• E-mail
• Print
• Comment
• Font Size
• Digg
• del.icio.us
• Discuss article

Reading Public Health Research: Understanding Risk Factors and the Epidemiological Approach

Posted on: Friday, 30 September 2005, 06:00 CDT

By Albert, Steven M

Clear explanations for practitioners in aging.

The second fifty years of life are marked by an increasing prevalence of chronic disease (along with an increased risk of mortality from such disease), as well as an increase in disease burden, which is evident in persistent disability, increased use of medical and supportive care services, and the ever more difficult challenge of maintaining high function following hospitalization and rehabilitation.

As used here, the words risk and disease burden are technical terms drawn from epidemiology, "the study of the distribution and determinants of disease frequency" (MacMahon and Pugh, 1970). A basic understanding of epidemiology is critical for understanding the public health challenge of an aging world. Epidemiology provides tools for establishing the frequency of disease and potential differences in this frequency between populations or subgroups within a population. With these characteristics established, one can determine sources of variation in the risk of disease. Recognizing differences in the frequency and distribution of disease points to risk factors and hence the causal pathway, or etiology, of a disease. Once we have identified these pathways, we can break the causal chain. Epidemiology is a key element in the public health goal of developing interventions and preventive strategies likely to reduce the risk of disease and the burden associated with it.

In this brief review, we will examine key epidemiological terms used in aging and public health research, basic research designs for promoting public health among older adults, and selected studies that show the promise of a public health approach to aging.

KEY TERMS

In public health, the term prevalence refers to the number of people with a condition at a given time. Prevalence is expressed as a proportion, for example, the number of cases per 10,000 people ages 65-plus. Prevalence combines newly diagnosed and already existing cases. Thus, the prevalence of a given condition will depend both on the rate of new cases and on how long people live with the condition. Survival with a health condition, in turn, depends on available treatments, the rate at which the disease progresses, whether treatments are appropriately applied, other health conditions a person may face, environmental and social supports, and much more. Because of these features, prevalence estimates are most useful for assessing the most general impact of a health condition and for estimating the medical and supportive care needs of people likely to be affected by it.

Prevalence is commonly reported by public health agencies. For example, a recent compendium from the Centers for Disease Control and Prevention (CDC) (2003) reports that about 80 percent of people ages 65-plus have tine or more chronic conditions, and 50 percent have two or more. Nearly 20 percent have diabetes. Some 60 percent have arthritis. Some 6 percent to 10 percent have Alzheimer's disease, with AD prevalence increasing to nearly 50 percent in people ages 85-plus.

The term incidence, by contrast, refers to new cases in a population over a defined time interval. It is a rate -for example, the number of newly diagnosed cases per 10,000 people followed for one year, or per 10,000 person-years. Incidence is most useful for establishing ctiological factors. Because it is limited to new cases, differences in incidence between populations indicate differences in risk, and hence point to factors (i.e., differences between the populations) that lower or elevate risk. "Hie excess risk associated with an criological factor is usually expressed as a risk ratio (RR), which is simply incidence in the group with the eriologic factor divided by incidence in the yroup without.

Incidence studies have been very valuable for identifying risk factors, especially in the case of falling and disability. In the Yale FICSIT trial (Frailty and Injuries Cooperative Studies of Intervention Techniques), 35 percent of the intervention group fell over a one-year period, compared to 47 percent of those in the control group (which did not receive the intervention) (Tinetti et al., 1994). As part of the intervention, medication use for people in the intervention group was evaluated and adjusted, leading to a smaller proportion of the intervention group taking four or more medications (6; percent in the intervention group, compared to 86 percent in the control group). The results pointed to polypharmacy, that is, taking more than one medication, as a risk factor for falling. A second example is found in research with a sample population from the Established Populations for the Epidemiologic Study of the Elderly (EPESE). Here, poor performance on simple measures of lower-extremity ability (balance, leg strength, gait) among people not reporting disability at baseline was a strong risk factor for onset of disability over four years. For example, a third of the people with the poorest gait (defined as performance in the lowest quartile) went on to develop disability with activities of daily living (ADLS) (e.g., difficulty or need for help in bathing, dressing, grooming, or using the toilet). The risk of ADL disability was 15.5 percent, 8.3 percent, and 6.2 percent in quartiles representing progressively better performance (Guralnik et al.. 1995). The difference in incidence clearly established the importance of lower-extremity ability for future risk of disability and allowed investigators to identify a preclinical disability group that could be targeted for prevention efforts.

