Obtaining Population-Representative Data from Nonclinical Samples

An important contribution of the combination of bioindicators with data from social surveys is the documentation of function in nonclinical populations. Estimates of prevalence based on clinical populations—and even on some of the more customary epidemiological field studies—depend on highly selected samples, difficult to control and difficult to generalize to the entire population. Diagnosis of an illness depends upon numerous factors. It is affected by the probability that a person recognizes and acknowledges an “aberrant” status; by the likelihood that he will seek treatment; and by the location at which treatment is sought, the resources available for treatment, and compliance to the treatment protocol. We know that there is heterogeneity in all of these factors. Women, for example, are more likely to seek medical attention than men, especially for certain kinds of illness (Kroenke and Mangelsdorf, 1989; Cohen and Rodriguez, 1995; Walling et al., 1994; Russell et al., 1992). Examination procedures, test protocols, and guidelines for the administration of protocols may depend on where the patient presents (Aaron et al., 1996). And access to care and the quality of care are tied to social and financial resources. Estimates of the prevalence of a disorder and of the experience and attributes that characterize those whom a disease affects are therefore intertwined with differential rates of care-seeking behaviors and modes of therapy. Heterogeneity in these behaviors across sex, ethnicity, and position in social orderings can bias assessment of risk and of the burden of disease. The advantage of the social survey machinery is clear: a large, well-drawn sample allows us to draw inferences about population parameters; it allows comparisons within population subgroups and across time and place.

While data from both the HRS and the Taiwan Study can be examined cross-sectionally, the advantages of their longitudinal design are especially compelling. In particular, a longitudinal design can provide information on population-representative disease trajectories. A longitudinal design allows one to track the emergence of disease, to identify the important correlates of differences in disease trajectories, and to explore how differences in outcomes are related to differences in earlier experiences and behaviors. For example, longitudinal data would allow one to focus on people who make a transition between rounds. Such persons could be screened for characteristics that might have identified them earlier as being at elevated risk for illness. Longitudinal data would also make it possible to follow outcomes relative to behavioral responses.

Another important contribution of the longitudinal design is the documentation of patterns of biological markers across both an individual's life span and across the population. How, for example, do biological markers themselves change as a function of change in the environment? In the context of the HRS, one could also ask questions about the degree to which individual and family choices about medical care, purchase of insurance, or asset accumulation or “decumulation” interact with biological indicators. As scientific knowledge about the relationships among biological, health, and socioeconomic spheres expands, it may also become possible to gauge the potential for behavioral responses in these and other domains by households, insurance companies, health providers, and government policymakers. Of course, troublesome ethical and confidentiality issues concerning HRS respondents arise precisely because of these potential interactions and behavioral responses.

Even in the cross-section, however, the data can be valuable. For example, increasingly, we have come to understand that there is remarkable diversity across populations in biological parameters (Weiss,1998a; Campbell and Wood, 1994). Systematic cross-cultural studies may reveal differentials that are of substantive interest. Results from the Taiwan data, for example, can be compared with analogous U.S. data from the MacArthur Study of Successful Aging. The MacArthur sample was drawn from participants in three Established Populations for the Epidemiologic Study of the Elderly. The three communities were in the Eastern United States. The participants responded to a 90-minute interview in which health status was assessed, and they provided blood specimens and 12-hour overnight urine specimens (Berkman et al., 1993; Seeman et al., 1997). The participants ranged in age from 70 to 79 years old (the Taiwan participants were 67 to 94). Results for a number of parameters differ between the U.S. and Taiwanese elderly. Blood pressure readings were higher among the Taiwanese than the U.S. sample. Unlike in the United States, women in Taiwan had slightly higher readings than men. The waist/hip ratio among Taiwanese men was lower than their U.S. counterparts, but higher among Taiwanese women than among U.S. women. The ratio of total to HDL cholesterol for both Taiwanese men and women was lower than in the United States, but the sex differential was reversed: in Taiwan, the ratio was higher for women than for men. Glycosylated hemoglobin was lower in Taiwan than in the United States among both men and women. Average levels of DHEAS were higher for both men and women in Taiwan than in the United States, and HDL cholesterol—at least among men—was also higher. On average, using an invariant scale, these differences put the Taiwanese elderly at higher risk than the Americans of health disorders related to hypertension, and at lower risk (despite the slightly higher average age of the Taiwan sample) of illness related to indicators of cholesterol (ratio of total to HDL), the waist/hip ratio, glycosylated hemoglobin, and DHEAS (Goldman et al., 1999). But how these differentials change over time, what their environmental and behavioral correlates are, and how they are connected to differentials in health outcomes are questions that can be addressed only by the use of longitudinal data.

