Vaccine Safety Datalink Project: A New Tool for Improving Vaccine Safety Monitoring in the United States
Objective. To fill the large “gaps and limitations” in current scientific knowledge of rare vaccine adverse events identified in recent reviews of the Institute of Medicine.
Methods. Computerized information on immunization, medical outcomes, and potential confounders on more than 500 000 children 0 to 6 years of age is linked annually at several health maintenance organizations to create a large cohort for multiple epidemiologic studies of vaccine safety.
Results. Analysis of 3 years of follow-up data shows that 549 488 doses of diphtheria-tetanus-pertussis (DTP) and 310 618 doses of measles-mumps-rubella (MMR) vaccines have been administered to children in the study cohort. Analyses for associations between vaccines and 34 medical outcomes are underway. Screening of automated data shows that seizures are associated with receipt of DTP on the same day (relative risk [RR], 2.1; 95% confidence interval [CI], 1.1 to 4.0) and 8 to 14 days after receipt of MMR (RR, 3.0; 95% CI, 2.1 to 4.2). The diversity of vaccination exposures in this large cohort permits us to show that an apparent association of seizures 8 to 14 days after Haemophilus influenzae type b vaccine (RR, 1.6; 95% CI, 1.2 to 2.1) was attributable to confounding by simultaneous MMR vaccination; the association disappears with appropriate adjustment (RR, 1.0; 95% CI, 0.7 to 1.4).
Conclusion. Preliminary design, data collection, and analytic capability of the Vaccine Safety Datalink project has been validated by replication of previous known associations between seizures and DTP and MMR vaccines. The diversity in vaccine administration schedules permits potential disentangling of effects of simultaneous and combined vaccinations. The project provides a model of public health-managed care collaborations in addition to an excellent infrastructure for safety and other studies of vaccines.
Immunizations are among the most cost-effective and widely used public health interventions.1 The incidence rates of vaccine-preventable diseases in the United States and most countries worldwide2 have decreased dramatically during recent decades (Table 1). No vaccine is perfectly safe, however. Increased vaccine use necessarily results in an increased number of true vaccine reactions as well as adverse medical events coincidentally associated with vaccinations.3 The number of both types of reports to the Vaccine Adverse Event Reporting System (VAERS) in the United States, approximately 10 000 per year, now exceeds the reported incidence of most vaccine-preventable childhood diseases combined (Table 1).
Because few vaccine-preventable diseases are currently eradicable, most immunizations must be continued indefinitely. One important way to minimize vaccine injuries is to improve our understanding of vaccine safety and thereby foster the development and use of safer vaccines.4 Close monitoring of vaccine safety should also help prevent the loss of public confidence in immunization programs and the subsequent resurgence of vaccine-preventable diseases, as experienced with pertussis in several countries5-7 and more recently with diphtheria.8
Despite the importance of vaccine safety, the Institute of Medicine (IOM) recently found that serious “gaps and limitations” exist in both the knowledge and infrastructure needed to study vaccine adverse events.9,10 Among the 76 types of vaccine adverse events reviewed by the IOM, the scientific evidence was inadequate to assess definitive vaccine causality for 50 (66%). The IOM also noted that “if research … [is] not improved, future reviews of vaccine safety will be similarly handicapped.” These gaps in knowledge are attributable to several factors. Prelicensure controlled trials provide only limited safety data because of their relatively small sample size, short duration, and population homogeneity. Postlicensure studies are therefore needed to provide a fuller understanding of the safety of vaccines in general use.4
Historically, postlicensure studies of safety have relied on passive surveillance systems such as the VAERS.3 Because of methodological weaknesses, such as the potential for biased reporting and underreporting and lack of denominators or comparison groups, data from such case reports are usually not helpful in assessing risk or vaccine causality.4 Ad hoc retrospective epidemiologic studies of vaccine safety (eg, swine influenza vaccine and Guillain-Barré syndrome11 and pertussis vaccine and encephalopathy5), although potentially informative about vaccine causality, are costly, time consuming, and usually limited to assessment of a single event.