Interpreting risk can be difficult. Often we hear that a treatment or risk factor is associated with an apparently large difference, such as a 50 percent reduction in disease. However, the "50-percent reduction" is a measure of relative risk (a ratio), not absolute risk (a subtraction). The claim of a "so-percent reduction" can only be properly understood if we know the base rate of the event at hand. If 10 percent of a treatment group developed disease over follow-up, compared to 20 percent in a control group, we might very well be impressed by the so-percent reduction. But if 1 percent of the treatment group developed disease, compared to 2 percent of controls, we might be less impressed, even though this too is a 50- percent reduction in the treatment group compared to the control group. In the first case, the absolute difference in risk is 10 percent and in the second only 1 percent.

This difference in risk is also called "attributable risk." The absolute difference in risk is useful for establishing how much disease (or improvement) can be attributed to a particular exposure (or treatment). We can use this risk difference to calculate the proportion of disease that could be prevented by eliminating an exposure or risk factor. This proportion is called the "etiologic fraction." To return to our example from the EPESE cohort, the risk of disability over four years in people in the top quartile of lower- extremity performance was 6.2 percent. In people in the next quartile, risk of disability was 8.3 percent. The attributable risk or risk difference is 2.1 percent. If we divide the attributable risk by the risk in the group exposed to the risk factor (in this case, poorer lower-extremity performance, or 2.1/8.3), we obtain the etiologic fraction of 25 percent. Thus, 25 percent of the disability experienced by this group can be attributed to poorer performance in lowerextremity function.

An important consideration in the calculation of risk over time is change in the age distribution of a population. If a population includes an increasing proportion of older people over time (as we see globally as a result of declining birthrates on the one hand, and improvements in social conditions and medical care on the other), comparisons will have to standardize or "age adjust" to take this difference into account. Consider mortality in the United States between 1930 and 1998. Simply dividing the number of deaths by the number of people alive in any given year yields a change from about 1,250 deaths per 100,000 in 1930 to about 875 per 100,000 in 1998. But this difference, impressive as it is, still vastly underestimates the reduction in mortality. Once we recognize that the U.S. population has "aged" considerably over this period and adjust so that the denominators in the comparison have the same age structure, the difference is much larger, from 1,125/100,000 to 480/ 100,000 (CDC, 2000). This more accurate indicator takes into account the greater risk of death associated with older ages and the greater proportion of older people in more recent decades.

Disease burden is another important term used in public health. An advantage of the public health approach to aging is consideration of the consequences of disease, along with the epidemiological focus on etiological factors. The importance of this dual focus is well illustrated by research that examines both the prevalence of a condition and its impact. If we paid attention only to the prevalence of health conditions, arthritis and hearing impairment would be of prime concern for public health interventions. The prevalence of the two conditions in women ages 65-plus, for example, is 554.1 and 268.1 per 1,00\1-that is, half have arthritis, and a quarter have hearing impairment. The next most prevalent conditions are chronic obstructive pulmonary disease (125.6 per 1,000) and ischemic heart disease (120.7 per 1,000), with much lower prevalence. Diabetes and cancer are even less prevalent, with only 98.8 and 38.1 cases, respectively, per 1,000. However, if we consider the burden of each condition, that is, the proportion disabled as a result of the disease, we may want to rethink this ranking. For example, of women with hearing impairment, only 7.6 percent reported a need for personal assistance in daily self- maintenance tasks. In women with cancer, diabetes, and heart disease, the proportion reporting need for assistance rises to 33 percent to 38 percent. Women with arthritis, the most highly prevalent condition, were intermediate in reporting disease impact, with 26 percent reporting a need for personal assistance (Verbrugge and Patrick, 1995).

Once we consider the burden of disease we will likely consider adjusting priorities for public health efforts. For example, despite its lower prevalence, diabetes may be a higher priority for public health intervention than hearing impairment. Diabetes is more disabling, and prevention of a case of diabetes may confer more future disability-free years than prevention of hearing loss. This is an important consideration given the substantial proportion of life people can expect to live with disability after age 65. While mortality continues to decline at each age as it is postponed to later ages, Americans currently can expect to be disabled for about 15 percent of the total life span and nearly half the life span beyond age 65 (Erickson, Wilson, and Shannon, 1995).

RESEARCH DESIGNS

Central research designs in epidemiology include prevalence studies, case-control studies, observational cohort studies, and randomized intervention trials. These form a hierarchy, with evidence from prevalence studies considered weakest and evidence from clinical trials strongest. We illustrate each with examples.

Prevalence studies. Perhaps the most well known example of a prevalence study is the life table, which has recently been extended to calculate disability-free life expectancy. The standard life table uses mortality rates at a given time to calculate the likelihood of survival to any age. The life table assumes that age- based mortality rates in a given year are fixed and apply across the life span of a birth cohort (the life table also assumes that populations are stationary, with no immigration or emigration). Hence it is essentially cross-sectional, using current mortality by age to model the experience of a population as it ages. Life expectancy in this approach is simply the total number of personyears lived by a birth cohort divided by the number of people in the cohort. In 2002, life expectancy at birth among white men and women in the United States was -4.5 years and 79.7 years, respectively. Among African Americans, life expectancy \vas 68.8 years for men and 75-6 years for women, clear evidence of an important health disparity (CDC, 2004).