Calibrating Self-Reports with Other Measures of Health and Disease

The use of self-reported health-related information from surveys has the advantage—relative to clinical or field studies—of data from a representative sample of the population, but self-reported data (like other sources) are subject to a variety of errors. Evaluations of the quality of health-related data that have been collected through self-reported (or proxy) responses to survey questions have been performed in various ways, including comparisons with medical records and with physical exams (Beckett et al., 2000; Haapanen et al., 1997; Edwards et al., 1996; Strauss and Thomas, 1996; Turner et al., 1997). The addition of biological indicators to survey machinery provides another powerful, direct way to assess and calibrate self-reported data for certain kinds of information.

Reliability over time of self-reported information from health interview surveys may be inadequate for many purposes. A recent study by Beckett and her colleagues (2000), for example, shows that the consistency of reports of diabetes across surveys taken between 1989 and 1996 as part of the Taiwan Study of the Elderly averaged just over 80 percent. In the United States, the analogous figure from the National Health and Nutrition Examination Survey (NHANES) (1971/1975-1982/1984) was 78 percent. Reports of hypertension in Taiwan were even less consistent (averaging 73 percent), although the corresponding figure for the United States was higher—86 percent.

Estimates of validity, which are even bleaker than the estimates of reliability, serve to underscore the need for calibration. For example, fewer than half of the respondents who were identified as hypertensive based on (objective) blood pressure readings in the NHANES-I reported the condition in the interview (Beckett et al., 2000). In Taiwan, the figures for high blood pressure are equally troubling. Only two-thirds of the participants in the 1998 biological marker study who were identified as having at least moderately high blood pressure (>160 mgHg) based on the in-hospital measurement reported having the condition in the concurrent health interview; only 60 percent reported it at the time of the 1996 interview. These estimates from Taiwan are are not inconsistent with results from other studies. In Pakistan, for example, based on clinical evaluations that were conducted as part of the National Health Survey of Pakistan, 80 percent of men who were clinically diagnosed as hypertensive were unaware of the condition; the corresponding figure for women was 60 percent (Pappas, 2000).

Whether these discrepancies arise from the respondents' lack of awareness of the condition, failure to recall information provided by their physicians, or an unwillingness to report the condition, they can have an important impact on health, on estimates of the burden of disease, and on health policy.

Diabetes is a particularly important example because so many aspects of the condition are amenable to behavioral management. The earlier the diagnosis, the better the opportunities for behavioral modifications that can ameliorate the disease. The ability to screen for the condition in a representative and longitudinal study provides an important vehicle for its identification and remediation. The results from Taiwan illustrate this point. In 1998, only 56 percent of the participants with elevated glucose levels reported having diabetes to the examining physician as part of the disease history that was collected at the same time as the biomarkers. Only 48 percent reported the condition at the time of the 1996 interview.

A second important example, particularly among the elderly, relates to cognitive function. Currently, the HRS is considering a proposal to conduct a clinical assessment of dementia on a subset of respondents. A syndrome with several underlying causes and rapidly increasing incidence rates at older ages, dementia is projected to become an increasingly important burden to families and the public as rates of competing causes of death fall and more people survive to advanced ages. The HRS has the potential to become a valuable source of information on the economic and social burden of dementia for a nationally representative population. It provides detailed longitudinal information from the respondent or a proxy on their assets, income and health care utilization and costs, income and asset transfers between respondents and their children, assessments of informal and formal care, measures of cognitive functioning, and measures of basic and instrumental activities of daily living (ADLs/IADLs).

