![]() |
![]() ![]() |
||||||||||||||||
![]() ![]() ![]() |
|
||||||||||||||||
|
Perspectives on Epidemiologic Surveillance in the 21st Century
Abstract This paper describes the importance of epidemiologic surveillance
as a systematic, ongoing and population-based system for early warning
and program development in the 21st century. Such a system routinely
collects data on three classes of indicators (health outcomes, risk
factors and intervention strategies) to set up both an early warning
system (to identify associations and make predictions on health outcomes)
and a program development system (to assess the need for intervention
strategies, to plan and implement such strategies and to assess their
effectiveness). A comprehensive surveillance system must be systematic
(evidence-based selection of indicators, not hypothesis-driven), ongoing
(continuous data collection, including repeated surveys) and population-based
(whole population, or representative samples of the population). Such
a system need not be developed from scratch, but can be based on linkage
of existing databases and collection of additional information for identified
data gaps. The initial steps for selecting indicators and creating a
prototype framework for a comprehensive surveillance system are proposed
to stimulate further discussion. It is suggested that surveillance systems
should be used more widely in public health.
Introduction Epidemiologic surveillance dates back to the time of John Graunt, who published the Natural and Political Observations Made Upon the Bills of Mortality in 1662.1 Graunt's approach for the analysis of death certificates (Bills of Mortality), that volumes of data should be reduced to a few tables and that profit may be gained by analyzing these tables, is consistent with the modern technique of population-based epidemiologic surveillance.2 In the subsequent 300 years, however, the focus of health research shifted to sample-based studies: cross-sectional, cohort and case-control studies, and clinical trials.3-6 In recent decades, awareness of the limitations of sample-based epidemiologic studies has grown,7 along with recognition of the importance of population-based surveillance systems for measuring the health status of a population,8 for early warning of emerging health risks and for program development.9 Systematic and timely analysis of health trends has been identified as increasingly important for evidence-based policy and program development.10,11 At the same time, biophysical and socio-economic data have become invaluable in the understanding of relationships among human health, risk factors and interventions.12 It appears that epidemiologic surveillance may come back full circle in the 21st century and become once again the focus of health research. (It has been suggested that epidemiologic surveillance is not research per se, and can therefore only be called public health surveillance.13 But it is my opinion that health research can be conducted in the next century using well-maintained and well-validated surveillance databases.) This paper offers a point of view for debate on the blueprints for a systematic, ongoing and population-based surveillance system for the 21st century, with the hope of stimulating further discussion on this very important topic.
In a sample-based approach, investigators conduct custom-designed and mostly localized studies to investigate associations. However, these studies are prone to biases, including those that Sackett14 called "positive results bias," "hot topic bias," "wrong sample size bias," "expectation bias" and "data dredging bias," and a number of biases reviewed elsewhere by this author.15,16 (Population-based studies are also susceptible to these biases, but to a lesser extent than sample-based studies.) Consider a situation where an exposure and a disease are not associated. Due to problems in study design, data collection or analysis, or by chance, a sample-based study may incorrectly determine that there is an association. When published, this false positive study can create a hot topic bias, that is, more investigators will become interested in the topic. In this case, let's say 100 studies may be designed. If the type I error rate (significance level) of these studies is chosen at the conventional 0.05, then on average, 5 of the 100 studies will show false positive results. Since positive results are more likely to be submitted by investigators to scientific journals (positive results bias) and accepted by editors (editor's bias), this will lead to an even bigger hot topic bias, and another false positive research cycle will begin. (In the case of a hot topic, it is unlikely that all of the 95 correctly negative studies will go unpublished, and letters to the editor will almost certainly follow. Although the hot topic bias could thus be somewhat self-corrected, the overall tendency for the bias remains.) Through this biased process, an investigator can almost always "prove" something out of nothing. Results of such an unsystematic, non-population-based approach would likely focus efforts in certain narrow areas, ignoring some other major and real issues. While sample-based studies have provided a tremendous amount of knowledge, their limitations due to false positive research cycles must be recognized.
