Interactions of Drugs, Biologics, and Chemicals in U.S. Military Forces (1996)

The approach to the study of interactions advocated in Chapter 4 presupposes that a large proportion of all interactions will be unknown and unpredictable. Therefore, a comprehensive strategy for studying such interactions must depend on surveillance. This requires both alerting mechanisms to signal that unanticipated health effects may have occurred and confirming mechanisms to estimate the incidence of these events and to determine if they were due to some particular exposure. Alerting mechanisms include case reports as well as broad monitoring programs designed to search for changes in rates of clinically important events. Such efforts also encompass the analysis of accidents or so-called natural experiments. In contrast, confirmatory data collection and analysis will be more focused on specific hypotheses. Such confirmatory data may be obtained from cohort studies or case-control studies, as well as randomized experiments and intervention studies, which include cessation of use studies. Animal studies should also be considered, either as preliminary or concomitant investigations, to help establish biological plausibility (see Chapter 5).

Automated multipurpose databases are increasingly being used for epidemiologic purposes, including surveillance. Such databases are typically assembled by combining patient-level information from two or more separate files originally developed primarily for nonresearch purposes. Through record linkage, it is possible to create these person-level longitudinal files on a general or

ad hoc basis. Examples of such multipurpose databases used in epidemiologic studies of drug effects include those assembled from the records of health maintenance organization enrollees, Medicare or Medicaid populations, military or Department of Veterans Affairs cohorts, or some other defined populations. Such databases are considered to be population-based if the information that they contain is derived from some underlying population in the statistical sense, specifically, some group of known size and composition from which statistical samples might be drawn. The utility of multipurpose databases for epidemiologic studies largely depends on the ability to assemble such files through linkage of individual person-level records and the quality and the completeness of the information recorded in such files.

In carefully designed studies, multipurpose databases offer a number of advantages: the ability to conduct studies of uncommon health events or of understudied but well-defined populations, the minimization of study costs, a reduction in the amount of time needed to conduct a study, and the opportunity to study a large number of people. However, studies based on multipurpose data sets are affected by a number of potential problems, including completeness and the quality of the data. It is worth reiterating that data collected for administrative purposes are not necessarily suited to research purposes. In such instances, careful editing and cautious use of these data can help to minimize the effects of their limitations.

Although there are no perfect surveillance systems, much can be done with currently available databases. This chapter describes Army and DoD surveillance activities that are under way or in development, as well as other sources of material from which to assemble multipurpose databases. There are also brief discussions of two nonmilitary systems —the Vaccine Adverse Event Reporting System (VAERS) and MEDWatch—that can provide useful information.

The Army Medical Surveillance Activity, the Defense Medical Epidemiology Database, and the Uniformed Services Prescription Database Project (USPDP) are the three automated multipurpose database systems currently in development that can be used to monitor adverse health events and their potential causes. This section briefly describes these databases, their developmental histories, and their potential uses for studying the interactions of drugs, biologics, and chemicals, as well as plans for further development of surveillance systems.

The Army Medical Surveillance Activity is a multipurpose database whose component parts include the following: the Reportable Disease Surveillance System, the U.S. Army HIV Data System (USAHDS), the Acute Respiratory Disease Surveillance System (ARD), the U.S. Army Hospitalization Surveillance System (Individual Patient Data System [IPDS]), the Deployment Medical Surveillance System, and the Defense Manpower Data Center. Because each component database collects data on an individual person basis and because each database record is identified by Social Security number (SSN), the data in all of these systems are potentially linkable.

Much like the data gathering system for the Morbidity and Mortality Weekly Report of CDC, the Reportable Disease Surveillance System (RDSS) is a worldwide automated system that collects relevant information on a specified list of morbidity endpoints (reportable diseases—see list below). Data are gathered by Preventive Medicine Service personnel at U.S. Army installations. The individual case reports, including SSNs, are coded so that updated local, daily, weekly, and monthly reports can be generated by the Army Medical Surveillance Activity. Reportable diseases occurring among inpatients and nondeployed personnel are more likely to be reported than are those occurring among military personnel who are deployed for training or duty or who are being treated on an outpatient basis.