At the present time, however, the HRS is limited by the lack of a clinical diagnosis of dementia. Such a diagnosis would allow more precise cost estimates as well as additional epidemiological analyses. In response to this need, and with the advice of experts on the diagnosis and economic impact of dementia, the HRS is developing a supplementary proposal to conduct in-home clinical assessments of the dementia status and severity of a stratified random sample of 500-700 respondents, possibly with longitudinal follow-up to clarify the diagnosis of those with ambiguous status. This supplementary project would provide a highly valuable body of data on the prevalence of dementia in a nationally representative sample, together with a rich array of information with which to study the dynamics over time of the burden of this disease on families and the public. If the clinical diagnosis proves to be sufficiently well correlated with survey measures of cognitive decline and disability or Medicare diagnoses, it may be possible for researchers to impute the dementia status of HRS respondents who are not assessed in the supplementary clinical protocol.

An additional benefit related to “calibration” is made possible by combining survey data with biological specimens. The combination would provide significant material relative to documenting and understanding characteristics of individuals who provide self-reports that differ from the objective measures. The data could offer insight into how those characteristics are related to differentials in reporting. For example, based on analyses of the consistency over time of self-reports alone, Beckett and her colleagues (2000) found that in the United States differentials were related to sex and to cognitive status (women and persons with better cognitive scores were likely to be more consistent reporters). In Taiwan, where information on the severity of the condition was also available, it proved to be an important predictor of consistency. Objective measures of function would provide data that would permit us to assess validity as well as consistency.

Explicating Pathways and Elaborating Causal Linkages Between Social Environment and Health

Health at any age, but especially among the elderly, embodies and reflects the accumulation of complex interactions among genetic endowment, developmental influences that affect gene expression, and life experience including environmental exposures. A large literature in sociology, economics, and epidemiology has been devoted to documenting and exploring the associations between the social environment and health outcomes; another area of intensive research has been the relation between challenge (i.e., stress-provoking experience) and health. Figure 11-1 provides a diagram of a highly simplified model of the theory that underlies much of this research. It illustrates two important points. First, it makes explicit the physiological interactions with health, challenge, and the social and economic environment that are so often treated as implicit or “unobservable” by most social research. Second, it shows individual biological response linked to collectives via membership in networks and social aggregates and to choices about lifestyle, work, consumption, medical care, and insurance that influence and are influenced by health at the individual and familial levels. The importance of group membership—its power to parsimoniously explain individual behaviors—has long been recognized by social scientists; indeed, acknowledgement of such emergent properties of groups and their effects is, arguably, a hallmark of sociological research. Likewise, the role of prices and incomes working through markets and public policies are emphasized in economic models of health. The inclusion in Figure 11-1 of physiological function makes explicit the linkage between (aggregate) social and economic processes and the biological processes of an individual.

FIGURE 11-1

Linkages among the social environment, health, life challenge, and physiological response.

Over many years, social scientists have examined—and they continue to explore—the relationships among health, position in social hierarchy, and social connection that are shown in Figure 11-1. It is now well established that there is a simple, direct, monotonic relation between socioeconomic status—measured across the multiple dimensions of income, education, and occupation—and both morbidity and mortality. These differentials have been found across the full range of these hierarchies (Adler et al., 1994; House et al., 1994; Marmot et al., 1996) and have been replicated across the life span, across sex, and across time and place. Similarly, a large literature dating back as far as the 1800's (Goldman et al.,1995; Hu and Goldman, 1990) has established large differences in mortality associated with marital status; married people live longer than those who are not. This result—that better social connection is conducive to reduced mortality—also holds for nonmarital ties: higher levels of social integration and having close relationships lead to better survival and health outcomes (Thoits, 1983, 1995).

The bottom portion of Figure 11-1 links exposure to challenge with health. Again, a large and growing literature has established an association between stressful experience and health. Exposure to challenge is associated with (both) heightened and depressed function of the autonomic nervous system, with cardiovascular health, with gastrointestinal stability, and to immune response. Specifically, stressful experience has been shown to be associated with asthma, diabetes, gastrointestinal disorders, myocardial infarction, cancer, viral infection, autoimmune diseases, depression, anxiety and other psychiatric disorders, task management, and memory (Lupien et al., 1994, 1995; Weiner, 1992; McEwen and Stellar, 1993).