Recognizing the need to improve the capability to study vaccine safety, the Centers for Disease Control and Prevention (CDC) participated during the late 1980s in two pilot studies using large linked databases (LLDBs) of computerized vaccination and medical records.12-14 These studies helped validate the LLDB approach for vaccine safety studies. The need for a larger LLDB population for continuous assessment of vaccine safety prompted the CDC to initiate planning for the Vaccine Safety Datalink (VSD) project in 1989.15 This article provides an overview of the design of the VSD project and preliminary results and reviews the prospects and the limitations of the VSD to address the vaccine safety issues identified by the IOM, as well as those of future vaccines.16
The need to improve postlicensure monitoring of drug safety became widely recognized after the thalidomide disaster.17 Faced with the methodological limitations in passive surveillance for drug adverse events, pharmacoepidemiologists began during the 1980s to turn to LLDBs linking computerized pharmacy prescription and medical outcome records.18 These databases derive from defined populations, such as members of health maintenance organizations (HMOs), single-provider health care systems, and Medicaid programs. Because the databases are usually generated in the routine administration of such programs and do not require completion of an adverse event reporting form, the problems of underreporting or recall bias are reduced. Because these programs have enrollees numbering from thousands to millions, large populations can be examined for relatively infrequent adverse events. Denominator data on doses administered and the ready availability of appropriate comparison groups are also very useful. Therefore, LLDBs can potentially provide an economical and rapid means of conducting postlicensure studies of the safety of drugs and vaccines (Table 2).19,20
Study Sites and Population
In 1991, the CDC began a partnership with four HMOs to evaluate vaccine safety in children in a large-scale prospective study. The HMOs include Group Health Cooperative (GHC) of Puget Sound in Washington, Northwest Kaiser Permanente (NWK) in Oregon, Northern California Kaiser (NCK), and Southern California Kaiser (SCK) Permanente health programs. These sites were chosen for their research experience and the existence of or willingness to create HMO-wide computerized vaccination databases for this project. Other Public Health Service agencies, including the Center for Biologics Evaluation and Research of the Food and Drug Administration and the Division of Vaccine Injury Compensation of the Health Resources and Services Administration, have been important contributors to this ongoing project.
The initial study focused on children 0 to 6 years of age but is being expanded to include adolescents and adults. To be eligible for the study, each child must be a member of the HMO, be within predefined age limits, live within the catchment area of a given site's participating clinics, and receive vaccine in a study clinic. The power to examine any particular vaccine and adverse event association within the VSD depends not only on the frequency of the vaccination but also on the sensitivity and specificity of the case definition, the background rate for the event, and the magnitude of the risk. Previous LLDB studies reviewed adverse event experience after approximately 500 000 doses of diphtheria-tetanus-pertussis (DTP) vaccine in 10 years. Our study accumulates similar exposure experience after 2 years. With time, we expect that the VSD study should have adequate power to detect potential vaccine reactions with attributable risk of about 1 outcome per 100 000 vaccine doses (assuming a three-dose vaccine, relative risk [RR] of 1.6, and background rate of 1.2 × 10−5 to 1.6 × 10−3). This magnitude of attributable risk is similar to that suggested by acute encephalopathy after whole-cell pertussis vaccination5,9 or Guillain-Barré syndrome after swine influenza vaccination.11
Health service use information for each patient is computerized and continuously compiled by each HMO indexed by an unique identifier. These data were initially used primarily for internal HMO administrative and clinical patient treatment purposes but have been adapted to this study. Data collection in a standardized VSD format began on March 1, 1991, for three HMOs (GHC, NWK, and NCK) and October 1, 1992, for SCK. The data are organized into files containing demographic information, covariate information, vaccination records, and various types of medical outcome data (Table 3). The automated outcome data are collected from various sources at each site, such as records of hospitalizations (all sites), emergency department visits (all sites), and outpatient clinic visits (GHC, NWK, and NCK). Each site encodes their patients' clinical data with unique study identifiers before shipping the data to the CDC annually for merging and analysis, thereby preserving patient confidentiality. Institutional review boards at each HMO have approved this project, in which only analyses with aggregate data are presented.
