exposures, or one particular type of exposure is the important factor, or whether these are surrogates for an as yet undefined factor that is important, is uncertain.

Possible impact of vaccines on autoimmunity. On a numerical basis, vaccine-preventable infections represent a minute fraction of the overall infectious and microbial exposure in childhood. For immunization to have an impact on autoimmunity under the hygiene hypothesis, it would be necessary for one or more vaccine-preventable diseases to be particularly important for conditioning immunoregulatory immune responses. The gastrointestinal tract is proposed to play a particularly critical role in this process, so it would follow that immunizations that affect infection or colonization of the gut would be good candidates, but none of the childhood vaccines currently in use do so. (The rotavirus vaccine, which did affect the gut, was in use for too short a time to influence rates of autoimmunity.)

Data from animal models suggest that no one infection is likely to be key, but, rather, a global reduction in microbial contact could be a factor. For example, prior exposure to or infection with a variety of microbes can prevent type 1 diabetes in non-obese diabetic (NOD) mice, as can certain vaccines (reviewed in Bach, 2001; Bach and Chatenoud, 2001; Hiltunen et al., 1999; Singh, 2000). The role of infection is complex in the EAE model. The potentially protective effect of prior mycobacterial or B. pertussis exposure (Bach, 2001; Ben-Nun et al., 1993, 1997; Hempel et al., 1985; Mostaricka-Stojkovic et al., 1988) may be related directly to their use as adjuvants at the time of subsequent immunization with myelin proteins; thus the effect is not generalizable to autoimmune disease developing through more natural mechanisms, as in humans with MS. Nonetheless, if prior infection with one or both of these agents is assumed to be particularly important in establishing protection from autoimmune disease, the immunization schedule in the United States would have had no effect; use of BCG vaccine has never been recommended and, by analogy to the studies of EAE, administration of whole-cell B. pertussis vaccine—even as given in alum along with DT (Hempel et al., 1985)—should have been protective. It is possible that the acellular pertussis vaccines might lack the key components needed to provide protection against human autoimmune diseases, but even if this is so, the apparent increase in autoimmune disease began much earlier than the use of the acellular vaccine.

If immunoregulatory cytokines and regulatory T cells play an essential role in impeding the untoward inflammatory responses to normal microbial flora that result in autoimmunity and allergy, and their generation or function depends on microbial contact, it follows that the necessary microbial cues must be established early in postnatal life, in parallel with the development of effector T and B cell responses. Furthermore, such cues must either be persistent or sufficiently frequent to maintain these protective immunoregulatory mechanisms. Because these mechanisms have presumably been operative in all human populations for millennia and only recently perturbed, the associated microbial exposure must be both universally present and long established. None of the diseases prevented by the current U.S. immunization program meets those conditions, nor does tuberculosis or measles, the two candidates proposed from studies of heterologous vaccine-induced protection in Africa (Kristensen et al., 2000). Although tuberculosis may induce persistent infection, fulfilling one requirement, it has not been endemic worldwide until the relatively recent past, nor has measles (Cherry, 1998; Daniel et al., 1994). The more likely candidates are commensal bacteria and ubiquitous environmental microbes, the richness and diversity of which are reduced in hygienic urban environments, or the cumulative exposure to these nonpathogenic microbes and to various invasive infections rather than any specific infection.

The hygiene hypothesis is a model originally proposed based on epidemiological data. The biological mechanisms by which this model could explain an increase in incidence of autoimmune (or allergic) disease are substantial, and the biological evidence in support of the model is moderate to strong. However, the potential contribution of vaccine-preventable diseases as part of this model is minimal. In the absence of experimental or human evidence regarding mechanisms related to the hygiene hypothesis as a means by which multiple immunizations under the U.S. infant immunization schedule could possibly influence an individual's risk of autoimmunity, the committee concludes that this mechanism is only theoretical.