Early work on population-based health surveillance included cluster studies and ecologic correlation studies. In the former type of study, observation of unusual clusters of cases of a rare disease in population subgroups exposed to a common risk factor may lead to hazard identification. Ecologic correlation studies, on the other hand, seek to establish an association between exposure and disease occurrence by comparing disease rates among populations in different geographic areas subject to varying levels of exposure.17 An example of an ecologic correlation study that is very close to a surveillance system is the correlation study by Murata et al.18 These researchers looked at how mortality and incidence rates of cancers of the lung, colon and rectum correlated with 63 environmental variables (including population density, number of households living on welfare aid, number of hospitals, tobacco tax revenue, sulphur dioxide level, rainfall, traffic density and meat consumption) in 583 geographic districts of Japan from 1975 to 1979. However, like other similar studies, this study failed as a surveillance system because it was not performed on a long-term basis, results were not used in the systematic development of intervention strategies and there was no systematic information dissemination. Furthermore, cluster studies and ecologic correlation studies to date are susceptible to ecologic fallacies and other potential problems.16 The US Centers for Disease Control and Prevention (CDC) defined epidemiologic surveillance as the "ongoing systematic collection, analysis, and interpretation of health data essential to the planning, implementation, and evaluation of public health practice, closely integrated with the timely dissemination of these data to those who need to know."19 More recently, Monson pointed out that, as scientists of the 21st century, "rather than conducting short-term studies that will test a hypothesis, we must be developing systems to collect data on exposure and disease that will become part of the fabric of the community, the workplace, and the health care organization."20 The importance of a comprehensive, systematic, long-term and population-based surveillance system has been stressed. Recent textbooks by Halperin and Baker2 and Teutsch and Churchill13 have summarized the basic principles of public health surveillance. However, the form, content and operating principles of a comprehensive epidemiologic surveillance system for the 21st century have not yet been described in the literature.
In a comprehensive and systematic approach to surveillance, all associations with the same strength will have an equal chance of being detected. In the 21st century, scientists and public health officials should therefore concentrate more on systematic, ongoing and population-based data collection systems than on hypothesis-driven, short-term sample-based data collection. Important and useful data on health indicators, risk indicators and intervention indicators should be collected routinely, systematically and accurately for the community. (Many of the current databases may be routine and population-based, but neither systematic nor accurate.) These databases can then be analyzed systematically to monitor trends of health, risk and intervention in the community, to identify emerging health risks (early warning system) and to develop and evaluate evidence-based disease control and prevention programs (program development system). To develop a comprehensive early warning and program development system based on surveillance, the following three questions need to be answered.
A conceptual framework for health information, the Health Template, was put forward in 1991 by the National Task Force on Health Information21 (Figure 1). The Health Template classifies health information into three major areas-individual characteristics, external milieu and "health-affecting" interventions-with further subdivision into several levels of categories. The Health Template attempts to provide a framework for the subsets of quantitative information and a basis for possible development of forecasting and policy modelling. Therefore, it can be used potentially as a model for the selection of indicators for a comprehensive surveillance system. The problem with the Health Template, however, is that it is too ambitious. In the current version there are 44 categories for individual characteristics, 93 for external milieu and 59 for health-affecting interventions, a total of 196. Most of these categories are described in general terms, such as "air" or "social support," without working definitions. When fully developed, the whole classification scheme of the Health Template may have too many variables for a practical surveillance system. The list of indicator variables must be narrowed down to a manageable size for use in a comprehensive surveillance system. Several techniques may be used to select indicator variables, including literature research, Delphi surveys among experts22-24 and experts' consensus workshops.22 These techniques will involve development of a priori selection criteria.