The social environment affects exposure to challenge and it mediates the effects of challenge on health. Stressful experience is more frequent among those who are lower on SES ladders, and access to the resources that can potentially cushion its effects is more limited. A strong social network can reduce the physiological burden of challenge by affecting both the perception and interpretation of experience as stressful and by providing a safety net to lessen its effects.

Research from the HRS both confirms the well-known strong correlation between health and economic status and extends our understanding of it in several important ways. For example, the correlation between health and wealth, which can be measured for the first time in the HRS, is even stronger than the correlation between health and income. While many researchers have shown that socioeconomic status has a strong effect on health during earlier portions of the life cycle, studies using the HRS have shown that health has a powerful impact on the income and wealth of individuals over the age of 50. Although some of this influence operates through medical expenses, especially for the uninsured, a large portion of the effect operates through the impact of health on labor supply and earnings (see Smith, 1999).

Analogous work on the Taiwan survey data has established that, as expected, multiple dimensions of health are influenced by position in social hierarchies, by social networks, and by exposure to challenge. The probability, for example, of being unhealthy in 1996 (or having died by 1996) is significantly related to paternal socioeconomic status and to the respondent's education; to contact with friends, neighbors, and children; to participation in social activities; to exposure to chronic financial difficulties; and to having a spouse in poor health or a spouse who died in the recent past (Beckett et al., 1999). Based on the self-reported data, Yamazaki (2000) has shown that depression (measured using the CES-D, i.e., the Center for Epidemiological Studies Depression Scale) is significantly increased by exposure to challenges in the preceding three years and by recent daily strains, and significantly reduced by affectional exchanges (see also Ofstedal et al., 1999).

Apart from the HRS and the Taiwan Study, these associations have all been well documented by repeated, large-scale surveys. Still, little is known about how the social environment is linked to the physiological factors that influence well-being and mortality, so the biological pathways through which the social factors influence health remain largely unspecified. Similarly, while we are increasingly able to document the scale and dimensions of the (physiological) stress response, we have only a limited understanding of how that physiological response is affected by social factors. Without joint biological and social data, the linkages among challenge, social environment, and mental and physical health are certain to remain “in the black box.”

Social/psychological studies are beginning to be used to unpack this black box in a number of ways. The Taiwan Study, for example, has been designed specifically to allow us to specify more completely the effects of challenge on biological markers; to relate self-reported physical decline and measures of cognitive function to those biological markers; to explore the consequences for health of cumulative challenge, social advantage, and adversity; and to explore sex-based differences in these factors.

We found that the biological markers collected in the Taiwan pilot study were consistent with results based on the self-reported data and that they offered some help toward progress in elaborating the pathways. Using an index developed by the MacArthur Study of Successful Aging (Seeman et al., 1994, 1995) to summarize biological function, Goldman and her colleagues (1999) found that the index was related in expected directions to self-reported measures of both physical and mental well-being and function, to education and sex, to economic and financial status and strains, and to important life events including the death of a spouse or of a child.

To date, the HRS has relied completely on self-reports of an individual's subjective health status, health conditions, and functional status, and on their self-reports of medical utilization, health insurance, and out-of-pocket costs. Only in the case of cognitive status and depression does the HRS provide measures that are similar in part to those that would be obtained in a medical setting. Sometime this year, the HRS will complete a linkage with the Medicare records of respondents age 65 and over which will give researchers access to physician's diagnoses and administrative records of care received.

Just as biological markers can be used to elaborate links among health and social, demographic, and economic factors, the addition of the markers, and more generally, of health-related information to social surveys can provide important, much-needed assistance in confirming and disclosing medical hypotheses. Insights into the relationship between health and disease might also emerge from examining the combination of self-reported data on health and biological markers associated with particular diseases. As Weiner (1992:91-92) persuasively argues, health and disease are not the same: “A patient may have a disease and either be in good health or be ill. Conversely, a patient may be in ill health without having a disease … Most patients seeking medical care do not have diseases but are in ill health … Persons in ill health express their distress in bodily symptoms … yet the relationship of one to another manifestation of ill health … is largely unsolved.”