Table 1

Disability-Free Life Fxpectancy: Females Born in France, 1991

Table 1 shows the extension of the life-table metruxi to disability-free life expectancy. We begin with a birth cohort of 100,000 women bom in 1991. Column 3 "years lived in interval," shows the number of years lived by the cohort within each age interval, which reflects years lost to deaths at each age. We then apply the prevalence of disability at each age (defined here as need for help with ADLS, column 4, a figure available from health surveys) to calculate the total number of years without disability lived by the cohort within each age interval (column 5). As column 4 shows, the prevalence of disability increases dramatically with age; it is under 5 percent up to age 20, under 10 percent between ages 20-49, under 20 percent between ages 50-65, and well over 20 percent for ages 65-plus; indeed, a majority of women ages 85-plus are disabled. To obtain years without disability (column 5), we multiply the non- disability rate (column 4) by the number of years lived by the cohort (column 3). If we add the entries in column 5, we obtain the cohort's total disability-free years. This total is the first entry in column 6, "non-disability years, cumulative." If we then divide total disability-free years (column 6) by the number of people alive at the start of each age interval (column 2), we obtain disability- free life expectancy at each age. Among French women born in 1991, disability-free life expectancy at birth was 70.8 years.

Life-table estimates are useful for comparing different populations and population subgroups and also for tracking change over time within a population. Generalization of the approach to include quality-adjusted survival (such as disability-free survival) makes the tool extremely valuable.

Case-control studies. The case-control design is most useful for studying relatively infrequent events or rare diseases. In this approach, we begin by identifying people with a particular condition, perhaps from a clinic or disease registry. We then identify a group of people from the same source population (same clinic or community, for example) who could have developed the condition but did not. We use this control group to establish the base rate of a potential risk factor, such as an environmental exposure or prior experience of a clinical condition. If the proportion of cases exposed to some factor is significantly greater than what we see in controls, we say this factor puts people at risk for the disease. The comparison can be further refined by matching cases and controls on age, gender, education, locality, or related factors.

The case-control design presents the challenge of identifying the right control group. We want cases and controls to be similar in every regard except factors that cause disease, but we do not know these factors in advance and cannot be sure that all differences between the groups have etiological significance. Thus, the design is considered hypothesis-generating rather than definitive. The design has not been used widely in research in public health and aging because few conditions of interest are rare in late life. One variant of the design is an ecological comparison of case and control communities, as in a recent study of the deaths among older people during the 1995 heat wave in Chicago. In two communities matched for poverty level and age structure, one community suffered a case fatality rate of 40/100,000 and the other 4/100,000. The communities differed dramatically in social environment, as reflected in crime rates, use of public space, and contact with elders who did not leave their apartments, the prime group at risk for heat death (Klinenberg, 2002).

Observational cohort studies. The observational cohort overcomes the challenge of choosing the right control group by following a sample of people initially free of some condition of interest and recording who develops the condition. Thus, the design obviates the need to find controls to match to cases, since all have been followed from the inception of the cohort. Also, in this design the temporal order of risk factor and health outcome is clear: Factors assessed at baseline can be used to predict who develops the condition and at what point over follow-up. The key analytic tool in the cohort study is the survival model, where factors established at baseline (as well as changing conditions over time, in more sophisticated models) can be examined as predictors of the time to onset of some outcome of interest.

The observational cohort study is currently the workhorse of research in public health and aging. Most of the progress in measuring risk factors for mortality, dementia, cardiovascular disease, ADL disability, mobility limitation, falling, hip fractures, osteoporosis, Parkinson's disease, and other adverse outcomes has emerged from a series of key cohort studies, such as the Framingham Study; Cardiovascular Health Study; Health, Aging, and Body Composition Study; Alameda County Study; Washington- Heights Inwood Columbia Aging Study; Chicago Health and Aging Study; and many others.

Randomized intmvntion trials. Less common in public health and aging research is the randomized controlled trial, wherein people are randomly assigned to an experimental treatment, either to treat a condition (for people with a disease) or prevent it (for people at risk for a disease). The randomized, double-blind trial is the most rigorous research design, in public health as in clinical medicine, because (1) randomization makes it likely that risk factors (measured and unmeasured) are comparable across groups, (2) assignment of subjects to a condition by the researcher allows control ONXT the exposure of interest, and (3) blinding ensures that differences in outcome are truly the result of treatment status.