Notifiable Conditions Reported Through the Army Medical Surveillance System, January 1995

Amebiasis

Anthrax

Arboviral fever, unspecified

Asbestosis

Botulism (adult)

Botulism (infant)

Brucellosis

Campylobacteriosis

Carbon monoxide intoxication

Chancroid

Chemical agent exposure

Chlamydia

Cholera

Coccidioidomycosis

Cold weather injury (CWI)

CWI, unspecified

CWI, frostbite

CWI, hypothermia

CWI, immersion type

Dengue fever

Diphtheria

Ehrlichiosis

Encephalitis

Fatality, trainee

Fatality, occupational

Giardiasis

Gonorrhea

Granuloma inguinale

Guillain-Barré syndrome

Haemophilus influenzae, invasive

Heat exhaustion

Heat stroke

Hemorrhagic fever

Hepatitis A, acute

Hepatitis B, acute

Hepatitis C, acute

Hepatitis, unspecified

Herpes simplex

Influenza, unspecified

Influenza, type A

Influenza, type B

Kawasaki syndrome

Lead poisoning

Legionellosis

Leishmaniasis, unspecified

Leishmaniasis, cutaneous

Leishmaniasis, mucocutaneous

Leishmaniasis, visceral

Leishmaniasis, viscerotropic

Leprosy

Leptospirosis

Listeriosis

Lyme disease

Lymphogranuloma venereum

Malaria, unspecified

Malaria, falciparum

Malaria, malariae

Malaria, ovale

Malaria, vivax

Measles

Meningitis, bacterial

Meningitis, viral

Mercury intoxication

Mumps (adults only)

Mycobacterial infection

Pertussis

Plague

Pneumococcal pneumonia

Poliomyelitis

Psittacosis

Q fever

Rabies, human

Radiation, ionizing

Radiation, nonionizing

Relapsing fever

Reye's syndrome

Rhabdomyolsis

Rheumatic fever

Rift Valley fever

Rocky Mountain spotted Fever

Rubella

Salmonellosis

Schistosomiasis

Shigellosis

Smallpox

Syphilis, unspecified

Syphilis, primary/secondary

Syphilis, latent

Syphilis, tertiary

Syphilis, congenital

Tetanus

Toxic shock syndrome

Toxoplasmosis

Trichinellosis

Trypanosomiasis, African

Trypanosomiasis, American

Tuberculosis, multidrug-resistant

Tularemia

Typhoid fever

Typhus fever

Urethritis, nonspecific

Vaccine adverse event

Varicella, adult only

Yellow fever

USAHDS acquires and maintains the data for the Army's human immunodeficiency virus (HIV) testing program. This system reports the dates and results of tests, as well as medical information. USAHDS also maintains the registry of all individuals with confirmed cases of HIV infection, documents clinical evaluations at Army medical treatment facilities, and supplies database support to the Army/Navy Serum Repository.

All personnel are tested for HIV before entry into active duty and before overseas deployments, but no less often than every 2 years for active duty personnel when they are not on deployment status. Reservists are tested for HIV every 5 years. The unused sera from each HIV test are maintained in the Army/Navy Serum Repository. The availability of these serum samples is an important resource for research because the serum samples are identified by SSN and are thus linkable to all other identifiable data. They may be used in case-control as well as cohort studies, for example, analyzing predeployment and postdeployment serum samples to document various types of exposures.

The information for the ARD Surveillance System is collected only from the Army basic training centers. The data consist of counts of hospital admissions for ARD, throat culture results, and group A beta-hemolytic streptococcus-positive throat culture results. A weekly report tracks admissions for ARD and a surveillance index indicates the percentage of hospital admissions for ARD that are caused by group A beta-hemolytic streptococci. Data are aggregated at the unit level and are thus not directly linkable to other individualized data. However, because individual hospitalizations are also captured in the IPDS (see below), the data can be linked via the IPDS.