The utility of large-scale surveys in identifying constellations of illnesses and their diagnostic markers and correlates, particularly in relation to health conditions linked to dysregulation of the central nervous system, is great. These conditions include fibromyalgia, chronic fatigue, irritable bowel syndrome, and migraine headaches, hypothesized to be caused by the interaction of both genetic and environmental factors resulting in changes in immune function (Clauw and Chrousos, 1997). The environmental factors include physical trauma (automobile accidents, for example), but also emotional stressors, particularly experiences which are perceived as inescapable, which are unpredictable, or which occur without emotional supports. The medical community has increasingly become sensitized to, and aware of, the importance of behavioral and social pathways affecting disease onset and severity. The collection of psychosocial data as part of a medical history has become more widespread. But most medical data are collected from nonrepresentative populations.

Although we are not advocating the transformation of social or demographic surveys into medical investigations, they clearly have the potential for informing such work. The extent of the contribution will almost certainly depend on disease prevalence and heterogeneity in its distribution across the population.

Linking Genetic Markers with Survey Materials

The rapid advance of knowledge in genetics and the relationship of genetics to health and behavior have led to suggestions that we collect biological specimens for genetic typing from respondents of population-representative surveys. In particular, the discovery of the link between late onset Alzheimer's disease and the APOE e4 genotype (Corder et al., 1993) provides an example of the potential predictive value of genetic information in the context of studies of older persons. Recently, the National Long Term Care Survey (NLTCS) used cheek swabs to obtain DNA samples from which information about a respondent's APOE genotype could be determined.

Wallace (1997) discusses the value of collecting genetic information in population surveys, including both pedigree data and specimens from which DNA may be obtained and coded, with an emphasis on the determination of genotype frequencies in well-defined populations. What has to date received less attention is the potential value of genetic information to population surveys, such as the HRS, whose major goals involve understanding determinants and consequences of behaviors related to work, earnings, saving, retirement; to heath status, health care expenses, and utilization; and to formal and informal caregiving and intergenerational transfers of time and money. The authors of the present chapter, who themselves have no special expertise on genetic issues, hope that broader discussion will begin to clarify this potential. Meanwhile, in this section we offer some speculation about some possible uses of genetic information in social surveys, particularly in the HRS, and some discussions of limitations and dangers in the use of the HRS for this purpose.

As described by Hermalin (1999), if genetic markers are modeled as part of an individual's physiological structure, they can provide controls for predisposing factors that affect more proximate mid-level markers of function as well as downstream health outcomes. This potential benefit of genetic information—i.e., its power in explicating the black box of Figure 11-1—may outweigh, or at least precede, its near-term potential for discovering genetic links to chronic disease. As discussed by Weiss (1998b), the situation with chronic disease differs from single locus disorders that are inherited following well-identified Mendelian rules. In general, we cannot expect to find relationships that are even as straightforward as the APOE links to cardiovascular and Alzheimer's disease. Variation across populations, difficulty in identifying a small enough area on the chromosome to search for disease-associated genes, and the problems inherent in identifying continuous outcomes with particular genes may limit finding the connections.

Our experience with APOE genotyping in Taiwan underscores some of these points. In our pilot study, of the 110 blood specimens which were typed, only 18 cases (16 percent) were e3/e4 and no one was identified as e4/e4. This distribution is consistent with other estimates of prevalence of the e4 allele in Chinese populations (see, for example, Ewbank in this volume), but with such a small sample and such a low prevalence we found no links between APOE type and an individual's health outcomes. Whether the upcoming larger sample of 1,000 will provide enough power to discriminate signal from noise remains to be seen.

