Randomized Controlled Trials

Brachman Study Brachman and colleagues (1962) conducted the only randomized, placebo-controlled trial of the efficacy of a protective antigen anthrax vaccine. Information on events of immediate onset following immunization is presented here; information on the efficacy of the vaccine is reported in Chapter 3. The vaccine studied was not AVA but was an earlier formulation produced from the R1-NP strain of anthrax (see Chapter 7 for more details). The study was carried out from January 1955 through March 1959 in four mills in the eastern United States that processed raw imported goat hair, which was commonly contaminated with anthrax spores.

The worker population eligible for the study included 1,249 men and women with no history of prior anthrax infection. The numbers of study participants were not reported by sex. Study participants received subcutaneous inoculations of 0.5 milliliters (ml) of either vaccine or placebo (0.1 percent alum). The first three doses were given at 2-week intervals, followed by three additional doses given at 6-month intervals and annual boosters thereafter.

TABLE 6-1

Ad Hoc Studies of Immediate-Onset Adverse Events Following Anthrax Vaccination: Local Events.

Employees at two mills were examined at 24 and 48 hours after inoculation for evidence of local and systemic reactions. Two measures of local adverse reactions were used: (1) an “erythema value,” based on the size of the area of erythema at the injection site, and (2) a “reaction index,” based on all objective findings of erythema, induration (hardness), and edema (swelling from accumulation of serous fluid). The reaction index ranged from 0 (no reaction) to 4 (marked reaction). Significant (3 to 4+) reactions were observed in about 2 to 15 percent of immunized persons after the first through the fourth inoculations, in approximately 40 percent of immunized persons after the fifth inoculation, and in 15 percent of immunized persons after the seventh inoculation. The most common local reactions—erythema, pruritis, and a small area of induration—were mild and disappeared within 24 to 48 hours. Overall, local reactions of any type, from mild to severe, were observed following 35 percent of inoculations.

Local reactions of greater severity included edema, an area of erythema of 25 square centimeters (cm²) or more, local tenderness and pruritis, and small painless nodules that persisted for several weeks. Severe, edema-producing reactions occurred following 2.8 percent of vaccinations, with those reactions peaking at the sixth inoculation and none occurring following the first inoculation. In three inoculees, extensive edema extended from the deltoid to the midforearm or wrist and resolved in 3 to 5 days. Systemic reactions were observed in two of the vaccine recipients who experienced edema. The systemic reactions consisted of “some malaise of 24 hours' duration.” Overall, 6 working days were lost as a result of the reactions of edema. In all, three placebo recipients experienced mild reactions, which were not further described. No information about sex differences in adverse effects was reported, nor was there any long-term follow-up of study participants.

The study by Brachman and colleagues (1962) has several strengths in estimating the frequency of reactions following receipt of anthrax vaccine. First, there was a placebo group against which the rates of reported reactions could be compared. Second, the characteristics of the vaccinated and placebo groups were likely to be comparable initially because of the randomized assignment of study participants. Third, recipients were directly monitored for the occurrence of adverse events following vaccination, reducing the possibility of reporting biases. Fourth, the criteria for determination of the presence and magnitudes of the reactions were explicitly defined. Unfortunately, events following vaccination were monitored in only two of the four mills and were monitored only for the first 48 hours following inoculation. The largest disadvantage of the study is its limited size.

Dose Reduction and Route Change Study A pilot study at the U.S. Army Medical Research Institute of Infectious Diseases (USAMRIID) on immune responses to alternative AVA dosing schedules and alternative routes of administration included active surveillance for adverse reactions (Pittman, 2001b; Pittman et al., 2002). In that study, 173 U.S. military and civilian volunteers (109 men and 64 women) were randomized to one of seven groups, defined on the basis of dosing schedule and route of administration. Three experimental dosing schedules were tested: a single injection on day 0, injections on days 0 and 14, and injections on days 0 and 28. For each experimental dosing schedule, two groups were established; one group was inoculated subcutaneously and the other group was inoculated intramuscularly. A control group was administered AVA by the licensed six-dose schedule and subcutaneous administration.

Study participants were evaluated clinically for local and systemic reactions at 30 minutes, 1, 2, and 3 days, 1 week, and 1 month after each vaccination. Local reactions at the injection site were common, with significant differences related to route of AVA administration and to sex. With subcutaneous vaccine administration, tenderness at the injection site was observed following the administration of 70 percent of the doses, erythema was observed following the administration of 36 percent of the doses, and induration was observed following the administration of 15 percent of the doses. With intramuscular administration, those effects were observed following the administration of 56, 6, and 2 percent of the doses, respectively. Subcutaneous nodules, observed following the administration of 38 percent of the subcutaneous doses, were not noted with intramuscular administration. With subcutaneous administration of one to three doses of AVA, some local reactions were significantly more common in women than in men. For example, erythema was observed in 63.4 percent of women and 22 percent of men, a subcutaneous nodule was observed in 63.4 percent of women and 24.2 percent of men, and local tenderness was observed in 84.5 percent of women and 62.9 percent of men.

The incidence of systemic reactions did not vary by route of vaccine administration or sex. For subcutaneous administration, the most common systemic reactions that occurred following administration of the first three doses were headache (9.9 percent of doses), malaise (9.4 percent), and myalgia (4.4 percent). No serious reaction attributable to the vaccine was observed with either subcutaneous or intramuscular administration.

The USAMRIID pilot study has the advantage of being a prospective randomized controlled trial, although it did not have a comparison group that received no active agent. In addition, the critical component of sex was included in the analysis. The study confirms a sex difference in local reactions that cannot be wholly attributable to a bias in reporting. However, it does not find a sex-related difference in systemic reaction rates that has been reported in other studies. The results of the study suggest that the approved subcutaneous route of vaccine administration may produce more local reactions than intramuscular administration. Given the significance of the route of administration, the number of participants in the study was relatively small, and larger studies should be instituted in the future. The Centers for Disease Control and Prevention (CDC) is planning a more extensive study to test the route of administration and the number of doses needed for protection.

Other Controlled Studies

ROTC Cadets Inadvertent administration of higher than recommended doses of anthrax vaccine to Reserve Officer Training Corps (ROTC) cadets provided an opportunity to compare their reports of acute effects with reports from cadets who received the recommended doses (Gunzenhauser et al., 2001). During the summer of 2000, 73 ROTC cadets were scheduled to begin the AVA series before deployment to Korea. Twenty-five cadets received 1.0 ml of vaccine as the initial dose, twice the recommended amount. The other 48 cadets received the standard dose of 0.5 ml. All the cadets received the standard dose of the vaccine for the second immunization.

Information on symptoms experienced by the cadets following vaccination was collected through a voluntary survey administered a few days after receipt of each of the two doses. The recipients of the double dose had been advised of the dosing error, and most of them completed the voluntary surveys after the receipt of both doses (25 and 18 responses, respectively). Of those who received only standard doses, 12 cadets completed the survey administered after the first dose, and 42 cadets completed the survey after the second dose.

Among the cadets who received the double dose, 92 percent reported a sore arm after the first injection, 88 percent reported a lump at the injection site, and 84 percent reported swelling. For cadets who received the standard dose, the reports for these symptoms after the first injection were 83, 42, and 42 percent, respectively. Fever was the only systemic effect reported after administration of the first dose, reported by 12 percent of those who received the double dose and 8 percent of those who received the standard dose. Decreased performance was reported by 28 percent of the cadets who received the double dose and 8 percent of those who received the standard dose. Following administration of the second vaccine dose, 44 percent of cadets who had received a double first dose reported three or more local symptoms, whereas 26 percent of those who received a standard first dose reported three or more local symptoms. In the group that received the double dose, 22 percent reported tiredness and 11 percent reported headache, whereas 7 and 5 percent of those in the standard-dose group reported these symptoms, respectively. Reports of fever (7 percent) and nausea (5 percent) came only from the standard-dose group.