IPDS is an administrative database maintained under the U.S. Army Patient Administration Systems and Biostatistical Activity. Hospitalization discharge information on all military personnel, dependents, and retirees is maintained in this system, which includes up to 10 coded procedures and 8 coded diagnoses per discharge. Information is coded using the International Classification of Disease, Ninth Revision (ICD-9), and entered into the database from hard-copy records; information from hospitals in areas where personnel are deployed is coded and entered into the system on a delayed basis. Plans are under way to automate a tracking system (Army's Patient Accounting and Reporting Real-Time Tracing System) for in-theater hospitalizations throughout the Army. This system is now operational in Bosnia. Outpatient information is not included in this database.

The Deployment Medical Surveillance System maintains rosters of deployed forces for all major deployments. It has the potential to link the information on deployment cohorts with the medical event databases listed above. Using this system, one could, for example, compare deployed troops with matched controls and provide a link to their Army/Navy Serum Repository samples. Since deployed personnel and those receiving outpatient treatment may be missed in the reportable disease database, however, this linkage may not capture all the relevant health events.

Complete Army population data (demographic and occupational data) are linked with the Deployment Medical Surveillance System by SSN in a relational database by the Army Medical Surveillance Activity.

As of March 1996 the following data were stored on-line in a database available through the Army Medical Surveillance Activity.

Army Active Duty, Reserves, and National Guard

Individuals n = 3,398,778

Serum samples n = 8,227,156

Military Entrance Processing Stations (recruits)

Individuals n = 4,402,470

Serum samples n = 3,970,757

USAHDS

Individuals n = 5,680

Clinical evaluations n = 16,658

Reportable Disease Surveillance System

Years on-line 1994–1995

Total reports n = 11,862

Army/Navy Serum Repository

Total sera banked n = 17,134,763

Individual Patient Data System (IPDS—Active Duty, Army)

Years on-line 1989–1995

Total discharges n = 756,214

The Army Medical Surveillance Activity has the typical strengths and weaknesses of an automated multipurpose database, which are described in the last part of this chapter. The particular strengths of the Army Medical Surveillance Activity and the proposed Defense Medical Epidemiology Database (described below) include the following: a variety of outcomes databases that are all linkable by SSN; components that are, by and large, already on-line and working; currently generated weekly and monthly reports from the Reportable Disease Surveillance System; and relatively easy addition of surveillance endpoints to the list of reportable diseases.

The particular weaknesses of the Army Medical Surveillance Activity include the fact that the system is not yet incorporated throughout DoD (although plans for a Defense Medical Epidemiology Database are in progress), and few exposure data are available. In addition, no outpatient data are available in an automated system, and data for reservists not on active duty are generally not included in any of the databases discussed. Specifically, no automated systems collect baseline data on exposures or outcomes before deployment or follow-up data after deployment. This deficiency makes medical surveillance of the reservist population almost impossible, save for short-term surveillance when these personnel are on active duty. Given the general plan to reduce the number of active duty forces, necessitating an increased reliance on reservists, the inability to conduct surveillance activities among reservists is a substantial limitation for the study of interactions.

Other sources of information that may be incorporated into the Army Medical Surveillance Activity include the following: disability data, active duty Army hospitalizations at Navy and Air Force medical treatment facilities, and reportable diseases for Air Force and Navy personnel. In addition, health risk appraisal data are being incorporated into the system.

Plans to create a triservice surveillance system are under way. Another potential strength would be linkage to the USPDP (described below), which plans

to forge links with hospital, outpatient, and claims databases, in addition to collecting prescription drug information.

The Defense Medical Epidemiology Database will integrate the Army Medical Surveillance Activity with the epidemiologic capabilities of the Air Force and Navy to create a triservice surveillance system. There will be Internet access to triservice reports, and data will be available through the Defense Medical Epidemiology Network. SSNs will provide the means to a common linkage within and across service-specific databases.

