September 24, 2008
Population-Level Risk Factors, Population Health, and Health Policy
By Naumova, Elena N Cohen, Steven A
Putnam and Galea challenge the epidemiological framework underlying conventional public health research. They underscore the need to address population-level macrosocial determinants of health in epidemiological studies. Studying the population-level factors that ultimately determine population health is surely essential to policy development and implementation, as public health policy is disseminated at the population level. What impedes incorporation of macrosocial determinants of health into the epidemiological domain, leaving today's playing field to individual-level socioeconomic conditions, or micro-social determinants - often outside the health policy domain? The idea that in addition to individual-level risk factors, macrosocial factors operating at the population level affect health is not new. In 1973, Mervyn Susser pointed out that "a determinant can be any factor...that brings about change for better or worse in a health condition" (i) and does not necessarily occur on the individual level. Because of resistance to incorporating population-level factors in epidemiological studies, the potential for social epidemiology to be a meaningful paradigm for researchers, policymakers, and practitioners of public health has yet to achieve its full potential.
In addition to the barriers delineated by Putnam and Galea, we note numerous other practical explanations as to why macrosocial determinants of population health have yet to be fully explored in the epidemiological literature. Availability of data needed to undertake population-based studies stands out. Sometimes reliable population data on diseases and contextual factors of interest are not available. For outcomes or diseases, national and local surveillance data quality and coverage vary by disease and location (2). Accurate and timely population-based data may be limited for contextual sociodemographic factors, such as income.
In the United States, some of the data most widely used in research can be obtained from the US Census Bureau, but the national census is conducted only every 10 years. For the intercensal years, interim data exist for socioeconomic and demographic characteristics, but the data coverage on some measures is sporadic. With the development and implementation of the American Community Survey by the US Census Bureau, more reliable data for small areas and on a continuous timeframe should be more readily available in the near future (3). Both the demography and epidemiology communities would benefit from collection and dissemination of population data.
As Putnam and Galea explain, compared to more traditional individual-level risk factor-based epidemiology, few reliable, flexible, and applicable summary measures of population health and specific disease burden exist. Susser and Susser elaborated upon this view while suggesting that the lack of standardized and validated population-based methodologies may account for their underuse (4). But several statistical tools suitable for population- level analysis do exist.
Multi-level or hierarchical linear modeling, for example, is a class of techniques involving sophisticated modeling of population- and individual-level factors that has contributed greatly to the advancement of public health knowledge (5). Mixed effects models and generalized estimating equation (GEE) procedures constitute two powerful methods of multi-level modeling that can accommodate the longitudinal nature and correlated data structure inherent in population-level analyses. Despite their utility for capturing population-level influences on health outcomes, several inherent obstacles to their universal implementation persist: proper training in advanced statistics; model specification to suit the hypothesis being tested; and difficulty interpreting model parameters.
Conceptualization of these models is substantially more complex than traditional linear models, thus full understanding of the underlying concepts behind these methods requires advanced training, which is generally lacking in many public health masters and doctoral programs. Interpretation of these models is not always transparent and visualization of results is often less than straightforward. seemingly minor mistakes in specification of multi- level models can lead to erroneous conclusions about the role of macrosocial factors as they relate to population health. Multi- level modeling procedures are much more computationally demanding than standard linear regression models and require specialized software and are often prohibitively expensive and difficult to use. Complex modeling procedures such as GEE and mixed effects models are not included in all statistical software packages. Where statistical programs do contain these modeling procedures, running similar analyses using different statistical packages may yield different results.
More emphasis should be placed on developing sound methods and measures to study larger-scale, population-level determinants of health that can, either independently or together with other factors, modify health. With the goal of making these methods more easily understandable for use and interpretation by the public health community, biostatisticians, epidemiologists, and other public health professionals should work together to create guidelines for using these methods.
More practical challenges confront those implementing a population-level approach to health research. Two overarching conceptual obstacles must be overcome to realize fully the potential and need to incorporate these factors into the standard epidemiological research paradigm. The traditional belief that the best way to assess causality is with a randomized, controlled trial is first. Biostatisticians have devoted enormous resources to and developed comprehensive methods to analyze data from randomized controlled trials. In comparison, while statistical techniques to analyze population-level socioeconomic data with population-level health outcomes exist, they have not, as Putnam and Galea explain, been explored as deeply.