Consideration of the HRS suggests additional concerns. As knowledge about the relationship between genetics and health increases, it is quite likely that private and public institutions and individual behavior will be affected in profound ways. One example concerns the availability, pricing, and demand for health insurance. If genetic information at birth is predictive of health conditions that occur later in life, will there emerge a market for “genetic insurance”? Providing insurance to populations with heterogeneous risks is already an important and difficult issue both at the theoretical level and for practical policy, in part because differences in information between those supplying and demanding health insurance lead to possibilities of adverse selection and instability of insurance markets (see Phelps, 1992, for a discussion). Moreover, the inequalities in access to or pricing of health insurance that may arise because of advances in genetic understanding are similar to those already confronted by individuals with differing pre-existing conditions. Collection of genetic information on HRS respondents might enable future researchers to study how individuals and families respond to new sources of perceived risk and attempt to deal with these risks through existing or newly created institutions. Other examples of potential research interest might be the role of genetics in the intergenerational transmission of health status or studies of the interaction between gene expression and measures of the economic and social environment over the life cycle.

These examples of research possibilities from the collection of genetic information in the HRS also serve to warn us about the limitations and dangers of the HRS as a vehicle for such data collections. The most obvious limitations of the HRS are (a) that data collection begins at a relatively late age, thus potentially obscuring and confounding important gene-environment-behavioral interactions that begin earlier in life,² and (b) that HRS respondents consist of genetically unrelated individuals (i.e., spouses), thus reducing a researcher's capacity to distinguish genetic from behavioral and environmental effects.

The collection of genetic information could also pose a serious threat to the value of the HRS. First, it is possible that HRS respondents would be frightened or alienated by requests for biological specimens. In this regard, the experience with the Taiwan survey effort reported in this paper provides some reassurance as does the experience in collecting cheek swabs in the NLTCS. Second, depending on the nature of the informed consent that would be used to collect specimens, there is considerable scope for contamination of the behavior of HRS respondents. For example, if we were to report back to a respondent that he had the APOE e4 genotype and explain what that implies for the probability of being stricken with Alzheimer's disease, the respondent might try to purchase long-term care insurance, increase asset accumulation to pay for care, or “spend down” assets to qualify for Medicaid, experience divorce, and so on. This scenario suggests that it is important to craft informed consent agreements in ways that consider the inevitable tension between the possibility of affecting the behavior of respondents with our overarching obligations to the participants. We do not see a simple solution to this problem.

Respondent Burden

For studies that have been driven primarily by questions related to the social sciences—whether demographic, economic, or sociological in nature—the additional burdens imposed on the participants by the collection of bioindicators may constitute an important potential concern. We see two particularly relevant aspects: first, whether agreement to participate in the bioindicator study itself is adequate; and second, whether participation in subsequent rounds of interviews is compromised.

For the HRS, it will be important to consider the impact on respondent cooperation before deciding to engage in the collection of clinical data or biological specimens. Indeed, if the HRS proceeds with the clinical dementia assessments described above, a careful study would be carried out to determine the effect of this additional effort on respondent cooperation in future waves of the survey. Similarly, any effort to collect biological specimens would first be piloted on a subset of respondents to determine effects on survey participation rates.

To date, the experience in Taiwan and elsewhere has been encouraging. In Taiwan, participation in the biomarker project was high. Most refusals to participate were unrelated to the collection procedures. Just under 72 percent of the targeted respondents completed the full protocol: they replied to a short survey that was administered by a public health nurse; they collected an overnight 12-hour urine specimen; they underwent a physical examination and provided a brief medical history in hospital; and they had blood drawn and provided a spot urine specimen in hospital. Another 23 percent of the targeted respondents could not be located or refused to participate. Of these nonparticipants, about a quarter were out of the country or out of reach of the hospital (temporarily located in another city, for example). More than a quarter had just recently completed a physical exam (either privately, or under the sponsorship of the National Health Insurance Program, which has made specific provisions for free examinations for the elderly). Only a small percentage (about ten percent) of the refusals were related to concerns about the procedures, the results, or the time commitment required for the protocol (Goldman et al., 1999). We also found that there was no negative effect of participation in the biomarker protocol on participation in the subsequent 1999 interview. Ninety-six percent of the (surviving) participants in the biomarker study responded to the 1999 interview, compared with a follow-up value of 92 percent for the entire population of survivors. That said, however, the 1998 collection of biomarkers was based on a subsample of the older respondents, that is, those who were age 69 and above. Our current fieldwork (due to begin in April 2000) will target people ages 53 and above. We recognize that recruitment of members of this younger cohort into a protocol that requires approximately two to three hours away from work may be more difficult than our experience with the older sample. The results from the elderly sample also lead one to suspect that another type of self-selection is operating: that the more highly motivated respondents agreed to participate in the bioindicator study. Whether this pattern is sustained when we examine a larger subsample of respondents or whether the motivation is related to health status are questions that we will examine carefully as the current fieldwork goes forward.