USPDP is an automated multipurpose database of comprehensive prescribed drug information that serves the military health service system. Although many of its components are in place, some are still under development. The purpose of USPDP is to analyze the distribution of medications to troops and all DoD health care beneficiaries and to analyze the uniformity and consistency of the prescription benefit provided.

As of July 15, 1996, the database contained 28,493,924 records on the prescriptions filled for 2,058,010 participants over a 25-month period at 31 Army and Navy medical treatment facilities. Sixty-two percent of the prescriptions were new, with the remainder being refills. The mean age of the recipients was 32.6 years, and 49 percent of the recipients were female. Limiting the analysis to active-duty personnel, the 10 most common medications, ranked by therapeutic class and based on the prescriptions filled at 22 Army USPDP sites, are listed in Table 3-1.

The particular strengths of USPDP include the ability to describe drug use rates and characteristics and to examine health policy changes, drug safety, and drug effectiveness. The limitations of this system include the lack of immunization data, an inability to define a temporal association between exposure and outcome by using only the information in the database, the lack of information on other potential confounders such as smoking, and an inadequate sample size for studying very rare events.

Future plans for USPDP include the following: expansion of the system to include the Air Force medical treatment facilities and all of the remaining Army and Navy sites, performance of analyses standardized by time, inclusion of mail service prescriptions, and the gathering of additional data on prescriptions filled for deployed personnel. In addition, plans to link the hospitalization, ambulatory visits, and CHAMPUS (civilian-provided health care for military health care beneficiaries) claim databases are being developed.

A proposal has been submitted to expand the current system into a Universal Pharmacy Patient Profile. This would be a computerized system maintained by all pharmaceutical dispensers that will enable pharmacists to detect any potential drug interactions among current and newly prescribed medications across the entire military health service system, as well as provide a universal, automated record of the prescriptions that have been dispensed to an individual. In addition to including all of the medical treatment facilities, the Universal Pharmacy Patient Profile would create an electronic profile in a central database and would be accessible on-line by military and civilian pharmacies.

Although they are not strictly applicable to the study of interactions among deployed military personnel, the databases of the U.S. Department of Veterans

Affairs (VA) provide an important capability: the study of long-term health effects by linking military and veteran records. A number of the relevant databases are briefly discussed here.

One of the VA's larger automated health databases is the Patient Treatment File, which is analogous to the Army's IPDS. An automated record is created for every inpatient discharge from a VA medical center. Each record contains the individual's SSN and is thus linkable to the military database records. Although only about 12 to 17 percent of veteran hospitalizations occur within the VA system (Ahuja, 1994), this amounts to roughly 1 million episodes per year.

As is the case with the Army, no automated VA outpatient database exists. However, plans for such a system are under way, and an automated database is to be created starting October 1, 1996. As with inpatient episodes, relatively few veterans' outpatient visits (15 to 20 percent) occur in a VA setting (Ahuja, 1994), but again, the amount of data is large: more than 20 million outpatient visits per year.

The VA's Veterans Benefits Administration administers two databases that are potentially useful in epidemiologic studies. The Compensation and Pension file includes records of all veterans currently drawing compensation for a military service-connected disability. Although these veterans number some 2 million, the automated medical information available about them is limited; for example, diagnoses are recorded using VA-specific codes rather than International Classification of Disease codes.

The Beneficiary Identification and Records Locator Subsystem (BIRLS) is an automated administrative database that identifies all veteran beneficiaries (and, as of the early 1970s, all veterans) and gives the location of their VA claims folders, which contain the paperwork related to any claims for veterans benefits. Of particular interest is the fact that BIRLS contains a date of death for deceased veterans, and, because reporting of death to the VA is quite complete (Page et at., 1995; Page et al., 1996), this produces an opportunity to use BIRLS to follow up on all-cause mortality. In addition, death certificates are generally located in the claims folders of the deceased individuals, so that cause-specific mortality can be determined by using death certificates requested from the claims folder. However, the SSN is not always available for every BIRLS record, although it should be available for all recent records.