Recent efforts have made good use of the randomized intervention design. Results from these studies have shown many areas in which public health interventions can promote healthy aging. To name a few, nursing home placement can be delayed for people with Alzheimer's disease through supportive counseling (Mittelman et al., 1996); the risk of tailing can be reduced through monitoring of medications and home environment (Tinctti et al., 1994); independence and social engagement can be heightened through the use of a brief program of occupational therapy (Clark et al., 199"); and cognitive skills can be bolstered through cognitive remediation efforts (Ball et al., 2002).

CONCLUSION: THE PROMISE OF A PUBLIC HEALTH APPROACH TO AGING

This survey of the cpidemiologic approach to public health and aging, described more fully elsewhere (Albert, 2004), is well illustrated in a final study that knits together many of the themes presented here. A large observationa\l cohort study established the preventive benefit of influenza vaccination for older adults not just for influenza-based hospitalization, but also for all-cause mortality, hospitalization for cardiac and cardiovascular disease, and hospitalization for any cause (Nichol et al., 2003). The result was replicated over two time periods and three sites, used large samples, and showed a large difference in risk according to vaccination status-all study features and outcomes that indicate that vaccination truly modifies the risk of poor outcomes. Using a randomized intervention, the most rigorous approach, to measure effectiveness of treatments like vaccination would be unethical, because randomization would require withholding a valuable treatment from pan of the study population. So, in such cases, the observational cohort may be the best design possible. What about the unexpected benefit of vaccination for the risk of hospitalization for cardiac and cardiovascular disease? This outcome is plausible once we recognize that preventing influenza in people with chronic disease eliminates a health perturbation that, left unchecked, initiates a downward spiral of declining health and hospitalization. Preventing this downward spiral means less hospitalization and less risk of the poor outcomes associated with hospitalization, such as falls, nosocomial infections (those contracted in the hospital), delirium, reduced function upon discharge, and in-hospital mortality.

As this review has shown, an impressive array of epidemiological tools is available for research in aging and public health. The value of this public health approach is dear, with only a fraction of the most important findings described here. Progress in this area is accumulating and will undoubtedly accelerate as gerontologists adopt a public health approach and public health researchers focus on the second fifty years of life (Albert, Im, and Ravcis, 2002).

REFERENCES

Albert, S. M. 2004. Public Health and Aging: An Introduction to Maximizing Function and Well Being. New York: Springer Publishing Company.

Albert, S. M., Im, A., and Raveis, V. 2002. "Public Health and the Second 50 Years of Life." American Journal of Public Health 92: 1214-6.

Ball, K., et al., for the ACTIVE Study Group. 2002. "Effects of Cognitive Training Interventions with Older Adults: A Randomized Controlled Trial." Journal of the American Medical Association 288: 22-1-81.

Centers for Disease Control and Prevention (CDC). 2000. Deaths, United States, 2000. National Vital Satistics Report, Vol. 38, No. n. Hyattsville, Md.: National Center for Health Statistics.

CDC. 2003. "Trends in Aging-United States and Worldwidc." Mortality and Morbidity Weekly Review 53: 102-6.

CDC. 2004. United States Life Tables, 2002. U.S. National Center for Health Statistics, National Vital Statistics Report, Vol. 53, No. 6.

Clark, F., et al. 1997. "Occupational Therapy for Independent- Living Older Adults." Journal of the American Medical Association 278:1321-5.

Erickson, P., Wilson, R., and Shannon, 1.1995. Tears of Healthy Life. Healthy People 2000, Statistical Notes, No. 7. CDC.

Guralnik, J. M., et al. 1995. "Lower-Extremity Function in Persons over the Age of 70 Years as a Predictor of Subsequent Disability;" New England Journal of Medicine 332:556-61.

Klinenberg, E. 2002. Heat Wave: A Social Autopsy of Disaster in Chicago. Chicago: University of Chicago Press.

MacMahon, B., and Pugh, T .F. 1970. Epidemiology: Principles and Methods. Boston: Little, Brown.

Mittelman, M. S., et al. 1996. "A Family Intervention to Delay Nursing Home Placement of Patients with Alzheimer's Disease." Journal of the American Medical Association 276:1725-31.

Nichol, K. L., et al. 2003. "Influenza Vaccination and Reduction in Hospitalizations for Cardiac Disease and Stroke Among the Elderly." New England Journal of Medicine 348: 1322-32.

Tinetti, M. E., et al. 1994. "A Multifactorial Intervention to Reduce the Risk of Falling Among Elderly People Living in the Community." New England Journal of Medicine 331: 821-7.

Verbrugge, L. M., and Patrick, D. L. 1995. "Seven Chronic Conditions: Their Impact on U.S. Adults' Activity Levels and Use of Medical Services? American Journal of Public Health 85:173-82.

Steren M. Albert, Ph.D., M.Sc., is associate professor. Mailman School of Public Health, Columbia University, New York, N.Y.

Copyright American Society on Aging Summer 2005

Source: Generations

More News in this Category