One of the most appropriate designs for a study of macrosocial determinants of population health is not the randomized, controlled trial, but rather the ecological study. The literature surrounding study design itself has long regarded the ecological study design to be one of the weakest for determining causation of individual risk factors to individual outcomes (6). Studies of population-level macrosocial determinants of population health do not purport to draw individual-level conclusions about risk factors and health outcomes. To assess how population-level factors influence health at the aggregate level, ecological study designs are the logical choice. Just as the ecological study is useful for evaluating population- level associations, analysis of spatial clustering and time-series modeling can be used to assess spatio-temporal associations between population-level risk factors and outcomes. Time-series analyses, although often regarded as weak study designs to assess causality, can indeed demonstrate one of the key criteria for causality mentioned by Bradford Hill: the exposure must precede the outcome (7). To assess temporal aspects of population-level macrosocial determinants of health almost demands active use of time-series analysis. Characteristics inherent in or affecting the entire population, such as herd immunity, income inequality, and their spatio-temporal variability, would have little meaning on the individual level. They can be best evaluated through the use of ecological, spatial, or time-series analyses (8).
The Gini Index, a measure of population-level inequality, is an example of a measure that can be defined only on the population level. This population-level characteristic has been a subject of increasing exploration and analysis in the medical and public health literature. The relationship between income inequality and health has been explored on several geographical levels and in various settings. Lynch et al. found that after controlling for the absolute income level, wealthy countries with higher income inequality tended to have higher infant and child mortality rates than similar countries with lower income inequality (9). Gravelle and Sutton found similar spatiotemporal associations in the United Kingdom between income inequality and infant mortality, as well as cause- specific mortality later in life (10). Herd immunity, defined as the indirect protection of populations from infection resulting from the presence of immune individuals (n), is another example of an epidemiological concept that cannot be defined on the individual level, and is reserved for populations. An influential analysis that assessed the long-term temporal trends in vaccination of school children and excess mortality comparing Japan and the United States ultimately led to the new CDC recommendations for universal influenza vaccination for children to achieve indirect protection of other subpopulations (12). For decades, demography and economics have embraced the major macro-level social determinants with respect to societal development, large-scale demographic and epidemiological transitions, and mortality. These fields have focused traditionally on the complex social interactions, environmental changes, and economic and technological advances that have either led to or hindered the development of societies. Their approaches can be adapted for epidemiology and public health research, including the development of new perspectives and blending of interdisciplinary perspectives that transition from the individual to the population level. Challenging customary epidemiological paradigms by incorporating demographic techniques and perspectives will surely inform public health research and policy, providing a more complete picture of the factors that contribute to population health. One such contribution would be to account for population-level spatio- temporal changes in disease patterns and transmission. Similarly, the concept of susceptibility to disease demands a life-span approach, not just a life-stage approach, incorporating the idea that social and economic development on the local, national, and even global levels can ultimately change disease patterns for populations (13).
A cohort approach is another example of extending demographic techniques to critical population health problems. An age- periodcohort approach was used to assess trends in breast cancer mortality rates over time in Spain (14), as well as for lung cancer mortality patterns across nations (15). The cohort perspective centers on the potential for health outcomes later in life to be affected by the common experience to a set of large-scale socioeconomic and environmental conditions at a common age contemporaneously throughout the life span. To further the understanding of populationlevel determinants of health, the public health research community should adopt and incorporate the principles and methodologies commonly used in demographic research.
The aforementioned transitions toward a more population-based, life span approach to health have the potential to affect public health policy downstream in major ways. Policy, whether health- related or not, is established at the national, state, and municipal levels rather than at the individual level. Policies that ultimately affect population health - vaccination recommendations or requirements, redistribution of wealth, environmental policies, behavioral modifications, and countless others - have far-reaching consequences and can affect health in ways previously unknown or unintended.
In an ideal world, rigorous scientific research informs policies adopted to maximize population health. However, there are several important considerations that often impede this process, ranging from logistics to economics, and the politics mentioned by Putnam and Galea. Consider the recent policy recommendation for universal influenza vaccination of children (i2.,i6). Clinical trials can readily determine the effectiveness and safety of such a policy for individual child. But vaccination of schoolchildren may have indirect benefits in the adult and elderly populations through herd immunity. To study this question demands a population-based approach that cannot be achieved through clinical trials. Population-based, macro-level analysis affords the researcher, and ultimately policymakers, the ability to assess in the population the indirect benefits of such a policy. Research can address macrosocial effects and their interactions with influenza rates in the population rates, measuring vaccine shortfalls, vaccine coverage trends, the associations with socioeconomic characteristics, and numerous other properties of the population.
Individual risk factor-based epidemiology is critical to understanding disease dynamics and prevention, remaining relevant to innumerable conditions affecting the population. Nonetheless, these individual risk factors are not distributed randomly in the population. Ultimately, population-level determinants of health are those driving forces that both serve to distribute those individual risk factors, and to mitigate population health through collective experience. Incorporation of population-level determinants involves both subtle and sweeping changes to the current public health research framework. Surely, these changes will not occur overnight.