The strength of the VA databases for the study of interactions lies in their linkability to military records (via SSNs), which in turn provides opportunities to create automated multipurpose databases for long-term follow-up. In addition, the individual databases contain large numbers of records. Their main

limitation is their incomplete coverage of the veteran population, although less so in the case of mortality information.

Two civilian surveillance systems are relevant to the study of interactions: VAERS and the FDA medical products reporting system (MEDWatch). The DoD Directive on Immunizations and Chemoprophylaxis (Army Regulation, 1995) requires that adverse events related to vaccines be reported to VAERS; there is no corresponding regulation for reporting to MEDWatch.

It is important to emphasize that neither system contains records that are directly linkable to military databases because neither system's records contain SSNs. This sharply restricts the utility of these databases for follow-up studies of military populations. Instead, both VAERS and MEDWatch can serve as alerting mechanisms, uncovering potential interactions that could be further studied in the military system. Reporting of adverse military events to MEDWatch is not currently required but would materially strengthen that system.

VAERS was established in 1990 to collect reports of adverse events following vaccination with licensed products. This system is managed jointly by CDC and FDA.

The objective of VAERS is to aid in the detection of previously unrecognized reactions to vaccines, to detect any increases in known reactions, to identify preexisting conditions that may promote reactions, and to determine whether particular vaccine lots result in an unusual number or types of reported adverse events.

The primary data included in VAERS are date of birth, description of adverse event, outcome, date of vaccination, date of onset of adverse event, and all vaccines administered. Additional information includes patient recovery, test results and laboratory data, other vaccinations within four weeks of the reported vaccination, other medications taken at the time of vaccination, any illness at the time of vaccination, and preexisting conditions.

Most of the reports in VAERS are for children under 10 years of age. A summary of the information for reports for adults is provided in Table 3-2

(many of the vaccines listed in Table 3-2 are those that are administered to military personnel).

The information is supplied by the following four sources:

• Manufacturers, 39 percent

• Patient or parent, 2 percent

• Health care professional, 25 percent

• State health coordinator, 34 percent.

All deaths that are reported to the system are investigated by FDA physicians. Serious events are followed-up by VAERS contractors for recovery status, and selected cases are reviewed in depth by FDA physicians. Less serious events are occasionally investigated if unusual manifestations or complications appear.

VAERS is most useful for identifying rare and previously unrecognized reactions to vaccines, particularly newly marketed vaccines, and for monitoring the safety of individual vaccine lots. By directive (Army Regulation 40-562, 1995), adverse vaccine events among military personnel are to be reported to VAERS; typically, reporting to VAERS is performed by preventive medicine personnel.

VAERS is a passive system; therefore, underreporting, reporting bias, and the lack of a denominator (i.e., the total number of persons exposed or vaccines administered) create analytic challenges for epidemiologic studies. The estimation of true rates of vaccine-induced serious adverse events and the detection of small to moderate differences in the capacity of individuals to react to different vaccine lots are not feasible with the VAERS system. Establishment of causal relationships between vaccines and most types of adverse events is rarely feasible from the data.

MEDWatch was established in June 1993 to gather information on serious adverse events following the administration of drugs and biologics and on problems with medical devices regulated by FDA (Kessler, 1993). This system is managed by FDA.

The objectives of MEDWatch are to facilitate the voluntary reporting of serious adverse events and problems with drugs, biologics, and medical devices; to continue monitoring the safety of new drugs and devices; and to aid in the detection of previously unrecognized reactions.