This opportunistic study provides information on adverse events following the receipt of a double vaccine dose and standard vaccine doses. The increased dose produced local and systemic effects similar to those experienced after receipt of the standard dose, but a greater proportion of the recipients experienced the effects. In both groups, there were fewer reports of systemic effects (e.g., fever, tiredness, and headache) than local effects. No severe adverse events or hospitalizations occurred among either group of cadets after the administration of either dose. The number of subjects in the two study groups is small, however; and all ascertainment of health effects was through self-reports.

Uncontrolled Studies

Investigational New Drug Data The committee reviewed five annual reports submitted from 1967 through 1971 by CDC to the Division of Biologics Standards of the National Institutes of Health in support of the Investigational New Drug (IND) application required for licensure of AVA (CDC, 1967–1971). The committee received the information as the result of a Freedom of Information Act request made by an earlier Institute of Medicine (IOM) committee. The committee also received copies of unredacted progress reports from the files of the current vaccine manufacturer, BioPort.

Two vaccine formulations were administered under the IND application. The formulation originally developed at Fort Detrick, Maryland, by Wright and colleagues (1954) and manufactured by Merck, Sharp and Dohme was distributed only during the first reporting year. The other formulation was AVA, which was manufactured by the Michigan Department of Public Health by the methods described by Puziss and colleagues (1963). AVA was administered over the entire course of the study. Because distribution of the older Merck vaccine was discontinued, some study participants received both formulations over the course of their vaccinations.

The annual reports submitted in support of the IND application reflect experience from the administration of almost 16,000 doses of AVA to about 7,000 people in an observational study with no control subjects. The study was designed as an open-label, multicenter study that provided investigational vaccine to individuals at risk of exposure to anthrax at nine U.S. sites and one foreign site. Most of the study participants were textile workers with potential contact with contaminated goat hair or wool; a minority were laboratory personnel engaged in research or diagnostic procedures involving anthrax. Adverse events following vaccination were reported by the investigators administering the vaccines, who were required to record any reaction observed 48 hours after administration of the vaccine and to notify the National Communicable Disease Center (NCDC) of any severe reactions. A reporting form for each vaccinee was returned to NCDC following administration of the initial series of three doses or after the administration of a booster dose. It is not clear whether investigators examined or at least contacted vaccine recipients to ascertain reactions (active surveillance) or whether they relied on reports from recipients (passive surveillance).

Local reactions were classified as mild, moderate, or severe. A mild reaction was defined as an area of erythema (redness) only or of measurable edema or induration of ≤3 cm in diameter. A moderate reaction was defined as an area of edema or induration of >3 cm to <12 cm in diameter. A severe reaction was defined as any reaction measuring >12 cm in diameter or accompanied by marked limitation of arm motion or axiliary node tenderness. Mild reactions were reported following the administration of 8.4 percent of the AVA doses, moderate reactions following the administration of 0.9 percent of the AVA doses, and severe reactions following the administration of 0.2 percent of the AVA doses. Systemic reactions, including chills, fever, aches, and malaise, were reported for four vaccine recipients. All reactions were self-limited. Adverse event rates were not reported by sex, and no surveillance for later-onset effects was conducted.

This study suggests that AVA has a reaction profile comparable to that of other toxin-based vaccines such as tetanus toxoid. No investigations were reported of the mechanisms of the few systemic reactions that were noted. The reactogenicity of the vaccine appeared to vary by lot. The use of one lot manufactured by Merck was discontinued because it was more reactogenic than another available lot. Conclusions from the study are limited by its observational design. Furthermore, reports of adverse events are not available by sex or age, and in the early part of the study more than one vaccine formulation was administered.

Special Immunizations Program Safety Study Additional information on the immediate-onset reactions following administration of AVA is available from observations of inoculations given between 1973 and 1999 through the Special Immunizations Program at USAMRIID at Fort Detrick, Maryland (Pittman, 2001a; Pittman et al., 2001a,b). The study participants were those laboratory and maintenance workers who required access to the biocontainment facilities where Bacillus anthracis was studied.

The study population included 1,583 at-risk individuals, of whom 79 percent were men, 81 percent were aged 18 to 40, and 86 percent were of European origin. Over the reporting period, these individuals received 10,722 AVA doses, administered in accordance with the licensed schedule. The median number of doses received was six. A total of 273 (17.2 percent) study participants received 10 or more doses, and 46 (2.9 percent) participants received 20 or more doses. Most participants also received several other vaccines during the study period.

Surveillance for the occurrence of local and systemic adverse events was passive: the vaccine recipients self-reported adverse events if they sought treatment or if they believed the adverse event should be recorded in their health records. When participants came into the clinic to report adverse events, study staff evaluated and recorded the adverse events, classifying them as local or systemic.

Local reactions of some type were reported for 3.6 percent of the 10,722 doses of vaccine administered. Erythema and/or induration (E/I), the most common local reaction, was reported for 3.2 percent of the doses administered. Of the 1925 doses administered to women, E/I was reported following 7.3 percent. Men received 8,797 doses, of which 2.2 percent resulted in reports of E/I. Doses given to individuals ages 18 to 40 were followed by significantly more reports of reactions at the injection site than were doses given to those over age 40, but this result was not adjusted for sex. After adjustment for age and sex, reports of erythema, induration, and injection-site warmth were significantly higher for doses given to participants of European ancestry than for those given to African Americans.

Of the 32 vaccine lots used during the study, 50 or more doses were administered from only 19 lots. There was significant variation by lot in the incidence of injection-site reactions, ranging from 0 to 22.1 percent. Receipt of vaccine from the one lot that was found to have traces of squalene contamination did not produce an elevated rate of local reactions (5.3 percent of doses administered, which was the fifth highest rate of local reactions among the 19 lots). (See Chapter 4 for information on laboratory analysis for the presence of squalene in AVA.)

After controlling for vaccine lot and sex, reporting of E/I after the administration of one dose was found to be associated with an increased likelihood of reporting of E/I after the administration of the next dose (odds ratio [OR] = 13; 95 percent confidence interval [CI] = 8.7 to 21). The relative risks of a second reaction were 6.9 (95 percent CI = 4.3 to 10.9) for women and 14.3 (95 percent CI = 8.8 to 23) for men. However, a reaction to one dose was not a satisfactory predictor of a subsequent reaction for either women or men because most reactions occurred in persons who had received a previous dose with no resulting reaction.

Systemic reactions of any sort were reported following the administration of 101 of the 10,722 doses (1 percent) and were more frequent for doses administered to women than for doses administered to men. Overall, the most commonly reported reactions were headache (0.4 percent of doses), malaise (0.4 percent), myalgia (0.3 percent), nausea (0.1 percent), and dizziness (0.1 percent). For both women and men, those who reported a systemic reaction following the administration of one dose were more likely to report a systemic reaction following the administration of a subsequent dose. However, the majority of systemic reactions occurred after the administration of a dose to individuals who had already received vaccine doses with no systemic reaction. No sustained adverse events were noted, but long-term follow-up was not conducted.

Since the adverse events were self-reported and were not uniformly recorded for all who were vaccinated, this study may be most useful for the establishment of trends in subgroups. Absolute estimates of local or systemic reactions cannot be given. However, the study suggests that females and younger individuals more commonly report local reactions. The rates did vary by vaccine lot, but this may be confounded by secular trends in reporting reactions. The study also confirms that a prior reaction is predictive of a future reaction. However, the study cannot distinguish between elevated rates of reporting of adverse events versus an actual increase in the reaction rates.

Fort Bragg Booster Study Pittman and colleagues (1997, in press; Pittman, 2001c) conducted a study to assess the persistence of antibodies to B. anthracis 18 to 24 months after initial vaccination during Operations Desert Shield and Desert Storm and to assess the safety and immunogenicity of a vaccine booster dose. Study participants were recruited from active-duty personnel at Fort Bragg, North Carolina, in 1992 and 1994. All participants were volunteers and signed an informed-consent form to participate in the study.