All vaccinations given within the HMO study population, either routinely or for special indications, are recorded and entered into a computer database. For all patients, automated data include: (1) the vaccine type, (2) the date of vaccination, and (3) concurrent vaccinations. For most patients, additional automated data available include whether vaccinations were obtained in or outside the HMO, as well as information required by the National Childhood Vaccine Injury Act of 198621: (1) the manufacturer, (2) lot number, and (3) site of vaccination.
Adverse Medical Events (Outcomes) of Interest
International Classification of Diseases (ICD)-9 codes for all hospitalizations and emergency department visits are compiled for the VSD study cohort. Automated diagnostic codes for routine outpatient clinic visits are currently available on approximately half of the cohort. This proportion will increase substantially as the HMOs switch to automated outpatient record systems. For initial study, the project identified 34 principal medical outcomes of possible association with vaccinations (Table 4) based on a thorough review of the literature and the IOM reports. In addition, the safety of certain vaccination practices are to be evaluated. These include simultaneous versus combined vaccination of various antigens and the relative adequacy of observation of contraindications to vaccinations. The list of research questions is amended yearly based on evolving issues in the medical literature and new vaccination policy needs.
The 34 outcomes are identified in the medical outcomes files by one or more ICD-9 diagnostic codes. Additional outcome information is obtained from selected ancillary sources (eg, procedures and laboratory tests). These sources may be used both for case finding and to supplement primary diagnostic codes. For example, positive blood cultures may enhance the accuracy of a discharge diagnosis of bacteremia. The case definitions for each of these 34 outcomes, consisting of one or more different ICD-9 and other diagnostic codes depending on source, have been determined iteratively and continue to evolve with the review of each year's analyses.
As a potential adverse event of immunization, death occurring soon (eg, 60 days) after vaccination is important to assess. However, deaths in children often occur in the home or outside of the routine health care system (eg, at an accident site or as a result of sudden infant death syndrome). Therefore, we maintain surveillance of state death reports in addition to monitoring hospitalizations and emergency department visits within the HMO. State death records are linked to LLDB cohort patients using a probabilistic matching algorithm.22Once identified, all deaths in study cohort patients are further validated by review of medical and autopsy records.
Factors that affect both vaccination status and incidence of medical events may confound observational studies of vaccine safety.23 Therefore, we obtain from state birth certificate tapes relevant information, such as parental education and occupation, birth weight, and Apgar score. Similarly, additional information on socioeconomic status is obtained on the VSD study cohort by linking the zip codes and street addresses of the patients with their respective census tract blocks via “geocode.”24
Data Quality Control Procedures
To study potential rare associations between vaccines and adverse events, large and accurate databases are needed.25 Routine procedures for assuring the quality of HMO databases vary by type of database and HMO. Inpatient databases are important for hospital and financial management; therefore, data quality is high because of staff training, standardized coding protocols, reliability monitoring, and routine audits. Both routine data quality audits required by hospital accrediting agencies and special audits conducted by other research projects have found inpatient databases to be complete and accurate.26 Drug, laboratory, radiologic, and referral databases contain user-based data that involve clinical and administrative data essential for delivery of medical services. Such information is constantly monitored for accuracy and is used in other research projects.27
In addition to routine quality checks for each of the databases, a random 2% sample of the study populations (1% at the larger sites: NCK and SCK) is selected periodically to review the quality of automated vaccination and diagnostic data. For these samples, vaccination and diagnostic data are abstracted from the medical records and compared with the automated data. Preliminary data from the first year of the VSD found nearly perfect agreement between automated and abstracted hospital discharge diagnoses.28 The percentage agreement between the dates of the abstracted and automated emergency departments diagnoses ranged from 60% to 87%. The percentage agreement between abstracted and automated vaccination dates ranged from 70% to 98% across vaccines (with the exception of DT) and HMOs. The primary source of disagreement was the incomplete entry of all vaccinations into the database. Continuous feedback to HMOs is provided to identify potential means of improving the quality of the automated data.