Beyond Taiwan, experience with the collection of biomarkers relative to participation and nonresponse has also been encouraging. The DHS, which has extensive experience with anthropometric and anemia measurement, reports that nonresponse (i.e., nonparticipation) has not been an important issue; to the contrary, they found that the biomeasures were an important motivation for participation in the studies (Vaessen, 2000). Even with a more extensive health examination in Pakistan, a similar experience was reported (Pappas, 2000).

Financial and Logistical Constraints

Financial and logistical constraints are considerable and are related to the choice of biomarkers and to location. Logistical problems for the potential collection of biomarkers in the HRS are exacerbated by the fact that it is a nationally representative survey of the United States with primary sampling units (PSUs) scattered across the country. For instance, in the proposed dementia assessment, two-person teams consisting of a nurse and a psychometric technician would conduct assessments in the homes of HRS respondents. Transportation of “flying squad” teams is a major component of the cost of such an endeavor.

For the Taiwan study, as for the HRS, PSUs are located geographically throughout the country but, of course, the total area is much smaller. The collection of the 12-hour urine specimen also adds substantially to both the expense and the complexity because it involves an additional visit to the respondent's home for the delivery of the collection supplies, and because the equipment and assay costs for cortisol and the catecholamines are high. Also, in Taiwan, phlebotomy must be performed by, or under the supervision of, a physician, so the cost of drawing blood is higher than it might be in locations where a technician could be employed. However, that cost is offset in Taiwan by the substantially lower charges for the blood and urine assays and the lower interviewing costs. In all, we estimate that the cost per respondent is about U.S. $400 for the interview, in-hospital physician's examination, and the blood and urine collection and tests. The logistics are not trivial. For the Taiwan study, the logistical constraints were significantly eased—and the response rate maintained at a high level—because of the experience of the Taiwan National Institute of Family Planning under the direction of Dr. Chang Ming-Cheng; the Director of Research and Planning, Mr. Chuang Yi-Li; and the project liaison, Ms. Lin Yu-Hsuan. The Institute has over 40 years of experience in fielding large-scale surveys and has well-established ties with public health nurses and other members of the medical community.

In general, it is difficult to imagine trying to accomplish these tests without adequate access to good communications and transport systems and to reliable refrigeration. In particular, challenges arising from inadequate cold chain equipment have been experienced by a number of studies (Vaessen, 2000; Makubalo, 2000).

The Taiwan protocol also involves extensive cooperation with the local public health nursing staff and with hospitals throughout the country. Drivers and interviewers who are knowledgeable about the local geographic areas are needed to find respondents; deliver and explain the urine collection equipment; and pick up, drop off, and return participants for the hospital visit.

Finally, other logistical considerations are significant as well. These include disposal of biohazardous waste materials, access to and training for a cadre of professional and technical assistants, quality control, and ensuring consistency of laboratory assays.

The Potential to Compromise Research Objectives

The collection of biomarkers also has the potential to exacerbate problems that affect all surveys. One concern common to virtually all research, for example, is the well-documented possibility that the research process itself affects behaviors that we wish to study. In the case of studies that include bioindicators as measures of health status, this concern is salient. Our earlier discussion of APOE testing suggested this possibility. More generally, this concern extends beyond genetic testing. Our clear obligation to provide participants with information related to their own health that is identified in the course of the study, indeed our obligation to encourage them to take action to address risk, may directly affect the course of the trajectories that we originally proposed to study. The data may become less representative of the population as a whole and reflect the effects of the intervention of the study itself.