The form used in MEDWatch is one page long, and the following primary data are included in this system: patient information (age, sex, and weight), adverse event, date of adverse event, description of adverse event, other relevant patient history, results of relevant tests, description of suspect medication or device, and reporter information. The patient's identity, if provided, is held in the strictest confidence by FDA. The reporter's identity may be shared with the manufacturer unless requested otherwise. However, the reporter 's identity is not released to the public.

The information to MEDWatch is supplied directly by physicians, pharmacists, nurses, and any other professionals working in health facilities or industries where drugs and medical devices are used and manufactured. This is in contrast to VAERS, which receives the majority of its reports from manufacturers (who are mandated to submit any reports that they receive from the medical community and from patients).

As with the VAERS system, the strengths of MEDWatch include the ability to detect a large number of serious adverse events whose similarities would normally remain undetected and the ability to recognize reactions that were not previously observed in newly marketed drugs. Limitations include the fact that this is a passive system, which can lead to underreporting, reporting bias, and the inability to determine a denominator (i.e., the total number of persons exposed or drugs administered). The ability to estimate the true rates of drug-induced serious adverse events and to establish a causal relationship is extremely limited with this passive system.

The FDA's involvement in postmarketing drug surveillance includes: monitoring approved drug use, monitoring the occurrence of serious adverse drug events associated with the use of approved drugs, and initiation of selected epidemiologic studies to estimate risk or test specific hypotheses (Arrowsmith-Lowe and Anello, 1994). Drug manufacturers are required to report adverse drug events that they are aware of, largely through spontaneous reporting from physicians and pharmacists, to the FDA (Kessler, 1993). In addition to conducting analyses of reported adverse drug events, the FDA supports a number of cooperative agreements to provide it with access to data on the safety of pharmaceuticals, including vaccinations. In general, the FDA's approach to postmarketing surveillance requires a variety of data sources, including the use of large, linked databases. However, it is noteworthy that the FDA lacks regulatory authority to require postmarketing or Phase 4 studies for approved drugs. The efforts of the FDA are often augmented by studies performed on a voluntary basis by pharmaceutical companies and, in some cases, academic and health care delivery organizations.

The surveillance tools currently available to the military comprise a series of potentially linkable automated databases (Grabenstein et al., 1992). Opportunities for creating automated multipurpose databases have already been seized, with the Army Medical Surveillance Activity and the USPDP being two examples. Additional opportunities to create linked databases remain, a fact recog-

nized in the planning for the triservice Defense Medical Epidemiology Database.

In general, the strength of these large databases is their outcome data, and their limitation is their exposure data, with prescription drug data being the exception. Data on the use of nonprescription drugs are not available, and vaccine data are not captured. No chemical and environmental exposure databases like the ones described above exist, but in some instances such databases are being created after the fact (see Chapter 5).

Although the availability of computerized diagnoses has been considered a strength of the automated databases described above, there is an accompanying, underlying weakness as well. The process of coding medical information in itself can change that information. Frequently, sentinel events will be harder to recover from a system once they have been coded. For example, the process of coding can inappropriately group together medical conditions, combining, say, 3 sentinel cases and 30 unimportant cases in a single general category. Although this does not preclude the use of automated diagnosis information as an alerting mechanism, it makes meaningful surveillance much more difficult. Because newly discovered interactions are not likely to mimic exactly previously described disease, and may indeed have unique presentations, the successful use of surveillance data to monitor interactions depends on problem recognition as well as reporting. Once a decision has been made to follow up some series of sentinel events, it may well be necessary to secure additional information on such cases by returning to hard-copy records for the appropriate level of detail. Finally, medical coding conventions lag behind developments in medical science and practice, so that the diagnoses that one may wish to retrieve from a system may not yet be available in coded form.

Given the inherent difficulties in identifying and confirming unknown interactions, the use of surveillance tools cannot ensure success. However, the current surveillance systems—with an expanded list of reportable conditions, additional linking of databases, and additional sentinel data from relevant civilian systems—can provide sentinel-event reporting for further investigation by the military. Furthermore, they represent one of the few practical ways to approach the problem of interactions.