| Operational Challenges in Longitudinal Studies |
| Although the longitudinal design is essential to the determination of age changes in individuals, it cannot resolve all the difficulties inherent in cross-sectional studies. Furthermore, longitudinal studies have a number of limitations of their own. What seems to be a simple straightforward question - "How does aging, or the passage of time, affect performance in individual subjects?" - turns out to be a demon in disguise. Many pitfalls in design, subject selection, data collection, and data analysis may undermine or negate the assumption that changes in serially collected measurements are due to aging. These problems include the following: |
| 1. Recruitment and Screening of Subjects |
| A primary concern in the selection of subjects for a longitudinal study is their commitment to continued participation in the study and their geographic stability over long periods. This requirement limits the sampling procedures that can be used and must be taken into consideration in the generalization of conclusions drawn from any longitudinal study. |
| The study may require the inclusion of procedures that prove tedious or distasteful to many people. Subjects selected at random will show both a high initial rate of refusal to participate and a high drop-out rate when presented with a test schedule that includes uncomfortable and time-consuming procedures. To this extent most longitudinal studies, including the BLSA, have compromised true representativeness in order to obtain loyalty and cooperation from their subjects. |
| Some studies from their inception exclude subjects who present clinical or laboratory evidence of disease. Although this procedure may initially limit the study to healthy subjects, it does not avoid the problem of a subject who subsequently develops chronic illness. If the subject is then dropped from the study, a great opportunity to trace the historical development of a disease is lost. Hence subjects who developed diseases were not dropped from the BLSA, although observations made on them were no longer included in analyses of age changes. |
| Another approach is to accept subjects with diagnosed disease, but to exclude measurements made on them from data analyses designed to characterize normal age changes. Serial observations on these subjects as a subgroup can be of great value in distinguishing the effects of aging from those of aging plus disease. |
| Although the presence of disease will confound the interpretations about aging in both cross-sectional and longitudinal studies, diseases are more apt to be discovered in subjects in a longitudinal than in a cross-sectional study because of the extended time during which longitudinal subjects are seen and tested. Findings that may be equivocal at one testing can be re-examined on subsequent visits for verification of diagnoses. |
| 2. Attrition |
| Subject losses must be expected as a longitudinal study progresses. Younger subjects are more likely than old ones to move away from the area of the study or to lose interest and motivation. As the subjects become older, death and disability become major factors (Wilson and Webber, 1976). On the other hand, useful research data may emerge from comparison of measurements in subjects who have survived with those in subjects who have died. This may result in development of new methods of predicting the likelihood of death. |
| Drop-outs due to loss of contact or to subjects' refusal to continue participation pose a problem in the interpretation of results, particularly when it is evident that those who have left the study differed systematically from those who have remained. |
| The degree, to which findings from longitudinal studies are distorted by attrition whatever its source, depends on the aspect of aging that is being investigated (see Chapter III). Since some variables are influenced more than others by attrition, each variable in each study must be examined for the drop-out effect. |
| 3. Expansion of Subject Panel |
| Unlike most longitudinal studies, the BLSA is designed to maintain a specified number of subjects within each age decade throughout its course. When new subjects are introduced, it is important that they resemble the original sample as closely as possible. Ideally, this requires careful description and matching of the original and new populations. Although the BLSA did not attempt such a matching, the self-selection strategy employed in the recruitment of its participants has tended to maintain the character of the sample (see Chapter III). |
| 4. Strategies of Analytical Design |
| It is often assumed that the differences among serial observations within cohorts followed longitudinally represent the effects of aging. This is not necessarily true: A number of non-maturational effects or factors may also induce differences in serial measurements. Changes in measurements made serially over time may be due to: a) changes in procedures; b) systematic methodological error; c) period effects - environmental or cultural changes that may influence all members of the population under study; or d) aging effects. |
| An aging effect is present if the dependent variable is a function of age regardless of the subject's birth year or of the period or time of observation. A period effect is present if the value of the variable changes systematically as a function of the time of observation and not as a function of age. A birth-cohort effect is present if the value of the variable changes systematically as a function of the subject's birth year rather than of his age. |
| Cross-Sequential Design/Time Sequential Design -- Table I.1 and Table I.2 |
| One of the primary difficulties in the analysis of longitudinal data from a single birth cohort is the confounding of period effects with age changes. Traditional longitudinal designs attempt to circumvent the cross-sectional confounding of aging effects with generational or birth-cohort effects by following the same group of individuals over two or more times of measurement, and thus at two or more ages. Such designs, however, are subject to the confounding of age with period effects. Changes that occur between the first and second measurements may be due to intervening historical events rather than to aging; in some tests, previous exposure or practice may be responsible. |