While the possibility of such “contamination effects” should be considered seriously, it is also important not to exaggerate the problem. The information provided to a respondent by participating in a survey is typically only a small part of the information that is used in making important decisions. The contamination effects themselves may also be of independent scientific interest to researchers interested in the effects of interventions such as screening tests. Overall, it is important to ask whether there are mechanisms that meet acceptable ethical standards and allow for the collection of biological material without creating unacceptable risks of distorting the data either through the data collection process itself or by altering the behavior of respondents.

Providing a Benefit to the Participants—Diagnosis, Treatment, and Counsel

A primary consideration in the design of the Taiwan biomarker study has been ensuring that our participants benefit from their involvement in the study. We have taken a number of concrete steps to secure this result. The protocol simplifies access to a free health examination that is more comprehensive than the nationally funded exam for the elderly. The Institute arranges and confirms the appointment at the hospital, and arranges transportation for the participant to and from his home. Throughout the entire hospital visit the participant is attended by a member of the Institute staff; this staff member is there to answer questions and ensure the comfort of the participant. After the hospital routine, the participants are offered a light breakfast and a given a token gift. The results of the tests are reported promptly and with follow-up directions, information, and resources. In all, the Taiwan protocol cuts through hospital red tape, eases the hospital procedures, provides more extensive testing than the free national exam, and promotes follow-up.

Not all the decisions have been easy. For example, we have had persistent concerns about balancing obligations to report test results to the participants against the uncertainties of interpretation for some of the tests—the ones which are not routinely used in medical examinations or which are at the experimental stage. This concern has been particularly thorny in regard to the 12-hour urine protocol from which we obtain data that are important only for our research. We have also struggled to find the right balance between informing our participants about unusual test results and encouraging them to seek follow-up care while trying to avoid alarming them unnecessarily.

Informed Consent

Ensuring informed consent has been a serious concern throughout our work. Several factors related to the collection of biomarkers make informed consent an unusually complex task; among an aging sample, issues of cognitive function add to the complexity. Among the Taiwanese elderly, the inability of many of our respondents to read compounds the difficulties.

We have addressed these concerns in part by using a multistage process. Project goals and the protocol are described in a letter that is sent to each participant prior to the first visit by the public health nurse. During the nurse's visit, a second statement is read to each potential participant. Consent is obtained separately for the nurse's interview and for participation in the biomarker protocol. Following the visit by the public health nurse, those who have agreed to participate in the biomarker protocol are sent a letter confirming the date and time of their hospital visit and explaining the urine collection and hospital protocols. Another explanation is provided at the time the urine collection equipment is distributed to the participants. A final consent statement is given and read to the participant on the date of the hospital visit. At each stage, the respondents are reminded that their participation is voluntary and that they can choose to stop at any time.

If we look at data from our initial survey in 1989, they show that most of our elderly respondents—about 71 percent of the men and 83 percent of the women—reside with their adult children. Our experience to date has been that these children have played an active role in the informed consent decision process. In our 1997/1998 biomarker study, only seven of the participants had cognitive impairments that created concerns regarding comprehension of the study protocol. In these instances, the guardian (generally the child) provided permission for the physical examination. This small number, however, may in part be related to the residual refusal rate discussed earlier. Some of the targeted respondents who refused were too ill, possibly reflecting an effect of selection on our sample that may create analytical concerns. Although the longitudinal social surveys do use proxy reports for respondents who are unable to participate for various reasons, including cognitive disability, we know that study attrition is related to proxy status (Beckett et al., 2000).

Confidentiality

The need to protect the identity of survey participants and the confidentiality of the data that they provide is a paramount consideration for any study. Its importance relative to the collection of biological and health indicators cannot be overemphasized. The sensitivity of the data, the (currently) unknown implications of biological materials that may be established in the future, and the extensive linkage across both private and government-run data archives raise the potential consequences of violations to new levels that require strict oversight.