Torrence, Mary E.
Medical device epidemiology is the study of the prevalence and incidence of use, effectiveness, and adverse events associated with medical devices in a population. Identification of data sources is needed for medical device epidemiology. Large data sources are frequently used for epidemiologic studies 1,2 because descriptive or case-control studies can be done quickly, a large population can be defined, and temporal trends can be determined. These attributes are important for regulatory or public health crises. Yet these data sources can be limited, expensive, and potentially biased. 1,2 The challenges in identifying data sources are finding data on specific devices, including manufacturer, brand, and model, and on patient exposure to those devices.
The Center for Devices and Radiological Health (CDRH) at the Food and Drug Administration (FDA) has identified both government and private sources of medical device data (Table 1). Government sources are usually more cost efficient and contain nationally representative populations so that national estimates can be determined. Data sources exist because of focused surveillance or administrative programs. Private sources are more expensive but can contain more specific, detailed data on devices and patients.
Agency for Healthcare Research and Quality: Nationwide Inpatient Sample
One of the most useful and cost-efficient data sources is the Nationwide Inpatient Sample (NIS). These data are from the agency’s Health Cost and Utilization Project-3. Data are provided on CD-ROM, and yearly updates are inexpensive. This data source is a 20% stratified probability sample of U.S. community hospitals and the hospital stay records. NIS contains all discharge data from approximately 1,000 hospitals located in 22 states, which represents about 7 million hospital stays. The hospitals are stratified by ownership/control, number of beds, teaching status, urban/rural location, and U.S. region. Long-term hospital facilities are not included. Data include demographics, primary and secondary diagnoses and procedures, admission and discharge status, length of stay, number of days of procedures, hospital characteristics, and expectant pay source. The typical discharge abstract contains patient-level clinical and resource-use information. Examples of medical device issues that might be investigated are the temporal trends in pacemaker use and the comparison of alternative cardiac procedures and outcomes.
The stratified probability sample allows a determination of a national estimate. Hospital and discharge weights along with information for calculating variances of estimates are included. Inpatient stay and hospital weight files link to patient demographic, clinical, and payment information. This database is good for implantable devices or when device use is associated with a procedure code (eg, pacemaker, pacing leads, or heart valves). 3 These data include hospital identifiers that permit linkages to the American Hospital Association’s database and county identifiers that permit linkages to the Health Resources and Services Administration’s Bureau of Health Professions’ Area Resource File. NIS is the only national hospital database with charge information on all patients, regardless of payer status.
Study is limited by the availability of inpatient data, and only devices used in an inpatient setting can be studied. A device cannot be studied if it cannot be linked to a procedure or diagnosis code. One of the biggest flaws is that the data are not current. Data may be as much as 4 years behind; therefore, newly marketed devices cannot be studied.
Centers for Disease Control and Prevention Surveillance Systems
The multiple surveillance efforts of the Centers for Disease Control and Prevention (CDC) provide an opportunity for the study of specific medical devices. The Dialysis-Associated Diseases Survey is an annual survey of all dialysis facilities regarding pyrogenic episodes, serologic tests, dialysis techniques, and disease-control measures. It also provides data on dialysis and water treatment systems. The National Nosocomial Infections Surveillance System, established in 1970, has selected hospitals that routinely report their nosocomial infection surveillance data for collection in a national database. 4 Since 1986, hospitals have had the flexibility to design their own surveillance programs. Surveillance components are hospital-wide, adult and pediatric intensive care units, high-risk nursery, and surgical units. Examples of possible devices to be studied are: devices related to vaginal and cesarean-section deliveries (eg, forceps and vacuum extractors), devices related to intensive care units and to high-risk nurseries (eg, catheters, ventilators, and monitors), and surgical devices (eg, coronary artery bypass graft, cardiac surgery, and joint prosthesis). Other possible CDC surveillance systems for data are the Behavior Risk Factor Surveillance System, National Program of Cancer Registries, and Pregnancy Risk Assessment Monitoring System.