The study population consisted of 495 male Operations Desert Shield and Desert Storm veterans who received one to three primary doses of AVA in 1990 or 1991. Only 5.5 percent received a single AVA dose; 70.5 percent received two doses, and 24.0 percent received three doses. Of the total, 91.3 percent received both AVA and pentavalent botulinum toxoid (PBT) in separate arms, and 8.7 percent received AVA only. For the booster study, participants who had originally received both vaccines also received (in separate arms) booster doses of both vaccines. Most participants (62 percent) were aged 30 to 39, and 92 percent were non-Hispanic Caucasians.

Adverse reactions to the booster vaccination were monitored through active surveillance. All study participants were assessed at 30 minutes after receiving the booster dose and subsequently on days 1, 2, 3, and 7 following vaccination. In addition, 86 percent of the participants returned after 24 to 36 days for an additional evaluation. Any individuals with reactions that were present at 24 to 36 days were monitored until the reactions resolved.

Because of the field setting of the study and the use of active surveillance, local reactions were common for both vaccines. Among the subjects who received only an AVA booster, an E/I reaction of <5 cm in diameter was observed in 27.9 percent, and an E/I reaction of 5 to 12 cm was observed in 4.7 percent. None of the study participants who received only an AVA booster had an E/I reaction of >12 cm in diameter. Among those who received both AVA and PBT, E/I reactions of these sizes were observed in 16.0, 9.3, and 0.7 percent of participants, respectively. Local reactions commonly occurred within 4 days of receipt of the booster dose.

Systemic reactions were also frequent following receipt of the vaccine booster dose, but the investigators noted that the study participants were also engaged daily in strenuous physical exercise, which could produce systemic reactions as well. The most common systemic reactions reported by those who received only an AVA booster dose were muscle ache (23.3 percent), headache (9.3 percent), malaise (7 percent), joint pain (7.0 percent), and fever (2.5 percent). In the much larger group of individuals who received both the AVA and the PBT booster doses, systemic reactions were reported by a greater percentage of study participants: muscle ache, 31 percent; rash, 17.3 percent; malaise, 16.8 percent; headache, 16.6 percent; joint pain, 13.1 percent; and fever, 2.7 percent. Anorexia (3.8 percent), nausea (3.5 percent), and breathing problems (0.2 percent) were reported only by study participants who received both booster doses simultaneously.

This study prospectively recorded the occurrence of local and systemic reactions following administration of an AVA booster to men previously primed with AVA vaccine. Because most vaccinees also received PBT, the results are difficult to interpret in terms of the effects specifically related to AVA. Those who received both vaccines simultaneously had more systemic reactions. Although fewer local reactions were noted in those who received both vaccines than in those who received AVA only, these reactions may have been recorded less often when systemic reactions were noted. The results do suggest that local and systemic reactions commonly occur within 4 days of vaccination but that these reactions are transient.

U.S. Forces in Korea Hoffman and colleagues (Hoffman et al., submitted for publication; CDC, 2000) analyzed information on adverse events following vaccination of U.S. forces in Korea with AVA between August 1998 and July 1999. The study was conducted at a single military clinic where a structured medical note was used as a survey instrument. In the study, 2,036 men and 495 women reporting to the clinic for vaccination with AVA completed a short questionnaire that requested information on sex, health status, vaccination history, and the occurrence of local or systemic reactions following the previous vaccination.

For both men and women, local reactions were common and were generally minor and did not lead to impairment of work performance. Women and anyone who had a reaction following receipt of a prior dose of AVA or who was taking medications were significantly more likely than their counterparts to report adverse events. Nodules and erythema were statistically more common in women. For example, reported rates of occurrence of nodules following receipt of one of the first three vaccine doses ranged from 49.8 to 62.4 percent for women, whereas the rates were 21.4 to 28.9 percent for men. Women also reported higher rates of localized itching, ranging from 20.4 to 37.0 percent, whereas the rates were 5.5 to 7.5 percent for men. Among those reporting a single reaction, men more consistently reported high rates of pain (12.4 to 16.9 percent) than women (2.9 to 5.0 percent). Decreased activity without a loss of time from work was the most common consequence of adverse events, reported by 6.7 percent of women and 5.7 percent of men following receipt of the third vaccine dose.

These data indicate that minor adverse events following the receipt of AVA are common and that rates are generally higher among women than among men. Since the service personnel reported adverse events at the next vaccine administration, however, there may have been some underreporting of adverse events and some selection bias in the reporting of those events. However, the results confirm those of other studies that women report more adverse events following vaccination with AVA but that they also report less pain than men.

Tripler Army Medical Center Survey The nature and frequency of adverse events following vaccination with AVA were assessed in a group of 601 military health care workers at Tripler Army Medical Center in Hawaii who began receiving vaccinations in September 1998 (AVIP, 2001; CDC, 2000; GAO, 1999; Wasserman, 2001). The study population included 416 men and 185 women, and the overall median age was 28 years. Enlisted personnel accounted for 71 percent of the group. Self-administered questionnaires were used to collect data on adverse events following each dose. Data on localized reactions were collected retrospectively for the first three doses and prospectively for subsequent doses. All data on systemic reactions were collected prospectively.

Local reactions were common and were generally highest following receipt of the first dose. The reactions included an area of redness, a lump or a knot at the injection site, and localized itching. Reports of local reactions were significantly higher for women than men, with an adjusted OR for any local reaction of 2.44 (95 percent CI = 2.04 to 2.93). Redness with a diameter of >5 cm was reported by 37 to 43 percent of women, whereas it was reported by 15 to 23 percent of men. Similarly, a lump or knot at the injection site was reported by 81 to 93 percent of women and 56 to 65 percent of men. Edema involving the lower arm was reported by 8 to 14 percent of women and 7 to 10 percent of men.

Muscle ache was the most commonly reported systemic reaction, reported following 45 percent of doses administered to women and 41 percent of doses administered to men. The study participants also reported whether their ability to perform their duties was affected and whether they had medical visits related to vaccination. For both indicators, reports were higher for women than for men. Four to 12 percent of women and 2 to 6 percent of men reported limitations in their ability to perform their duties. Four to 14 percent of women and 2 to 5 percent of men reported that they made an outpatient medical visit following vaccination with AVA.

This study monitored a group of subjects over the full series of six inoculations, but by the sixth dose about 50 percent of the original participants had been lost due to reassignment, separation from the military, or medical and other exemptions from vaccination. No control group was included, and it is uncertain how reports from medical personnel would compare with those from other military personnel.

9th Airlift Squadron Survey A survey of the members of the 9th Airlift Squadron at Dover Air Force Base in Delaware collected information on the systemic symptoms that they experienced following vaccination with AVA (Tanner, 2001). Vaccination of this squadron with AVA began in the autumn of 1998. In January 2000 a questionnaire was distributed by mail to all members of the squadron except administrative workers, a group that had not yet been vaccinated. Respondents were asked to provide information that included the number of vaccine doses they had received, whether they had experienced any of 16 specified systemic symptoms at any time since receiving their first vaccine dose, and whether they had sought treatment from the flight surgeon's clinic for symptoms or had lost time from duty. Respondents were also asked to describe their symptoms and any formal diagnosis that they had received.

Of 265 questionnaires mailed, 139 (52 percent) were completed and returned by vaccinated squadron members. Two other respondents had not been vaccinated, and so their responses were not included in the tabulations; 11 questionnaires were undeliverable. Responses were received from nine women, but the survey results were not reported separately for men and women. Joint or muscle pain was the most common symptom, reported by 42 percent of the 139 respondents. Other common symptoms included loss of energy or tiredness (31 percent), reduced concentration (27 percent), and short-term memory loss (26 percent). Reports of some systemic symptoms noted in other studies included itchy skin (21 percent), headaches (19 percent), and chills or fever immediately following vaccination (11 percent). Overall, 89 respondents (64 percent) reported one or more symptoms. Of these, 41 reported that they had sought treatment for one or more symptoms. The author noted that an unspecified number of untreated respondents added comments indicating that they had refrained from seeking treatment because of concerns about the quality of care available or the potential loss of flight status. In addition, 24 of the respondents with symptoms reported that they had missed more than 1 day of work; 10 had not returned to full duty.