We hope that, in time, public health researchers will more fully explore these important contextual variables to improve population health domestically and abroad; these researchers will better understand the importance of these factors in all aspects of risk factor epidemiology. The public ultimately contends with any health policy aimed at improving public health. To us, it surely makes sense to address potentially modifiable factors at the population level, factors that can be addressed to protect the health of the population.
1. Susser M. Causal Thinking in the Health Sciences. New York, NY: Oxford University Press; 1973.
2. Weber SG, Pitrak D. Accuracy of a local surveillance system for early detection of emerging infectious disease. JAMA. 2.00352.90: 596-598.
3. US Bureau of the Census. American Community Survey (ACS)- About the Data (Methodology): Multiyear Estimates Study. Website: http:// www.census.gov/acs/www/ AdvMeth/Multi_Year_Estimates/ overview. html, accessed 2.9 January 2.008.
4. Susser M, Susser E. Choosing a future for epidemiology: II. From black boxes to Chinese boxes. Am J Public Health. 1996;86:674- 8.
5. Fitzmaurice GM, Laird NM, Ware JH. Applied Longitudinal Analysis. 1st edition. Hoboken, NJ: Wiley; 2.004.
6. Tower RL, Spector LA. The epidemiology of childhood leukemia with a focus on birthweight and diet. CnY Rev CHn Lab Studies. 2007544:203-42.
7. Last JM, editor. A Dictionary of Epidemiology. 3rd edition. New York: Oxford University Press; 1995.
8. Zeger SL, Irizarry R, Peng RD. On time series analysis of public health and biomedical data. Ann Rev Pub Health. 2.00652.7:57- 79.
9. Lynch J, Smith GD, Hillemeier M, Shaw M, Raghunathan T, Kaplan G. Income inequality, the psychosocial environment, and health: comparisons of wealthy nations. Lancet. 20015358:194-200.
10. Gravelle H, Sutton M. Income related inequalities in self assessed health in Britain: 1979-1995. J Epidemiol Comm Health. 2003557:125-9.
11. Fine PEM. Herd immunity: history, theory, practice. Epidemiol Rev. 1993515:265-302.
12. Reichert TA, Sugaya N, Fedson DS, Glezan P, Simonsen L, Tashiro M. The Japanese experience with vaccination schoolchildren against influenza. N Engl J Med. 20015344:889-96.
13. McMichael AJ. Prisoners of the proximate: loosening the constraints on epidemiology in an age of change. Am J Epidemiol. 19995149: 887-97.
14. Cayuela A, Rodriguez-Dominguez S, Ruiz-Borrego M, GiIi M. Ageperiod-cohort analysis of breast cancer mortality rates in Andalucia (Spain). Ann Oncology. 2004515:686-8.
15. Liaw Y, Huang Y, lien G. Patterns of lung cancer mortality in 23 countries: application of the age-period-cohort model. BMC Public Health. 2.00555:2.2-9.
16. Centers for Disease Control and Prevention (CDC). ACIP Provisional Recommendations for the Prevention and Control of Influenza. 2.5 March, 2.008. Website: http://www.cdc.gov/vaccines/ recs/provisional/ downloads/flu-3-zi-o8-5o8.pdf.
ELENA N. NAUMOVA and STEVEN A. COHEN
Tufts University School of Medicine, Boston, MA, USA
Correspondence: Elena N. Naumova, Tufts University School of Medicine, 136 Harrison
Avenue, Boston, MA OZIII, USA. E-mail: [email protected]
Journal of Public Health Policy (2.008) 29, 290-298.
ABOUT THE AUTHORS
Dr. Elena N. Naumova is a professor at the Tufts University School of Medicine and Director of InForMID, the Tufts University Initiative for the Forecasting and Modeling of Infectious Diseases. Her expertise is the development of analytical tools for time series and longitudinal data analysis applied to disease surveillance, exposure assessment, and studies of growth. Her research activities span a broad range of research programs in infectious disease, environmental epidemiology, molecular biology and immunogenetics, nutrition and growth.
Dr. Steven A. Cohen is an assistant professor at the Tufts University School of Medicine and a member of InForMID. His primary training is in population studies pertaining to public health policy and programs. He is interested in the modeling and prediction of infectious disease dynamics in the United States and assessing health care utilization practices in the US elderly population.
Copyright Palgrave Macmillan Limited 2008
(c) 2008 Journal of Public Health Policy. Provided by ProQuest LLC. All rights Reserved.