These are multiple data sources from one agency that can contain device information. Questions can be added to the surveillance system. Cost and planning efforts vary with the programs. The Nosocomial Infection Program is the only source of national data on the epidemiology of nosocomial infections in the United States. 4 Service-specific, ward-specific, and overall rates can be calculated.
In the Nosocomial Infection Program, the amount of surveillance as well as the type of data collected varies among hospitals. All of the surveillance systems are dependent on participation. Some data are missing or inaccurate.
Centers for Disease Control: National Center for Health Statistics (NCHS) Surveys
This agency is responsible for numerous health surveys on a national level. Some of the surveys are the National Home and Hospice Care Survey, National Mortality Followback Survey, National Health Interview Survey, and National Health and Nutrition Examination Survey. The National Health Interview Survey is a principal source of information on the health of the civilian U.S. population and is used to monitor trends in disability and illness. It is a cross-sectional survey with a multistage area probability design to achieve representative sampling of households. The questionnaire topics are basic health, demographics, and current health. The current health topics change depending on data needs. CDRH used the 1988 Medical Device Implant Supplement to the National Health Interview Survey to provide a nationally representative study of the prevalence of implants including artificial hips, knees, and intraocular lenses. 5,6 The National Health and Nutrition Examination Survey consists of a standardized physical examination including blood samples, health interview, and laboratory testing to provide data on the health and nutritional status of the population. It is used to estimate the number and percentage of the population with selected diseases and risk factors; monitor trends; and determine the relationship of diet, nutrition, and health. CDRH has utilized this survey to look at the prevalence of latex allergies in the population. In the National Home and Hospice Care Survey, agencies providing home health and hospice care are sampled. In addition to questions on assistive devices such as wheelchairs, crutches, contact lenses, and hospital beds, CDRH added questions to look at certain traditional medical devices that are being used in the home such as ventilators, oxygenators, nebulizers, catheters, infusion pumps, blood glucose monitors, and restraints. The National Mortality Followback Survey is a multicomponent survey of a nationally representative sample of individuals more than 15 years of age who died in a particular year. This survey uses information from death certificates, other administrative sources, and an interview with a relative of the deceased. It collects information on utilization and access to health services, activities, drug and alcohol use, medical device use, industry, and occupation. This information looks at the etiology of disease, demographic trends in mortality, and other health issues. CDRH used the 1993 National Mortality Followback Survey to determine the quality of life in pacemaker recipients. 7
The samples are population based, so national estimates can be determined on various health issues. Several health issues can be studied at the same time. Interviewers are trained. Other surveys can be linked to the National Health Interview Survey, such as the National Survey of Family Growth.
Studies are expensive and there is a significant time lag. Only commonly used devices can be studied, and each survey must be planned years in advance.
Department of Defense: The Birth Product Line Study
This study initiated by the Department of Defense included approximately 9,714 women admitted, with a resultant delivery, into one of the 97 military facilities from October 1992 through September 1993. Complete data were collected from a randomized sample of admissions from each military facility. Information was abstracted from medical records from the first prenatal visit to discharge of the mother and neonates. More than 800 relevant clinical variables were identified to characterize the demographics, conditions, and care of the patients. These data have been used by CDRH to look at chorionic villus sampling, and the risk of congenital defects, as well as methods of assisted deliveries and possible risk factors.
Numerous areas of prenatal information can be studied because data such as postpartum conditions, marital status, prenatal screening tests, level of education, follow-up visit, and readmissions are collected. Chart abstractors are trained.
The data are older, and some prenatal care has already changed. Data may be incomplete and require extensive labor to obtain. The population represents a specific subgroup and findings may not be generalizable. Military care may differ from care given in other settings.