This survey provides both quantitative and qualitative information on the self-reported systemic symptoms experienced by members of a single Air Force squadron who responded to the survey. An unofficial survey may elicit reports of symptoms from persons who would be reluctant to report those symptoms as part of an official DoD study or disease surveillance program. On the other hand, if persons with symptoms are more likely to respond to the survey, the rates of occurrence of symptoms after vaccination with AVA will be overestimated. In addition, no information is provided about the timing of these self-reported symptoms in relation to the time of receipt of AVA. It is difficult to interpret the results of a survey that asks for the occurrence of events “at any time” since receiving AVA. The author acknowledges that the low response rate and the lack of a control group are important limitations of the study.

Any Health Outcomes: Uncontrolled Studies

Multiple Vaccines Studies Three published studies (Peeler et al., 1958, 1965; White et al., 1974) provide longitudinal information on a population of skilled laborers and laboratory workers at Fort Detrick, Maryland. The members of this study population received multiple doses of many different vaccines, including an anthrax vaccine, because of the potential for occupational exposure to virulent microorganisms. The 99 white male workers in the initial cohort (Peeler et al., 1958) were selected for the study because they had the longest and most intensive vaccination histories among the 700 employees receiving a continuing schedule of multiple immunizations. The workers began receiving vaccinations in 1944 and were evaluated for these studies in 1956, 1962, and 1971, respectively. Both acute and more persistent changes in health status following vaccination were ascertained through complete medical histories, physical examinations, laboratory tests, and reviews of outpatient and immunization records.

An earlier IOM report (IOM, 2000) described the three studies in some detail. The well-studied cohort showed no evidence of illnesses attributable to intensive immunization over a 25-year period or any serious, unexplained illnesses. Reaction rates for immediate-onset events, whether local or systemic, were not reported for any individual vaccine. The concerns over the high incidence of abnormal liver function test results, lymphocytosis, and abnormalities observed by electrophoresis of serum raised in the two earlier studies (Peeler et al., 1958, 1965) were increasingly dispelled by the third report (White et al., 1974). The laboratory changes were often reversible, as shown in the second study (Peeler et al., 1965), and in the third study (White et al., 1974), conducted 10 months after the termination of the immunization program, laboratory values had normalized. The serum abnormalities seen earlier are probably explainable by an increased level of fast mobility gamma globulins. The overall mortality rate for the cohort was within the expected range.

In general, data from these studies do not suggest that repeated exposure to AVA along with other vaccines is associated with later-onset health effects, but the studies are of limited value for evaluating the safety of AVA. Because the cohort was exposed to many vaccines in addition to an anthrax vaccine, indications of any deleterious effects could have been due to any of the vaccines received. In addition, during the time frame covered by these studies, workers received both earlier anthrax vaccine products and the currently licensed AVA product. Although the number of anthrax vaccine recipients increased over the study period from 28 to at least 76, without apparent coincident changes in the health status of the group, the small size of the cohort provides little statistical power to detect increased risk of illness.

Reproductive Outcomes: Controlled Study

Pregnancy, Births, and Birth Outcomes Wiesen and colleagues (Wiesen, 2001a,b; Wiesen and Littell, 2001) compared the reproductive experiences of vaccinated and unvaccinated women on active duty in the U.S. Army. The study population consisted of 4,092 women, ages 17 to 44 years, who had been assigned to Fort Stewart or Hunter Army Airfield in Georgia between January 1999 and March 2000. Of this group, 3,135 received at least one dose of AVA; 962 were unvaccinated. The vaccinated and unvaccinated women were similar in age and marital status. A similar percentage of women in each group was African American, but a smaller percentage of the vaccinated group was white (28.8 versus 36.4 percent) and a larger percentage was of another race (19.7 versus 12.9 percent).

The analysis compared the two groups in terms of pregnancy rates, the proportion of pregnancies resulting in live births, and the incidence of adverse birth outcomes. The size of the study population provided an 80 percent power to detect a 20 percent decline in the pregnancy rate. Pregnancies, births, and birth outcomes were determined from a review of medical records at Fort Stewart. For 54 of 85 women who left Fort Stewart during their pregnancies, records from the Defense Enrollment Eligibility Reporting System (DEERS) could be used to determine if a birth occurred. Vaccination against anthrax had no effect on pregnancy rates (OR = 0.9, 95 percent CI = 0.7 to 1.1), with or without adjustment for marital status, race, and age. There also was no significant difference between vaccinated and unvaccinated women in terms of the proportion of pregnancies that resulted in a live birth (adjusted OR = 0.8, 95 percent CI = 0.7 to 1.0). In addition, the adjusted ORs for low birth weight (OR = 1.3, 95 percent CI = 0.2 to 6.4) and structural abnormalities of cosmetic or surgical significance (OR = 0.7, 95 percent CI = 0.2 to 2.3) showed that there was no statistically significant elevation of risk for the infants of vaccinated women.

The strengths of this study include the size of the study population and the resulting statistical power to detect changes in pregnancy rates. Also, the retrospective ascertainment of outcomes from a source independent of exposure reduces potential observer and follow-up biases. However, information on certain outcomes was obtained from two sources: medical records and DEERS. It may have been better to use only information from DEERS to ascertain the outcome and then verify agreement with the medical record, where available. The study may be interpreted as providing some assurance that vaccination with AVA has no major adverse effect on reproductive potential or reproductive outcomes. However, because specific birth defects occur at low rates and different birth defects can have different causes, failure to detect an excess overall rate of birth defects does not necessarily rule out an elevation of risk for a specific, although rare, birth defect. In addition, the power of the study to detect meaningful decreases in birth rates or birth weights associated with maternal AVA exposure is unknown. The study does not directly address the risks of vaccination during pregnancy. Women known to be pregnant were exempted from vaccination, but some women with early unconfirmed pregnancies may have been vaccinated. Thus, more definitive conclusions regarding pregnancy outcomes will depend on additional study.

Air Combat Command Study

Rehme and colleagues (Rehme, 2001; Rehme et al., 2002) performed an opportunistic retrospective cohort study in which they compared the postdeployment use of Air Force ambulatory health care services by AVA-vaccinated and unvaccinated Air Force personnel following their return from Southwest Asia (SWA). The study population was identified by linking Air Force records on visits to a medical treatment facility in SWA between January 1, 1998, and September 10, 1998, with DoD records on vaccination against anthrax. Of the personnel with records of a visit to a medical treatment facility in SWA, 4,045 persons had a record of the receipt of at least one dose of AVA and 1,133 persons had no record of the receipt of any doses of AVA. Men accounted for 84 percent of the vaccinated group and 85 percent of the unvaccinated group.

No difference in the rate of postdeployment use of Air Force ambulatory health care services within the 6 months following vaccination with AVA was found between the vaccinated and the unvaccinated personnel. Women were more likely than men to have had a postdeployment ambulatory health care visit (relative risk 1.38, 95 percent CI = 1.30 to 1.46), but there was no difference between vaccinated and unvaccinated men (relative risk = 0.95, 95 percent CI = 0.88 to 1.02) or between vaccinated and unvaccinated women (relative risk = 0.99, 95 percent CI = 0.88 to 1.12). The analysis also examined ambulatory health care visits in terms of 17 broad ICD-9-CM diagnostic categories as well as 16 specific diagnoses, including diabetes, thyroid disorders, anemia, and headache. For vaccinated personnel, the risks of any of these diagnoses were comparable to or lower than those for unvaccinated personnel.

The uniformly negative results support the general conclusion that AVA does not lead to increased use of health care services by DoD personnel, and the inclusion of only deployed personnel in the study attempts to control for a possible “healthy soldier” effect. This is analogous to the “healthy worker” effect observed in studies of occupational groups compared with the general population.