Inference that a vaccine causes an adverse event may be drawn if higher rates are consistently observed among vaccinees compared with nonvaccinees, especially if results are consistent for the different HMOs and for several study years. Because of high vaccine coverage within the HMOs, for most outcomes of interest, there are too few unvaccinated persons, and those not vaccinated may not constitute representative comparison group for epidemiologic analyses. Alternatively, an analysis of “risk interval” is used (Table 2). For each outcome of interest, time intervals after vaccination are selected during which an adverse event would be expected if an association exists. These intervals are defined a priori based on biological and clinical considerations. For the study population, incidence rates of adverse events within and outside the specific risk interval for the study population are then compared using appropriate statistical methods, after controlling for potential confounders.
Several statistical analytical designs are used in the VSD study to compare incidence rates, depending on the accuracy and completeness of the automated data. If the quality of the automated data for an outcome of interest is high, multivariate cohort methods (eg, Poisson or proportional hazards regression) are used. Otherwise, the automated records will require validation by medical record review for all persons suspected of having the outcome. Similar validation of the medical records of the random 1% to 2% of the study population selected for quality control assessment permits a more efficient means of providing the comparison group for a case-cohort analysis.29 Validation studies using nested case-crossover30 or case-control methods may also be done in parallel or sequentially, especially if a previously unknown association is detected. The medical records of cases, or cases and controls (also selected from the 1% to 2% samples if possible), will be thoroughly reviewed, and children or their parents may be interviewed to identify other potential confounders.
Because each HMO has its own set of medical and operating procedures, much of the first 2 years of the VSD study was devoted to developing protocols to standardize prospective data collection across sites. NCK and SCK incrementally expanded their automated vaccination registries throughout the HMOs. Consequently, the annual study population under surveillance grew from approximately 181 000 in 1991 to approximately 502 000 children younger than 7 years (approximately 2% of the US population in these age groups) by late 1996, encompassing 1 862 000 child-years of observation. Complete vaccination and other medical records are available now on a cohort of 242 000 children born into the study, which will grow by about 63 700 children annually. The 10 most common vaccines and vaccine combinations administered to the cumulative study cohort by late 1994 are shown in Table 5.
Figure 1 plots the vaccine coverage for specific vaccines by age for the VSD cohort. The delay between the recommended age for a vaccine dose and when 80% of the cohort actually receives it increases for older children. The lowest coverage of routine childhood vaccinations in the cohort was for hepatitis B (universal recommendations began after the start of the study) and the second dose of measles vaccine (administered at primary school entry at two HMOs and secondary school entry at the others). The highest coverage was for polio vaccination, exceeding pertussis-containing vaccines by about 5% to 10%; this may measure the proportion of children with concerns about the safety of pertussis vaccines.
The simultaneous administration of multiple vaccines is very common. For example, when children younger than 6 months in our study received DTP vaccine, 97% of them also received Haemophilus influenzae type b (Hib) conjugate vaccine at the same visit. Thus, at certain ages it is difficult, if not impossible, to differentiate the possible effects of certain vaccine combinations. The vaccination schedules at the four HMOs are not identical, however, and this provides opportunities for separating the possible adverse effects of some vaccine combinations. For example, in the second year of life, one HMO gives DTP, Hib, oral polio vaccine, and measles-mumps-rubella (MMR) at the same visit, whereas the other three sites commonly give MMR and Hib at one visit and DTP (or diphtheria-tetanus-acellular pertussis [DtaP]) and oral polio vaccine at a later visit.