Centers for Medicare and Medicaid Services
Medicare data can be used to study devices used primarily or exclusively in the elderly (65 or over) and reimbursed by the Health Care Financing Administration. The studied device needs to have separate codes for implantation, removal, and revision and to have associated relevant outcomes that are detectable in administrative data. Examples of studies that have been done are: incidence of implantation and outcomes for pacemakers, 8 short- and long-term gender-specific effects and survivability related to automatic implantable cardioverter defibrillators, and home medical device use (eg, oxygen, wheelchairs, hospital beds, or dialysis) by Medicare beneficiaries .9
The availability of both patient and physician location allows consideration of regional trends. It is a large computerized database, covering the majority of elderly in the United States, that contains cost and utilization data. It is a source of denominator and descriptive data for certain devices such as implants. The procedures and diagnoses are recorded in standard coding conventions.
This database requires considerable expertise to use properly and is costly. The data are not generally brand specific unless the device is the only one. No prescription data are available. The participants in managed care programs must be excluded from the analysis because claims data on these beneficiaries may be incomplete. This may become more of a problem as more of the elderly are encouraged to enter managed care. Some studies suggest that managed care plan members differ from standard Medicare recipients according to some important demographic characteristics. 10 Newly marketed devices or procedures may not have unique national codes, and there is a lag time for code assignment for new technologies. Codes are not always correct.
Consumer Product Safety Commission: National Electronic Injury Surveillance Systems
This data source is maintained by the Consumer Product Safety Commission and has been in existence over 20 years. This sentinel surveillance system was designed to capture information on product-related injuries that result in hospital emergency room treatment from a statistically representative sample of hospital emergency rooms. Data are collected on a broad range of injury-related issues and hundreds of product categories. Estimates can be made of the number of injuries associated with products that result in treatment at hospital emergency departments.
This data source contains statistically selected hospitals for participation that are representative of similar hospitals across the nation. Events that are reported are differentially weighted to produce national estimates of number and severity of injuries related to specific products. Data collectors are trained, and data can be electronically transmitted.
Although data already collected are free, adding questions about specific medical devices can be costly. The data are not specific for devices. Only information on events that result in an emergency room visit is available. Infections that might occur later cannot be related directly to devices.
Private Data Sources
The term “registry” differs in meaning depending on the agency or institution. A registry may simply be a database that is established to capture information for a specific clinical trial. Conversely, it may capture information on patients either undergoing a specific procedure or having a particular disease syndrome. Registries exist both in governmental agencies and in universities. For example, CDC maintains the National Program for Cancer Registries. The University of Pittsburgh maintains a dynamic registry for cardiovascular disease. Other registries include the Juvenile Rheumatoid Arthritis and the Chorionic Villus Sampling Birth Defects.
Registries can provide follow-up for a cohort. Registries are good for capturing dynamic information about rapidly changing technologies, eg, coronary stents. The usefulness and data vary with the purpose of the registry.
Registries are sometimes only set up to collect information for one focused study. Registries are expensive to maintain, and data are often limited. In addition, registries focused around one specific disease or condition may be biased in participation.
IMS America Hospital Supply Index
The IMS America Hospital Supply Index monitors medical-surgical supply purchases by 350 hospitals in the United States. It tracks more than 240,000 products in more than 1,500 product categories such as cardiovascular products, diagnostic supplies, dialysis and orthopedic supplies, and urological products. With these data, the total sales of particular device brands and models can be estimated and used as a rough estimate for a denominator. This data source provides information on the manufacturer and distributor; product name; manufacturer catalog number; item description; dollar and unit volumes; and region, size, and type of hospital.
The comprehensive estimates are impartial and the vendor is customer-oriented.
It takes several months to get the most current data that do not reflect user characteristics such as age and sex. There is no standard code for specific brands and models. Users need to be familiar with the particular study device to decipher the meanings of the model or catalog numbers reported in the data. Unused, wasted, or even reused devices are not accounted for because the company only documents paid invoices. Data for other settings are available separately and are costly. The range of products available for the study depends on the extent of the plan’s policy.