In theory, molecular mimicry, bystander activation, and impaired immunoregulatory mechanisms might act in an additive or synergistic manner to affect the risk of autoimmunity. Considering molecular mimicry, bystander activation, and impaired immunoregulation collectively rather than individually, the committee concludes that there is weak evidence for these mechanisms as means by which multiple immunizations under the U.S. infant immunization schedule could possibly influence an individual's risk of autoimmunity.

Bystander Effects

The elimination of smallpox vaccine in 1972 and the substitution of DTaP (containing acellular pertussis vaccine) for DTwP (containing formalin-inactivated B. pertussis whole cells) in the 1990s removed two vaccine-based sources of microbial signals favoring Th1 and opposing Th2 responses (Ausiello et al., 1997; Rowe et al., 2001; Ryan et al., 1998). In theory, the replacement of the oral polio vaccine with an injected vaccine given in alum, along with the addition of other vaccines given in alum and administered as early as birth, favors the development of Th2 responses relative to Th1 responses. In mouse neonates (which have a developmentally less mature immune system than that of the human neonate), such a Th2 bias has clearly been shown in response to antigens in alum compared with antigens administered with microbial adjuvants (Adkins, 2000; Barrios et al., 1996; Siegrist, 2001). Alum also induces IL-4 production from human mononuclear cells (Ulanova et al., 2001). However, except for evidence supporting a more Th2-directed immune response to components of DTaP than to DTwP (both of which are administered in alum), direct evidence is lacking that alum-containing vaccines deviate the immune response of human infants to environmental antigens toward Th2 responses. (For more details regarding this point, see the discussion of Th1 versus Th2 responses related to bystander effects of immunization and autoimmunity.) If such a deviation were to occur, whether it would be sufficient to result in clinically manifest allergy to these antigens would depend on other factors that are as yet incompletely elucidated.

Nonetheless, the biological mechanisms by which immunizations that contain microbial stimuli favor Th1 responses and immunizations containing alum favor Th2 responses are well established. Although the impact of immunization on heterologous allergic responses is unknown, on balance the current routine childhood immunization schedule in the United States is less likely to favor Th1 responses to heterologous antigens and more likely to favor Th2 responses.

The committee concludes that there is weak evidence for bystander activation as a mechanism by which multiple immunizations under the U.S. infant immunization schedule could possibly influence an individual's risk of allergy.

Prevention of Protective Infections: The Hygiene Hypothesis.

The hygiene hypothesis is discussed in detail in the above autoimmunity section on “Heterologous Effects and the Hygiene Hypothesis”. In the context of allergic diseases, either a shift of the Th1−Th2 balance, a loss of immunoregulatory mechanisms that block untoward immune responses to environmental antigens, or both, could result in an increase in allergy. All but the first of these mechanisms are compatible with a parallel increase in allergic and autoimmune diseases. The same reasoning applied to the potential role of vaccine-preventable diseases in reducing risk of autoimmunity can be applied to the question of allergy, with one modification. If a specific type of infection or microbial exposure impaired heterologous Th2 responses, even if it did not play an important role in the generation of immunoregulatory cytokines or regulatory T cells, a vaccine that prevented the disease but was not an effective surrogate for the infection could contribute to the increased incidence of allergy. It does not appear that this occurs. Tuberculosis and measles infection have been proposed as agents that impair heterologous Th2 responses, although, unlike tuberculosis which is a strong Th1-inducing microbe, measles virus is unlikely to have such an effect (Wills-Karp et al., 2001) and neither disease meets the requirement of having been a long-time, ubiquitous infection of all human populations. Although the evidence is conflicting, it has also been proposed that the vaccines against these two diseases are effective surrogates for the infections in the prevention of allergy.

The hygiene hypothesis is a theoretical model, originally proposed on the basis of epidemiological data. The biological mechanisms by which this model could explain an increase in incidence of allergic diseases are substantial, and the model is considered to be moderately to strongly plausible. However, the potential contribution of vaccine-preventable diseases as part of this theory is minimal. In the absence of experimental or human evidence regarding mechanisms related to the hygiene hypothesis as a means by which multiple immunizations under the U.S. infant immunization schedule could possibly influence an individual's risk of allergy, the committee concludes that this mechanism is only theoretical.