Other studies have shown that both DTP and MMR vaccines are associated with febrile seizures,9,10,14 and this relationship was explored in the VSD study to illustrate its analytic capabilities. Because it is difficult to differentiate between febrile and other types of seizures without chart review, we analyzed all seizure types combined with the automated data. Using a Cox regression model stratified by HMO and birth date (children born in 3-day blocks), the relative rate of seizures in specific time windows after receipt of vaccine (same day, 1 to 3, 4 to 7, 8 to 14, and 15 to 30 days) was compared with periods before and more distant to the receipt of vaccine. Elevated rates of seizures, presumably mostly febrile, were found for both DTP and MMR, consistent with each vaccine's respective modes of biologic action (Fig 2). The risk varied for different periods after immunization. The risk of seizures after MMR, a live viral vaccine requiring a longer incubation period for viral replication, was delayed compared with that of DTP, a killed bacterial vaccine.
To examine the capability of the VSD cohort to differentiate the effects associated with simultaneous administration of vaccines, we examined the association between seizures and the administration of Hib and MMR vaccines. On crude analysis, a possible association was found 8 to 14 days after vaccination with both Hib (RR, 1.4; 95% confidence interval [CI], 1.2 to 1.8) and MMR (RR, 2.3; 95% CI, 1.9 to 2.9) vaccines. Prior experience suggests that the association with Hib may be an artifact of the frequent coadministration of Hib with MMR during the second year of life and not a true association with Hib. Because the two vaccines are also frequently administered separately, we were able to adjust our analysis via regression to show that there is in fact no association between Hib and seizures (RR, 0.9; 95% CI, 0.7 to 1.2), whereas the association with MMR persisted (RR, 2.42; 95% CI, 1.8 to 3.2).
On screening of automated data, we found 8 to 14 days after immunization an apparent association between measles vaccine and invasive bacterial disease (RR, 2.3; 95% CI, 1.1 to 4.6). This association was not validated by chart review, because the exposed cases were found to have undergone evaluation of fever of unknown origin (possibly caused by the vaccine) but did not actually have culture-proven sepsis. This example shows that the ICD-9 codes selected for this outcome had the desired high sensitivity for surveillance purposes but needed validation.
Ad Hoc Studies
In addition to the planned vaccine safety studies, the infrastructure created by the VSD project permits timely investigation of new hypotheses. For example, when changes in vaccine policy are considered, it is often necessary to develop new information to evaluate such a potential change. The Advisory Committee on Immunization Practices recently considered lowering the age of the tetanus-diphtheria (Td) booster from 14 to 16 years of age to 10 to 12 years of age. However, information on the safety of the Td vaccine booster dose given to a younger age group was lacking. In the VSD study, we identified cohorts of 12 626 and 3379 children in the two age groups, respectively, who had received Td boosters. Comparison of rates of emergency department and hospital use within 7, 14, and 30 days after Td showed some differences between the two groups, which disappeared when visits for trauma and suture removal were excluded (Table 6), supporting the safety of this schedule change. This analysis further illustrates the importance of adjusting for potential confounders in VSD studies. In this case, age itself affected rates of emergency department use independent of Td use.
Vaccines are generally administered to healthy persons, frequently infants and children. Therefore, the acceptable risk of adverse reactions to vaccines is lower than that for therapeutic agents for ill persons. This lower tolerable risk translates into the need to conduct studies to detect rare reactions (eg, attributable risks on the order of one per 105 to 106 doses).4Studies able to address such rare risks are possible only after licensure and general use, and they are large and expensive and may not provide conclusive results. For example, the National Childhood Encephalopathy Study was a case-control study that aimed to detect all hospitalizations in England and Wales for acute neurologic illness in children 2 to 35 months of age during a 3-year period.5 The study findings were controversial, because the conclusion was based on only seven cases of chronic encephalopathy observed within 3 days of DTP vaccination.32 These difficulties plus the limitations of passive surveillance3,4 largely account for the relative sparsity of data on vaccine safety in the recent IOM review.9,10
In recent years, developments in health care organization, health information systems, and pharmacoepidemiologic methods have improved our capability to study rare drug reactions.18 Walker et al12 and Griffin et al13 pioneered the use of such record linkage studies to evaluate vaccine safety. These studies were limited, however, by their relatively small sample sizes, retrospective design, and focus on the most severe reactions.9 The VSD study attempts to overcome these shortcomings by prospective collection of vaccination, medical outcome, and covariate data under joint protocol at multiple sites. Selection of prepaid health plans also minimized potential biases resulting from data generated from fee-for-service claims. Substantial efforts have been required to implement accurate automated vaccination record systems for our cohort, which represents approximately 2% of the children in the United States. A list of key research questions and how best to answer them within the VSD has been elaborated. Quality control procedures and methodological approaches have also been developed. After all this development, does it work?