Nevertheless, the study has important limitations. First, all personnel entered into this study must have visited an ambulatory health care setting at least once during their deployment, and according to the investigators, this eliminated more than 50 percent of the deployed personnel. This eligibility criterion might have the effect of masking an adverse effect of AVA if those not vaccinated experienced other health conditions that made them more likely to have used health care services. The uniformity of the findings, however, argues against (although does not rule out) this potential bias.

Second, there is no control for whether Air Force personnel sought ambulatory care in SWA or in the United States from sources other than Air Force treatment facilities. If personnel were more likely to receive AVA if they were to be deployed for longer periods in SWA and if medical care was less available in SWA, the results may mask a real effect of vaccination with AVA. Again, the uniformity of the results argues against (but does not rule out) this potential bias.

Third, if personnel who received AVA had medical problems that resulted in an early discharge from the military, they would not have been included in the analysis. The investigators clearly note this possibility but suggest that the time required to process a medical discharge would limit its effect. The available data offer no basis for evaluation of this possibility.

Finally, it is not known how service personnel were selected to receive AVA (perhaps it was random, but the committee does not know this).

U.S. Army Aviation Epidemiology Data Register

A study of U.S. Army aircrew members was conducted to assess clinical outcomes reflecting vaccine-associated adverse events among those who had been vaccinated with AVA (Mason et al., 2001, submitted for publication). The study used record linkage to assess those who were vaccinated with AVA and those who were not and to identify clinical outcomes. A total of 3,356 AVA-vaccinated aircrew personnel were matched to an equal number of unvaccinated personnel by age, sex, race, class of flying duties (aviator, flight traffic controller, flight surgeon), and type of service (active, reserve). Changes in medical condition were determined by comparison of information from medical examinations conducted before and after the vaccination date for the AVA-exposed individual in the pair. The clinical outcomes evaluated included weight change, an increase in blood pressure, anemia, increased intraocular pressure, stereopsis, hearing loss, vision loss, proteinuria or glycosuria, and increased fasting blood sugar levels. No outcome showed any positive association with receipt of the AVA vaccine. In fact, those vaccinated with AVA actually showed less weight loss and vision loss than those not vaccinated.

The apparent reduction in risk may be attributable to a healthy soldier effect. Those vaccinated may have been healthier than those not vaccinated, thus producing an apparent but spurious negative association with administration of AVA. In addition, the average time between examinations may have been different for the personnel vaccinated with AVA and those not vaccinated with AVA. Nonetheless, the results do not show any elevation in adverse outcomes due to AVA vaccination.

Naval Health Research Center DoD-Wide Surveillance of Hospitalizations

Sato and colleagues (2001a,b; Sato et al., 2001) used data from DoD databases on hospitalizations of active-duty military personnel to determine whether those who had received AVA had an excess of hospitalizations within 42 days of AVA vaccination compared with those who had not received AVA. The study included all U.S. military service personnel on active duty during the analysis period, which extended from January 1, 1998, through March 31, 2000. Hospitalizations in both military and nonmilitary facilities were ascertained. Hospitalization data were linked with the vaccination, demographic, and personnel data in DoD records. All personnel received other standard vaccinations.

The analysis was based on 2,651 hospitalizations and 120,870 person-years of observation in the group vaccinated with AVA (the vaccinated group) and 151,609 hospitalizations and more than 2.3 million person-years in the group not vaccinated with AVA (the unvaccinated group). For the vaccinated group, the risk interval was counted from the date of AVA vaccination until either the date of the first hospital admission, the date of receipt of the next dose of AVA, or the end of the analysis period (March 31, 2000). The hospitalization rates for this group were calculated per person-years of observation within 42 days of receipt of any AVA dose. For the unvaccinated group, the risk interval started January 1, 1998, and extended to the date of the first hospital admission, the date of separation from the military, or March 31, 2000. Hospitalizations were assigned to 14 major ICD-9-CM categories of disease on the basis of discharge diagnoses. Subsequent hospitalizations for the same condition were not counted for either group.

Relative risks for the vaccinated group versus the unvaccinated group were adjusted for age (in quartiles), sex, number of hospitalizations in the previous year, marital status, race/ethnicity, pay grade, duty occupation category, branch of service, and number of days deployed (in quartiles). Analyses were done separately for men and women. No adjustments for multiple comparisons were made in the interim analysis presented to the committee.

Vaccinated men and women had significantly lower relative risks than unvaccinated personnel for hospitalizations for each of the 14 broad ICD-9-CM categories examined. Adjusted relative risks for women ranged from a low of 0.18 for diseases of the blood to a high of 0.66 for neoplasms. For men the lowest relative risk was 0.30 for mental conditions, and the highest relative risk was 0.75 for diseases of the digestive system.

In these analyses, receipt of AVA was not associated with a significant increase in the risk of hospitalization within 42 days of vaccination for 14 major groups of disease, and, in fact, vaccinated personnel were hospitalized significantly less than unvaccinated personnel. The relative risks for hospitalization (the group vaccinated with AVA versus the group not vaccinated with AVA) were very low, with the relative risks for most personnel falling between 0.2 and 0.6. Relative risks less than 1.0 would be expected if service personnel who received AVA were also more likely to be deployed and deployment is associated with a healthy soldier effect. The possibility thus cannot be discounted that differences in the risk of hospitalization between personnel who received AVA and personnel who did not reflect differences in deployment status. Although analyses were statistically adjusted for quartiles of number of days deployed, this approach may not have been adequate to fully control for differences between deployed and undeployed military personnel in their underlying health status or in the manner in which health-related issues are addressed for predeployment personnel. Also affecting the interpretation of the current results is the possibility that the disease categories used may be too broad to detect increases in risks of individual diseases in the group vaccinated with AVA.

AMSA Analyses of DMSS Data Regarding Health Outcomes Following Vaccination Against Anthrax

In 2000 and 2001, AMSA prepared several reports that described analyses that were carried out with data available from DMSS to assess whether inpatient or outpatient medical visits are associated with vaccination with AVA. AMSA also carried out analyses in response to specific questions raised by the IOM committee. As a result, several different approaches to the analyses of the data available from DMSS were taken over the course of the committee's work. Each is described separately.

Screening Analyses In 2001, AMSA began a process of regularly using DMSS data for screening purposes. It has since produced two quarterly reports describing screening analyses of data available from DMSS and DoD's electronic immunization tracking system database (AMSA, 2001a,b). The databases were used to compare rates of hospitalization and outpatient visits between military personnel who had and those had not been vaccinated against anthrax on the basis of 14 major disease categories and 824 specific diagnoses (identified on the basis of three-digit ICD-9-CM codes). A third analysis on incident visits (first visits for a diagnosis) to inpatient or outpatient facilities was also conducted. For the April 2001 quarterly report on data for January 1998 to December 2000, a total of 757,540 person-years of observation for the group that had received AVA and 3,430,459 person-years of observation for the group that had not received AVA were included in the analyses.

The analyses found that, for all major diagnostic categories, crude and adjusted rates of hospitalization and of outpatient visits and incident visits (incident visits include inpatient and outpatient visits combined) were lower in the group that received AVA than in the cohort that did not. For specific diagnoses within each database (hospitalization, outpatient, and incident data; a total of 2,472 comparisons), however, the rates of some diagnoses were statistically significantly higher for the group that received AVA than for the group that did not. In many cases, these diagnoses (e.g., malaria, wounds, and trauma) were ones that are expected to occur at higher rates in service members deployed overseas than in those remaining in the United States. Since personnel receiving the anthrax vaccine were those most likely to be deployed to areas where risks of exposure to infectious disease are higher, these statistical associations do not raise questions for further analysis. Statistically significant elevations in rates for outpatient visits were also found for certain malignant neoplasms, portal vein thrombosis, and acute pulmonary heart disease, among others. These statistical associations can raise hypotheses to be tested further in additional analyses, such as those described in the sections that follow to try to account for the healthy soldier effect. AMSA plans to continue these screening analyses as additional data accrue.