| Longitudinal changes include age and period effects. Cross-sectional differences include age and cohort effects. Each set of differences is thus influenced by two of the primary effects, those of age, period, and birth cohort. Cross-sequential and time-sequential designs have been proposed to help untangle the confound. In the cross-sequential design, independent samples of individuals from the same birth cohort are compared at different times of measurement, and thus at different ages (Table 1). Since a given individual is measured only once, exposure or practice effects are eliminated. In Table 1 the vertical comparison confounds aging and the effects of birth cohort, while the horizontal comparison confounds aging and period effects. |
| In the time-sequential design, independent samples of individuals of a specified age are compared at different times of measurement (Table 2). Age and time of measurement are separated, but both are confounded with birth cohort. No clear-cut statistical separation of age effects from birth-cohort and period effects can be made. |
| It was originally assumed that, while each of these designs is ambiguous when used alone, it might be possible to separate out age, period, and birth-cohort effects if all were employed and analyzed simultaneously (Schaie, 1965; Baltes, 1968; Riley et al., 1972; Agnello, 1973; and Mason and Mason, 1973). It has since been demonstrated, however, that there is no single solution to the inevitable confounding of the three, and that interpretation of such analysis depends on the data, the goals of the investigator, and the state of knowledge in the area. Costa and McCrae (1982) discuss in greater detail the role of judgment in the interpretation of aging, period, and birth-cohort effects. |
| 5. Maintaining Uniformity of Methods and Quality Control |
| A longitudinal analysis requires special attention to the maintenance of uniformity of tests and testing conditions throughout the study. Continuous quality control is essential. Methods must be examined at regular intervals for consistency of results and stability of standards. |
| In the BLSA, replicate samples of blood, urine, and tissues are frozen and stored for re-analysis at a later date. Stored plasma samples have made it possible, for example, to validate the methodology used for the determination of cholesterol levels by repeating the analyses at one time on a random subset of samples collected over the entire span of the study (Hershcopf et al, 1982). |
| 6. Data Storage and Retrieval |
| Longitudinal studies generate special problems in the storage and retrieval of data. Thanks to the developments of computers, it is now possible to store an immense amount of data in such a fashion that the data can be updated as successive test cycles are completed and at the same time remain available for analysis. It is essential that the system and format of data collection be carefully planned in advance, with the advice of personnel trained in computer technology (Ramm and Gianturco, 1974). It is also essential that special precautions be taken to protect the stored data against catastrophic loss as well as to ensure confidentiality. |
| 7. Staffing |
| Longitudinal studies pose special problems in the recruitment and maintenance of a research staff. As Busse (1965) has pointed out, scientists who participate successfully in longitudinal studies possess distinctive personal characteristics in addition to their scientific qualifications. They must first of all be patient and willing to wait for longitudinal results to evolve. This does not imply that they will sit with folded hands during the early stages of the study; they will have the insight and initiative to examine data cross-sectionally and to look for significant relations among observations as they accumulate. They will generate new hypotheses that can be explored by the introduction of new tests and procedures. As a result, an effective longitudinal study will be dynamic and will not be limited by the initial test procedures. |
| Since longitudinal studies are apt to be multidisciplinary in their design, the successful investigator should be able to work with others as a member of a team. Each participating scientist should also be interested in other scientific disciplines and willing to communicate with other scientists in the solution of problems. |
| 8. Financing |
| Longitudinal studies in adults require stable funding for long periods of time if their full potential is to be realized. Although data analysis must be a continuing part of the program, significant longitudinal results cannot be expected in the early years of a study. Hence, a support system that requires the reporting of substantive longitudinal results at short intervals in order to maintain funding is inappropriate. Since short-term funding has in the past characterized most research support, few individuals or institutions have been prepared to initiate and carry out longitudinal studies. |
| Of primary concern to research administrators are the presumed high costs and the long-term commitment of resources. However, the ability of a longitudinal study to answer certain important questions about aging answerable by no other technique fully justifies the costs. Although costs may appear high in comparison with those of cross-sectional studies, the potential efficiency of having a population with known characteristics available for multiple satellite short-term cross-sectional studies covering the entire period of adult life greatly increases the cost effectiveness of a longitudinal study. The costs of recruiting multiple groups of well-characterized subjects for short-term studies may well exceed those of maintaining a single stable population. |