Although much remains to be done to improve the VSD, the early results are promising. Previously known associations between seizures and DTP and MMR vaccines have been validated in the prospective VSD cohort. This provides validation of the design, data collection, and analytic approaches of this project. The medical charts for many children with seizures identified from automated records are being abstracted. This will permit an evaluation of the accuracy of the automated system for a rigorous case-control analysis to distinguish between possible vaccine causation of first and subsequent seizures, as well as to characterize the types of seizures (eg, febrile versus other) associated with vaccinations.
In addition to studies to assess potential hypothesized vaccine associations, new ad hoc questions that arise from the VAERS, from changes in immunization schedules (eg, new vaccines such as varicella or the use of simultaneous vaccination) or from screening level cohort analyses in the VSD, can be addressed in a timely manner. For example, in response to concerns regarding a potential increased risk of arthropathy in adult women after rubella vaccination,33 the VSD database was used to identify a cohort of women who had rubella immunity testing during pregnancy and their subsequent rubella immunization status. In a retrospective review of the women's charts, no association of any chronic arthritic condition was found with receipt of rubella vaccine.34 The potential association between hepatitis B vaccination at birth and suspected neonatal sepsis work-up is being examined in another VSD study.
Each of the core databases created by the project (eg, vaccinations, medical outcomes, and potential covariates) has valuable applications. Each HMO in the VSD study uses its automated immunization records to set goals for improvement in vaccine coverage levels.35,36Children, clinics, and practices with inadequate immunization can be easily identified, and strategies can be developed to improve coverage. Research into barriers, missed opportunities, and recall systems for immunization have been performed.37 Documentation for school-entry immunization requirements is also easily retrievable. These records have been used to calculate the level of childhood immunizations as of the second birthday included as quality measures in the Health Plan Employer Data and Information Set.38,39Finally, the VSD databases provide excellent bases for construction of broader regional immunization registries.40
Studies of many other pediatric illnesses via the VSD are also possible. Taking advantage of the cohort infrastructure created, plans are underway to expand the VSD study to examine: (1) vaccine safety issues in adolescents and adults, (2) the impact of vaccination programs on incidence of vaccine-preventable disease (eg, new varicella vaccine), (3) cost effectiveness of specific vaccines, and (4) safety and immunogenicity of new combined vaccine schedules in prospective phase II and III clinical trials nested within the cohort.41
The diversity in vaccination practice at the four HMOs and the clinic-to-clinic and day-to-day variations in practice permit useful contrasts in safety experiences. A study contrasting the safety of the second dose of measles vaccine administered at entry to primary school (as recommended by the Advisory Committee on Immunization Practices) versus entry to secondary school (as recommended by the American Academy of Pediatrics) is currently underway. As demonstrated by the Hib, MMR, and seizure example, the size of the VSD population may also permit separation of the risks associated with individual vaccines from those associated with vaccine combinations, whether given in the same syringe or simultaneously at different body sites. Such studies will be especially valuable in view of the new combined pediatric vaccines currently in development.42
Should the VSD study identify a vaccine reaction, data on attributable risk will be available, thereby permitting accurate risk-benefit assessment by both the public and policy makers.43 Subgroup analyses may permit identification of risk factors, which may be useful in identifying contraindications to vaccinations. Research may then be launched to understand the pathogenesis of the reaction in these individuals, potentially leading to the development of safer vaccines. The incidence rates of reactions identified in the VSD should permit the evaluation and improvement of passive surveillance systems such as the VAERS. The VSD data will also be invaluable to other Public Health Service agencies sharing responsibility for vaccine safety. The Food and Drug Administration, in fulfillment of its regulatory responsibilities, is interested in potential product-specific differences in the vaccine safety profiles. The results of the VSD will also substantially enlarge the scientific basis for deciding whether to recommend compensation in alleged vaccine injury cases, fundamental to a fair and efficient vaccine injury compensation program.44
Amid these promises, a few caveats are appropriate. Although diverse, the population in the four HMOs currently in the VSD is not wholly representative of the US population in terms of geography or socioeconomic status. With current changes in health care organization and additional resources, it may be possible to broaden the scope of the VSD. In the interim, there is little reason to believe that these factors significantly influence the risk of vaccine reactions. More importantly, because of the high rate of vaccine coverage attained in the HMOs, few nonimmunized control subjects are available. The VSD must therefore rely predominantly on some type of risk interval analysis. The capability of this approach to assess associations between vaccination and adverse events with delayed or insidious onset (eg, autism) is limited. Similarly, the ability of the VSD to distinguish effects of combined or simultaneous vaccination fully may be limited should such practices become universal.