Hypothesis Testing Analyses AMSA also presented data to the committee to address specific concerns that had been raised regarding AVA (Lange et al., 2001a). As described above, the analyses compared rates of hospitalization and of outpatient visits for selected conditions among active-duty personnel who received one or more doses of AVA with the rates among those who had not yet been given AVA or who had never received AVA. Rate ratios were adjusted for differences between AVA recipients and AVA nonrecipients in terms of age, sex, rank, deployment, service, ethnicity, previous hospitalizations, calendar year, and occupation. Separate analyses for men and women were also done. Both the group that had received AVA and the group that had not received AVA could have received other types of vaccines.

Rates were calculated for the interval from January 1998 to June 2000 and included 515,389 person-years of observation for the group vaccinated with AVA and 2,873,751 person-years of observation for the group not vaccinated with AVA. The 12 inpatient and the 14 outpatient diagnoses selected for comparison were those for which concern in relation to AVA exposure had been publicly expressed or those that have been investigated in association with other vaccines. Inpatient conditions included arthropathies, asthma, connective tissue diseases, diabetes mellitus, Guillain-Barré syndrome, cardiac dysrhythmias, multiple sclerosis, thyroid disorders, and lymphatic cancers. Outpatient conditions included circulatory problems; endocrine or immunological conditions; genitourinary problems; connective tissue diseases; ill-defined conditions; and respiratory, skin, and nervous system diseases.

For each of the diagnostic categories examined, both the unadjusted and the adjusted rate ratios for hospitalization or outpatient visit rates for the group that received AVA compared with those for the group that did not receive AVA did not differ significantly from 1.0 (the ratio observed when the rates are equal). The rate ratios were less than 1.0 for nearly all of the diagnoses examined (ranges, 0.67 to 1.11 for hospitalizations and 0.68 to 0.84 for outpatient conditions), indicating lower hospitalization and outpatient visit rates in the group that received AVA than in the group that did not receive AVA. Lower rates in the group that received AVA were observed for all personnel combined and for the separate analyses among male and female soldiers.

These data indicate that there was no excess risk of selected adverse health events that required either hospitalization or an outpatient visit among active-duty military personnel receiving AVA over a 2.5-year period. In fact, the group that received AVA tended to have fewer hospitalizations or outpatient visits than the group that did not receive AVA.

Inferences about the safety of AVA based on these hypothesis-testing data are limited for several reasons. First, only selected diagnoses were examined, and thus the analyses do not address all possible risks. In addition, many of the diagnostic categories subsumed multiple medical conditions. Thus, risks associated with specific conditions within these categories might have been missed. Although deployment status was included as a covariate in the adjusted rate ratio analyses, this approach may not have been sufficient to account for the many differences in health status and reporting biases for those who are eligible for deployment and those who are not eligible for deployment.

Subsequent Analyses to Address the Healthy Soldier Effect To address concerns about inherent health-related differences in personnel who did and did not receive AVA because of deployment and to examine a wider range of diagnoses, in response to the committee's request, a second set of analyses were performed with the DMSS data (AMSA, 2001c). Again, several approaches were used, and in most of these analyses, service members served as their own controls. Tables are found in Appendix G.

Postimmunization Versus Preimmunization Analyses: Overall Analyses In the first analysis, the hospitalization rates in the time period after the receipt of one or more doses of AVA were compared with the rates in the period before the receipt of AVA for the population of service members who had received at least one dose of AVA. The analyses included the active-duty personnel who had received one or more doses of AVA between January 1, 1998, and December 31, 2000. Pre- and postimmunization cohorts were established on the basis of each individual's daily immunization status during that time frame. Therefore each individual could contribute a different amount of preimmunization time depending upon his or her time in the military prior to receiving AVA. Rate ratios (the rate after vaccination with AVA versus the rate before vaccination with AVA) were calculated for hospitalizations for 843 specific diagnoses (identified on the basis of three-digit ICD-9-CM codes) and were adjusted by Poisson regression methods for up to 11 covariates. Ratios were calculated only for diagnoses with at least five hospitalizations in each comparison group.

The results of these analyses were based on 11,436 hospitalizations during 478,093 person-years of observation in the preimmunization time period (crude rate, 23.92 per 1,000 person-years) and 21,436 hospitalizations during 738,382 years of observation in the postimmunization time period (crude rate, 29.03 per 1,000 person-years). The unadjusted overall rate ratio (the rate after vaccination with AVA versus the rate before vaccination with AVA) for hospitalization was 1.21. Hospitalization rates in the period after vaccination with AVA were higher than those in the period before vaccination for about one-half (414 of 843) of the diagnoses and were lower than those in the period before vaccination for the others. Of the conditions with rate ratios significantly different from 1.0, hospitalization rates in the period after vaccination were statistically significantly elevated for 15 conditions (see Appendix G, Table G-1) and were statistically significantly reduced for 12 conditions. One would have expected rates for about 42 diagnoses to be significantly different in the intervals before and after vaccination with AVA just by chance, given the large number of conditions examined. The significantly elevated rate ratios ranged from 1.31 (95 percent CI = 1.04 to 1.65) for inguinal hernia (ICD-9-CM code 550) to 5.14 (95 percent CI = 1.81 to 14.57) for carcinoma in situ of the breast and genitourinary system (ICD-9-CM code 233). The rate of hospitalization for diabetes mellitus was increased 3.46-fold (95 percent CI = 1.51 to 7.90) in the interval after vaccination with AVA.

Comparison of rates of hospitalization in the same individual before and after the receipt of AVA removes many of the biases inherent in comparing groups vaccinated with AVA and groups not vaccinated with AVA. However, one limitation of comparisons based on a single individual is that for very serious medical conditions (e.g., aplastic anemia or multiple sclerosis) the interval before vaccination with AVA will by definition have few or no events, since if such events had occurred, the soldier would likely never have been eligible to receive AVA.

Similarly, for a diagnosis generally made on an outpatient basis, such as diabetes, it is possible for the rate before vaccination with AVA to be artificially and differentially lower since those who had the disease and who had been hospitalized for it would be less likely to be deployed and therefore less likely to be vaccinated. A normal rate of hospitalization for the disease after vaccination would then appear to be an increase over the rate before vaccination, thus explaining the higher rate after vaccination with AVA without indicating that the vaccine caused the problem (particularly in the instance when that rate after vaccination with AVA remains below the expected rate for the population). In other words, the frequency of diabetes after receipt of AVA may appear to be elevated only because the rate in the time period before vaccination is especially low due to the healthy soldier effect. Whether this phenomenon explains the apparent higher risk after vaccination with AVA can be determined by comparing the rate before vaccination with AVA with the rate in those who never received AVA. If the rate before vaccination with AVA is significantly lower than that in those who were never vaccinated (as it is in the case of diabetes), it supports the conclusion that there is no increased risk attributable to AVA.

Postimmunization Versus Preimmunization Analyses by Time Window A second, similar analysis compared hospitalization rates for the same individuals for the period before immunization with AVA and two time periods after immunization: 0 to 45 days and more than 45 days. This analysis was intended to determine whether any excess risks following exposure to AVA might have been obscured in the previous analysis, which used a longer, open-ended postvaccination time frame. The approach to the analysis was the same as that described above, except that the period after immunization was divided into two time intervals. The unadjusted overall hospitalization rate ratio for the first time interval (0 to 45 days postvaccination versus prevaccination) was 1.08 (25.81 versus 23.92 per 1,000 person-years) and that for the second time interval was 1.25 (29.96 versus 23.92 per 1,000 person-years). Compared with the hospitalization rates before receipt of AVA, rates of hospitalization within 45 days of being given AVA were significantly greater than 1.0 for 13 of the 843 diagnoses examined (Appendix G, Table G-2) and significantly less than 1.0 for 7 diagnoses. Diagnoses for which adjusted rate ratios were statistically significantly greater than 1.0 included diabetes mellitus (adjusted rate ratio = 3.49, 95 percent CI = 1.39 to 8.79) and other disorders of the intestine (ICD-9-CM code 569; adjusted rate ratio = 4.16, 95 percent CI = 1.51 to 11.49). Most of the significantly elevated rate ratios in the first time period were associated with nonspecific diagnostic categories, such as other and unspecified disorders of the back (ICD-9-CM code 724). Given the number of diagnoses examined, significantly elevated rate ratios would have been expected for approximately 42 diagnostic categories just by chance.