Managed Care Organizations
The existence of managed care organizations offers the researcher a large, population-based data source. Although each organization may have variable data and specificity, these data often have centralized, computerized diagnosis information. Outpatient pharmacies may or may not be computerized. Differences in coverages may prevent generalization. As managed health care coverage continues to expand, these sources may become more important to consider.
Harvard Pilgrim Health Care
This database is a fully automated medical record system with computerized pharmacy records. It is the largest staff model managed care organization in New England with around 300,000 active members and an additional 700,000 historical members. Records have been available since 1990. Currently CDRH has done exploratory, descriptive studies to determine what devices could be studied (eg, cardiac devices, ventilators, or catheters). A recent study determined risk factors for mammography access. 11
All encounters are electronically recorded except for x-ray images and printed monitor strips. Most of the health care notes are searchable and coded. Implants and outpatient prescription devices have fully patient-linkable data. These data are available to do a study quickly with a large population. It does contain implant utilization, brand, and model information.
Medical equipment is not fully patient-linkable if the facility has more than one machine that may be used for the specific procedure. It is not good for inpatient drug exposure, and only serious diseases can be studied. The range of products studied depends on the health plan, and the data elements are variable. Often there is incomplete information on potential confounders. The results may not be generalizable.
The identification and use of large data sources for medical device epidemiologic studies are important. This information is needed to supplement current knowledge for public health and regulatory decisions. Both government and private data sources were identified. The government sources tended to be more representative of the population, were easily accessible, and cost effective. However, specific data regarding medical devices were limited. Data involving medical devices in women were even more limited. Private sources are more cost prohibitive but provide an opportunity for more focused and specific research. Increased awareness about medical devices is essential to enhance the ability of both government and private institutions to collect data on medical device usage. Expanded collaborations among institutions and agencies will also improve the availability of medical device data. As the percentage of women increases in the elderly, more attention is needed for study of women’s health care issues, specifically medical device use in women.
1. Stom BL, Velo G, eds. Drug Epidemiology and Post-Marketing Surveillance. New York: Plenum Press, 1992.
2. Strom BL, Carson JL. Use of automated databases for pharmacoepidemiology research. Epidemiol Rev 1990; 12: 87–107.
3. Daley WR, Kaczmarek RG. The epidemiology of cardiac pacemakers in the older US population. J Am Geriatr Soc 1998; 46: 1016–1019.
4. National Nosocomial Infections Surveillance System. Methods and Analysis. Atlanta: Hospital Infections Program, National Center for Infectious Diseases, Centers for Disease Control and Prevention, 1993.
5. Silverman BG, Gross TP, Kaczmarek RG, Hamilton P, Hamburger S. Epidemiology of artificial knee implantation in the USA. Knee 1995; 2: 95–102.
6. Sharkness CM, Hamburger S, Moore RM, Kaczmarek RG. Prevalence of artificial hip implants and use of health services by recipients. Public Health Rep 1993; 108: 70–75.
7. Hefflin BJ. Final-year-of-life pacemaker recipients. J Am Geriatr Soc 1998; 46: 1396–1400.
8. Lamas GA, Pashos CI, Normand SL, McNeill B. Permanent pacemaker selection and subsequent survival in elderly Medicare pacemaker recipients. Circulation 1995; 91: 1063–1069.
9. Silverman BG, Gross TP, Babish JD. Home oxygen therapy in Medicare beneficiaries, 1991–1992. Chest 1997; 112: 380–386.
10. Morgan RO, Virnig BA, De Vito CA, Persily NA. The Medicare-HMO revolving door: the healthy go in and the sick go out. N Engl J Med 1997; 337: 169–175.
11. Barton MB, Moore S, Shtatland E, Bright RA. The relation of household income to mammography utilization in a prepaid health care system. J Gen Intern Med 2001; 16: 200–203.
© 2002 Lippincott Williams & Wilkins, Inc.