The VSD also cannot easily assess adverse events not currently captured in existing HMO databases, either because they do not result in health care consultations or because the data are not automated. Important nonautomated data sources relevant to the VSD study (eg, results of neurologic consultations) have required manual abstraction, coding, and computerization. The patient enrollment, health care practices, and health information systems at each HMO are dynamic, which may either aid or impede study of specific outcomes. Coding errors occurs inevitably in all data files to some extent, resulting in a decrease in our ability to detect a true association. The current VSD is also unable to examine the risk of extremely rare events after infrequent vaccinations, such as Guillain-Barré syndrome after each season's flu vaccine. Because the VSD relies on epidemiologic methods, it may not successfully control for confounding and bias in each analysis,23 and inferences on causality may be limited. Finally, even if findings from the VSD may often be “negative” (ie, may show no elevations in risks in association with vaccination), one cannot absolutely “disprove” an alleged reaction.10,45
Despite these potential shortcomings, the VSD provides a new, essential, powerful, and cost-effective complement to our ongoing evaluations of vaccine safety in the United States. The capability of the VSD for reliable and consistent ascertainment should reassure the public of the adequacy of the surveillance for significant vaccine adverse events and the general safety of routine vaccine products. Enhanced public confidence is integral to maintaining or improving rates of vaccine acceptance at a time of rapid changes in vaccine schedules46 and introduction of new vaccines.16
We thank Drs Marie Griffin, Harry Guess, Edgar Marcuse, and Alec Walker for their invaluable insights as external advisors; the National Vaccine Program Office and the FDA for their financial assistance; and Bill Nichols, Phil Horne, and Dr Walt Orenstein for their unwavering support of the VSD project.
- Received April 1, 1996.
- Accepted June 14, 1996.
Reprint requests to (R.T.C.) National Immunization Program (MS-E61), Centers for Disease Control and Prevention, Atlanta, GA 30333.
↵‡ Other Vaccine Safety Datalink team members by site: Emmett Swint, MA*; Janet R. Hardy, MSc, MPH*; Thomas Payne, MD‡; Virginia Immanuel, MPH‡; Patti Benson‡; Julie Drake‡; Lois Drew§; Barbara Mendius, MS§; Paula Ray, MPH∥; Ned Lewis, MPH∥; Bruce H. Fireman, MA∥; Jennie Jing, MA¶; Michael Wulfsohn, MD, PhD¶; Marlene M. Lugg, DrPH#; Patricia Osborne, MS#; Suresh Rastogi, PhD**; Peter Patriarca, MD**; and Vito Caserta, MD, MPH (Division of Vaccine Injury Compensation, Health Resources and Services Administration, Rockville, Maryland).