In the second time interval (>45 days after vaccination with AVA), adjusted rate ratios for hospitalization were significantly greater than 1.0 for 20 of the diagnoses examined (Appendix G, Table G-2), including ratios of 3.44 (95 percent CI = 1.47 to 8.06) for diabetes mellitus and 2.61 (95 percent CI = 1.06 to 6.44) for other disorders of the intestine (ICD-9-CM code 569). Adjusted rate ratios significantly less than 1.0 were observed for 10 of the 843 diagnoses examined. No consistent pattern was observed when rate ratios for the first interval (0 to 45 days postimmunization) were compared with those for the second interval (>45 days postimmunization). That is, ratios were not uniformly either larger or smaller in the first interval than they were in the second interval. This suggests that any excess risks following vaccination with AVA did not aggregate in the immediate period after vaccination, nor were excess risks specifically identified for conditions that may take some time to develop and be recognized, resulting in hospitalization in the later interval after vaccination.

Postimmunization Versus Preimmunization Analyses by Dose A third analysis among persons ultimately vaccinated with AVA compared rates of hospitalization before receipt of AVA and after receipt of either one to three doses or four or more doses of vaccine. This analysis was designed to determine whether there is a dose-response effect between the amount of exposure to AVA and the risk of hospitalization for specific diseases. Results were based on 11,436 hospitalizations during 478,093 person-years of observation in the preimmunization time period (crude rate, 23.92 per 1,000 person-years), 5,832 hospitalizations during 184,273 years of observation in the cohort that received one to three doses of AVA (crude rate, 31.65 per 1,000 person-years), and 15,604 hospitalizations during 554,109 person-years of observation (crude rate, 28.16 per 1,000 person-years) in the cohort that received four or more doses of AVA. The ratios of crude overall hospitalization rates in the postimmunization time period compared with the hospitalization rates in the preimmunization time period were 1.32 for the cohort that received one to three doses and 1.18 for the cohort that received four or more doses.

In the cohort that received one to three doses, hospitalization rates for 23 diagnoses were significantly higher than the rates before vaccination with AVA (Appendix G, Table G-3), and those for 5 diagnoses were significantly lower than the rates before vaccination with AVA. Conditions with significantly elevated adjusted rate ratios included diabetes mellitus (rate ratio = 4.98, 95 percent CI = 2.02 to 12.25), asthma (rate ratio = 2.18, 95 percent CI = 1.37 to 3.47), and regional enteritis (rate ratio = 4.90, 95 percent CI = 1.55 to 15.44). In those who received four or more doses, the hospitalization rates were significantly elevated for 13 diagnoses (Appendix G, Table G-4) and were significantly reduced for 10 diagnoses compared with the rates in the prevaccination time period. Significantly elevated adjusted rate ratios in the cohort that received four or more doses included malignant neoplasms of the thyroid gland (rate ratio = 2.35, 95 percent CI = 1.01 to 5.48), diabetes mellitus (rate ratio = 3.05, 95 percent CI = 1.31 to 7.09), other disorders of the intestine (ICD-9-CM code 569; rate ratio = 3.24, 95 percent CI = 1.33 to 7.89), and osteochondropathies (ICD-9-CM code 732; rate ratio = 2.17, 95 percent CI = 1.09 to 4.32).

Rate ratios were not calculated for multiple sclerosis since there was only one case in the preimmunization time period (hospitalization rate, 0.21/100,000 population). The rates of hospitalization for multiple sclerosis were 7.60 and 3.25 per 100,000 population in the cohort that received one to three doses and the cohort that received four or more doses, respectively. The corresponding rate for those who never received AVA was 3.5/ 100,000 population. Thus, the rates of hospitalization for multiple sclerosis were similar in those receiving the greater number of AVA doses and in persons who had never been immunized with AVA.

It is also noteworthy, as mentioned earlier, that the prevaccination disease history of service members who received AVA because they were going to be deployed will, by definition, not include any severe, chronic conditions that would have disqualified them from deployment. For nearly all diagnostic groups, hospitalization rate ratios were smaller rather than larger for the higher-dose cohort. Thus, no dose-response effects of AVA and the risk of hospitalization were observed. A dose-response effect may not be observed, however, if persons with significant health conditions that required hospitalization, whether or not these conditions occurred in conjunction with exposure to AVA, did not receive additional doses of the vaccine. If this were the case, even in the presence of a true association, higher risk ratios would be expected for the cohort that received one to three doses.

Preimmunization Versus Nonimmunization Analyses The fourth analysis was somewhat different from the first three in that hospitalization rates in the time period before vaccination with AVA for those ultimately vaccinated were compared with the rates for those who were never vaccinated with AVA. This comparison would allow assessment of inherent differences in disease risk among those who received AVA at some time and those who never did. The results of these analyses were based on 11,436 hospitalizations during 478,093 person-years of observation in the cohort evaluated before immunization (crude rate, 23.92 per 1,000 person-years) and 109,893 hospitalizations during 2,890,037 person-years of observation in the cohort that was never immunized (crude rate, 38.02 per 1,000 person-years). The unadjusted overall rate ratio for hospitalization (preimmunization versus never immunized) was 0.63. The rate ratios for all major categories but one (diseases of the skin; adjusted rate ratio = 1.01) were less than 1.0, as would be expected if those who would receive AVA were healthier than those who never received AVA. These rate ratios ranged from 0.29 to 0.91, indicating for many conditions a substantial healthy soldier effect. Hospitalization rates for specific diagnoses of diabetes mellitus, regional enteritis, other disorders of the intestine, and multiple sclerosis were also significantly lower in the preimmunization cohort (rates provided in the interpretation section below). For five diagnoses (malaria; erythematous conditions; superficial injury of elbow, forearm, and wrist; toxic effect of carbon monoxide; and effects of air pressure), hospitalization rates were statistically significantly higher for the preimmunization cohort than in those never immunized (Appendix G, Table G-4). Overall, the results of these analyses confirm that those who ultimately received AVA were healthier as a group, even before receipt of the vaccine, than those who never received AVA.

Postimmunization Versus Preimmunization Analyses, Including Those Unvaccinated The final analyses, done separately for men and women, compared the rates of hospitalization for specific diagnoses during the period before receipt of AVA with the rates after receipt of the first dose of AVA. The population included all personnel on active duty between January 1, 1998, and December 31, 2000. In this comparison, the rates for the preimmunization cohort were based on those for all active-duty personnel, including those who never received AVA, whereas the postimmunization time period covered the interval after receipt of the first dose of AVA among those who were vaccinated. Twelve specific diagnoses were investigated: arthropathies and related disorders; asthma; diffuse disease of connective tissue; diabetes mellitus; disease of the ear and mastoid process; inflammatory and toxic neuropathy; cardiac dysrhythmias; lymphosarcoma and reticulosarcoma; multiple sclerosis; acute myocardial infarction; disorders of the thyroid gland; and diseases of the esophagus, stomach, and duodenum (Appendix G, Table G-5).