The committee concludes that there is weak evidence for the existence of any biological mechanisms, collectively or individually, by which multiple immunizations under the U.S. infant immunization schedule could possibly influence an individual's risk of allergy.

Carrier-Induced Epitope Suppression

This process was first described in model systems, but has become clinically important in the context of conjugate vaccines. In such systems, the carrier is a protein antigen, to which is conjugated (covalently linked to create a single molecule) a non-protein antigen. In clinical practice, conjugation of a bacterial non-protein antigen to a protein carrier has been used to convert a T cell-independent antigen (e.g., the type b capsular polysaccharide of Haemophilus influenzae or pneumococcal capsular polysaccharides) into a T cell-dependent antigen. Such conjugate vaccines are immunogenic in infants, inducing high-affinity IgG antibody and long-term immunological memory, whereas none of these features occurs when the polysaccharide alone is used as a vaccine. The basis for this is that the polysaccharide-protein conjugate will bind to and partially activate B cells that are specific for the polysaccharide, which internalize the conjugate, then process and present peptides from the protein component to CD4⁺ (helper) T cells specific for these peptides. This leads to activation of the T cells, that in turn help the B cells to produce high-affinity antibodies to the linked polysaccharide and mature into long-term memory B cells. In competition with these polysaccharide-specific B cells, are other B cells specific for the protein component of the conjugate. These B cells also present peptides to CD4⁺ T cells and in turn receive second signals allowing them to produce antibodies to the protein component of the conjugate vaccine. If the numbers of CD4⁺ T cells specific for the protein are limiting, then B cells specific for the polysaccharide component and B cells specific for the protein component of the conjugate are in competition with each other for limited numbers of CD4⁺ T cells capable of providing help (Insel, 1995). A variant of this can occur if multiple different carbohydrate antigens are conjugated to the same protein. In this case, the B cells specific for different carbohydrates may compete with each other for limited numbers of CD4⁻ T cells. The latter situation may account in part for reduced responses seen when multivalent pneumococcaltetanus toxoid conjugate vaccine was given along with H. influenzae type btetanus toxoid conjugate vaccine (Dagan et al., 1998).

Competition for Antigen Presentation

This describes a situation in which T cells responding to one antigen or infection compete with other T cells that are responding during the same time frame to other antigens or another infection, and one of the responses has a head start—it precedes the other by a few days or weeks. This gives the response to the earlier challenge a competitive advantage, such that it dominates and impedes the response to the delayed antigenic challenge or infection. Such competition is most readily observed in the context of strong CD8 T cell responses to viral infections (Chen, HD, 2001; Selin et al., 1998, 1999) or artificially manipulated immune responses (Kedl, 2000) in experimental animals. Bystander effects, including viral immune interference (see IOM, 2001a for more details) rather than competition for antigen presentation may affect responses to heterologous viral infections. An example of a heterologous effect in humans is the recent findings related to the timing of administration of MMR vaccine and varicella vaccine. If MMR and varicella vaccine are given at the same time or an interval of 30 or more days elapses between the administration of MMR and varicella vaccines, MMR and varicella vaccine efficacy is not compromised. However, if varicella vaccine is given after but within 30 days of MMR the relative risk for later development of breakthrough varicella, defined as cases of varicella that occur following exposure to wild-type virus >42 days after varicella vaccine administration, is 2.5 (1.3–4.9). The risk for breakthrough varicella is not increased when varicella vaccine is given within 30 days after DTP, Hib, OPV, IPV, or hepatitis B vaccines. These findings parallel those in earlier reports that had shown a reduction in responsiveness to smallpox vaccine following measles vaccine (MMWR, 2001c).

The committee concludes that there is strong evidence for the existence of biological mechanisms by which multiple immunizations under the U.S. infant immunization schedule could possibly influence an individual's risk for heterologous infections.