- VAERS =
- Vaccine Adverse Event Reporting System •
- IOM =
- Institute of Medicine •
- CDC =
- Centers for Disease Control and Prevention •
- LLDB =
- large linked database •
- VSD =
- Vaccine Safety Datalink •
- HMO =
- health maintenance organization •
- GHC =
- Group Health Cooperative (of Puget Sound) •
- NWK =
- Northwest Kaiser •
- NCK =
- Northern California Kaiser •
- SCK =
- Southern California Kaiser •
- DTP =
- diphtheria-tetanus-pertussis •
- RR =
- relative risk •
- ICD =
- International Classification of Diseases •
- Hib =
- Haemophilus influenzae type b •
- MMR =
- measles-mumps-rubella •
- CI =
- confidence interval •
- Td =
- ↵World Bank. World Development Report 1993: Investing in Health. New York, NY: Oxford University Press; 1993
- ↵Chen RT. Special methodological issues in pharmacoepidemiology studies of vaccine safety. In: Strom BL, ed. Pharmacoepidemiology. Sussex, United Kingdom: John Wiley & Sons; 1994
- Miller DL,
- Alderslade R,
- Ross EM
- Krantz I,
- Taranger J,
- Trollfors B
- ↵Howson CP, Howe CJ, Fineberg HV, eds. Adverse Effects of Pertussis and Rubella Vaccines: A Report of the Committee to Review the Adverse Consequences of Pertussis and Rubella Vaccines. Washington, DC: National Academy Press; 1991
- ↵Stratton KR, Howe CJ, Johnston RB, eds. Adverse Events Associated with Childhood Vaccines: Evidence Bearing on Causality. Washington DC: National Academy Press; 1994
- Schonberger LB,
- Bregman DJ,
- Sullivan-Bolyai JZ,
- et al.
- Griffin MR,
- Ray WA,
- Fough RL,
- et al.
- Walker AM,
- Jick H,
- Perera DR,
- et al.
- Wassilak SGF,
- Glasser JW,
- Chen RT,
- Black S,
- Mullooly JP,
- Thompson RS
- Frontiers in medicine: vaccines
- Strom BL,
- Carson JL
- Fine PEM,
- Chen RT
- Krieger N
- Mullooly J
- ↵Institute of Medicine. Reliability of National Hospital Discharge Survey Data. Report of a Study. Washington, DC: National Academy of Sciences; 1980
- ↵Fox LA, Stearns G, Imbiorski W, eds. The Administrator's Guide to Evaluating Records and the Medical Records Department. Chicago, IL: Care Communications; 1979
- Mullooly JP,
- Black S,
- Thompson RS,
- Glasser JW,
- Chen RT,
- Rhodes P
- Prentice RL
- Farrington CP,
- Nash J,
- Miller E
- ↵Ray P, Black S, Shinefield H, et al. Retrospective cohort evaluation of chronic rubella vaccine arthropathy. Abstracts of the 34th Interscience Conference on Antimicrobial Agents and Chemotherapy. Washington, DC: American Society for Microbiology; 1994:240. Abstract
- Davis R,
- Vadheim C,
- Black S,
- et al.
- Lieu TA,
- Black SB,
- Sorel ME,
- Ray P,
- Shinefield HR
- ↵Cordero JF, Guerra FA, Saarlas KN, eds. Developing immunization registries: experience from the All Kids Count Program. Am J Prev Med. 1997;13(suppl 1):1–128
- ↵Centers for Disease Control and Prevention. Request for Proposals No. 200-95-0947(P). Comprehensive Linked Data Collection of Medical Events and Immunization (Vaccine Safety and Development Datalink). Atlanta, GA: Centers for Disease Control and Prevention; 1995
- ↵Williams JC, Goldenthal KL, Burns DL, Lewis BP Jr, eds. Combined Vaccines and Simultaneous Administration: Current Issues and Perspective. New York, NY: New York Academy of Science; 1995
- ↵Hinman AR, Orenstein WA. Public health considerations. In: Plotkin SA, Mortimer EA, eds. Vaccines. Philadelphia, PA: WB Saunders; 1994:903–932
- ↵Rothman KJ, ed. Causal Inference. Chestnut Hill, MA: Epidemiology Resources, Inc; 1988
- Copyright © 1997 American Academy of Pediatrics