Among the women, there were 1,847 hospitalizations and 509,265 person-years of observation in the preimmunization cohort and 268 hospitalizations and 73,947 years of observation in the postimmunization cohort. Among the men, there were 11,684 hospitalizations and 2,858,865 person-years of observation in the preimmunization cohort and 2,361 hospitalizations and 664,434 years of observation in the postimmunization cohort. Among the men, none of the adjusted rate ratios (rates before vaccination with AVA versus the rates after vaccination with AVA) were significantly greater than 1.0, and the rate ratios ranged from 0.65 to 1.02. Among the women, however, the rate of hospitalization for multiple sclerosis was significantly increased for the postimmunization cohort compared with that for the preimmunization cohort (rate ratio = 2.14, 95 percent CI = 1.14 to 4.01). When analyses were restricted to incident cases so that multiple hospitalizations of the same woman would not be counted, the adjusted rate ratio for multiple sclerosis in the postimmunization interval versus that in the preimmunization interval was no longer significantly elevated (rate ratio = 1.26; 95 percent CI = 0.50 to 3.14).

The major limitation of this sex-specific analysis is that postimmunization rates (which are, by definition, based only on those for persons who received AVA) were compared with the preimmunization rates among all active-duty personnel. The latter group includes both those who would go on to receive AVA and those who were never immunized with AVA. Use of this comparison group would likely reduce the magnitude of any AVA-associated hospitalizations. On the other hand, it provides a somewhat “fairer” comparison for rates of hospitalization for severe conditions, such as aplastic anemia, that would have precluded ever receiving AVA.

Interpretation of Analyses of Data from DMSS Databases The committee emphasizes that the statistically significant associations observed above are not necessarily causal associations and, indeed, most likely are not causal associations. The interpretation of data such as these requires careful attention to several important but often subtle matters. For example, upon initial review of the postexposure versus the preexposure data (i.e., the initial analyses performed to evaluate risks while controlling for the healthy soldier effect), the results appeared to suggest an elevated risk of hospitalization for diabetes mellitus after receipt of AVA. At first blush, this could be evidence that AVA uncovers cases of diabetes that otherwise might not have been detected, as has been postulated for viral infections (Robles and Eisenbarth, 2001).

However, upon closer examination, a causal link appears to be unlikely. One possibility for observing a significant increase in rates of hospitalization for diabetes would simply be chance. In fact, 27 different conditions were found to be statistically significantly associated with AVA (15 conditions with rate ratios greater than 1.0 and 12 conditions with rate ratios less than 1.0), but 42 diagnoses would be expected to be significantly different in the periods before and after vaccination with AVA purely by chance because of the large number of conditions examined. By use of a conventional p value standard of .05, one would expect 1 in 20 findings to be statistically significant just by chance. However, in this situation that explanation appears to be unlikely. In examining the results stratified by sex, they are completely consistent. Yet there is only a 1 in 400 probability (0.05 × 0.05) that the results could be significant for both men and women independently purely by chance.

Instead, other patterns in the data make it clear that this association is unlikely to be causal. First, the elevated risk is present to the same degree in the time period >45 days after vaccination as in the time period 0 to 45 days after vaccination. This seems unlikely, although not impossible, if the mechanism was causal.

Diabetes is a common disease that is normally treated on an outpatient basis. Data on outpatient care, however, do not appear in the detailed DMSS analyses available to the committee at the time the report was written. A selection bias may affect data on hospitalizations for diabetes. If soldiers with known diabetes who were treated as outpatients were less likely to be deployed, they would be less likely to receive AVA. The result would be a lower than normal rate of hospitalization for diabetes before vaccination among those who would ultimately receive AVA. Comparison of a normal rate of hospitalization for diabetes after vaccination with this lower rate before vaccination would produce the false appearance of a positive association, and this false signal would persist, regardless of whether one were examining the time period right after the vaccination (0 to 45 days) or the time period thereafter (>45 days).

How can one be confident that the true explanation is this selection bias rather than a causal connection? A separate analysis compared the rates of hospitalization for any of 843 diagnoses in the prevaccination period with the rates for those who were never vaccinated. In general, the rates of hospitalization prevaccination were lower than the rates in the group that was never vaccinated, confirming the healthy soldier effect. The adjusted rate ratios varied, but most often they were about 0.7 or 0.8. However, for diabetes the comparable adjusted rate ratio was 0.12 (95 percent CI = 0.06 to 0.24). Thus, those who received AVA were dramatically less likely to be hospitalized for diabetes than those who were never vaccinated. The normal rate of spontaneous development of diabetes after vaccination would therefore falsely appear as an increased risk. The same was true when the prevaccination rates were compared with the rates in those who never received AVA for some of the other apparent signals, such as regional enteritis (rate ratio = 0.14, 95 percent CI = 0.06 to 0.35) and other disorders of the intestine (rate ratio = 0.28, 95 percent CI = 0.12 to 0.64). For multiple sclerosis a selection bias seems even more likely, with a hospitalization rate ratio of about 0.06 for the preimmunization cohort versus those never vaccinated with AVA (based on only one preimmunization case of multiple sclerosis).

Overall, the analyses of data from DMSS were very reassuring. They indicate that exposure to AVA is not associated with a significantly increased risk for any condition of later onset that cannot be otherwise explained by biases inherent in this type of analysis. Several possible “signals” were observed, however. Signals are the earliest indication of a possible causal relationship between an exposure and a health event. These conditions include diabetes, regional enteritis, and multiple sclerosis. The committee's judgment is that these signals are probably not causally linked to exposure to AVA but most likely are due to random error or biases. However, a causal link cannot be completely excluded. Thus, these signals deserve continued surveillance; in addition, ad hoc studies are required to further explore the possible links of these signals with exposure to AVA. Such studies could involve additional analyses with data from DMSS, as well as examination of medical records to validate the diagnosis and the timing of the onset of symptoms in relation to the vaccine exposure.

The committee was impressed by the creativity and rigor of the military professional staff working with the data in the DMSS databases and their productivity. However, the committee also counsels great caution in the use of approaches that use such data collected through automated systems for signal generation. As expected by chance alone, the rates of several diseases and conditions will predictably appear to be elevated in one group or another. Although random error and bias are likely explanations for these increases, other conclusions might also be drawn. In other words, these preliminary findings should lead to further examination of the data. The current DoD approach and organization focus on screening DMSS data for hypotheses. DoD should, however, devote more attention and resources to the evaluation of these hypotheses, as was begun in response to the committee's inquiries. As has been articulated in a set of good epidemiology practices developed for use with similar administrative and clinical data sets in civilian practice (Andrews et al., 1996), analysis of such data requires the exercise of great caution and a commitment to devote the necessary resources to explore the possible associations that might surface from such exercises. Chapter 8 discusses recommended improvements for use of DMSS data.

Thus, finding an increased rate of occurrence of one or more adverse events must be considered a signal until proper review provides an alternative explanation. Criteria for determination of which signals should be further evaluated need to be developed and routinely applied. At a minimum, a system for retrieval and review of primary medical records is required to be able to rule out coding and classification errors, to search for subtle but possibly explanatory variables that may confound an association, or to differentiate a true signal from a statistical chance event.

Finding: DMSS data are screened quarterly to identify statistically significant elevations in hospitalization and outpatient visit rate ratios associated with receipt of AVA. In this way, DMSS promises to be very useful as a tool for hypothesis generation.

Finding: The elevated rates of specific diagnoses in the various analyses of DMSS data are not unexpected per se; that is, they appear to be explicable by chance alone. The bias of selection of healthy individuals for receipt of AVA is also a likely explanation for some observed associations. Thus these elevated rates should not be automatically viewed as an indication of a causal association with the receipt of AVA. However, additional follow-up is needed.

Recommendation: AMSA staff should follow up the currently unexplained elevations in hospitalization rate ratios for certain diagnostic categories among the cohorts of AVA recipients. Studies might include additional analyses with the database or examination of medical records to validate and better understand the exposures and outcomes in question. A protocol should be developed to ensure that such follow-up regularly and reliably occurs after a potential signal is generated.

Finding: Examination of data from the DMSS database to investigate potential signals suggested by VAERS reports related to vaccination with AVA has not detected elevated risks for any of these signals for the vaccinated population, although continued monitoring is warranted.