|
||||||||||
|
|||||||||||
2. Where can these data be found? There are already many existing data sources in Canada. For example, there are national population-based databases, such as those on cancer incidence and mortality, congenital anomalities, hospital statistics and the National Population Health Surveys. Provincial population-based databases include British Columbia Health Surveillance, Manitoba Infant Deafness Surveillance, Newfoundland Disability Surveillance and Prince Edward Island Diabetes Surveillance. There are also national voluntary databases, such as those on cystic fibrosis, hemophilia, multiple sclerosis and muscular dystrophy. In the US also, there are numerous existing databases. In addition to the CDC-funded Emerging Infections Program operating in seven states, most of the other states already have population-based surveillance systems. For example, the Behavioral Risk Factor Surveillance System, which started in 1984, now covers all 54 American states and territories. Twelve years of data (1984-1995) are now available on CD-ROM. Record linkage of existing databases can be used to create new and more useful databases. In addition, new data collection systems will be needed for data not yet captured routinely. Currently, there are numerous potential surveillance systems that are not being used for surveillance in Canada and the US. Two of the oldest, birth certification and death certification, are not used optimally for surveillance, partly because of the poor data quality and partly because of the lack of linkage to other useful data. If, say, death records and hospital statistics (health outcomes) were linked with both behavioural risk factor surveillance data (risk factors) and programs/services data (intervention strategies), the result would be a comprehensive surveillance system that could put current public health resources to better use. It is important to know that many of the existing databases have been created for administrative purposes (i.e. not for use by epidemiologists), have not been validated and may contain a lot of surplus information of no use to epidemiologic surveillance. Therefore, these data sources should be evaluated, validated and used for providing data only for the selected indicator variables. Data pertaining to the same indicator variables may be cross-validated from multiple data sources. The useful portions of the separate databases may be physically combined to become a central, large database for the comprehensive surveillance system. Alternatively, the various databases may remain separate, with easy access for use through telecommunications.
The exact form and content of a long-term, comprehensive epidemiologic surveillance system for the 21st century is not known at this time. However, it is anticipated that the system will collect data on three classes of indicators: health, risk and intervention. Potential variables for each class are proposed in Table 1. |
|||||||||||
|
|||||||||||
The lists of indicators are proposed here for the purpose of inviting further discussion. For example, behavioural changes appear in one list but tobacco use and physical exercise are in another. Healthy eating is not included in any list. The actual selection of variables should be systematic and evidence-based as described in the next section, "Initial Steps." Risk factors can directly result in changes in health trends, and interventions can directly or indirectly (through their effects on risk exposure) result in health changes. Therefore, monitoring changes in risk indicators and intervention indicators can predict health changes and identify emerging health problems (early warning). If intervention programs are effective, they are expected to cause changes in both risk exposures and health outcomes (program development and evaluation) [Figure 2].
|
|||||||||||
|
|||||||||||
Initial Steps Initial steps for setting up a comprehensive, long-term surveillance system may include the following activities. 1. Conducting a series of round table discussion sessions to identify the stakeholders, purposes and priorities for the comprehensive surveillance system 2. Conducting an extensive literature review and literature survey to identify valid, reliable indicators of the biophysical and socio-economic environments and health outcomes; using meta-analysis25,26 to prioritize risk variables based on relative risks and attributable risks 3. Conducting a series of Delphi surveys22 (an initial survey to acquire indicators and subsequent surveys to rank indicators) among cross-disciplinary teams of experts to identify the set of indicator measures favoured by the experts for each of the health, risk and intervention areas 4. Conducting a series of experts' consensus workshops22 to refine the set of indicators and to develop ground rules and working definitions for the early warning and program development system for the chosen health outcomes 5. Determining the availability of existing databases for these indicators,
how to access such databases and 6. Evaluating the quality and developing methods for improving the quality of such existing databases 7. Identifying gaps in data availability and developing methods for collecting additional information for the surveillance system 8. Repeating steps 2-4 to identify, rank and refine the set of statistics to be generated from the surveillance system27 9. Repeating steps 2-4 to identify, rank and refine the methods of using surveillance data for public health practice (development and evaluation of prevention and control strategies) 10. Repeating steps 2-4 to identify, rank and refine the methods of timely information dissemination
Development of surveillance methodology is needed to accompany the development of a comprehensive surveillance system in the 21st century. Methodological challenges include (but are not limited to) the following areas: data collection, data analysis, data interpretation, public health practice, information dissemination and computer technology (Table 2). |
|||||||||||
|
|||||||||||
Discussion The establishment of a comprehensive, systematic, long-term and population-based epidemiologic surveillance system for early warning and program development, with indicator measures of health quality, risk factors and intervention practices, is very desirable for conducting research and setting health priorities in the 21st century. Such a system could be used to provide trend analysis, risk assessment and early warning of human health changes, to generate hypotheses for epidemiologic research, to produce evidence for program development and to evaluate prevention and control strategies. It is hoped that a comprehensive surveillance system could issue early warnings to the general population on emerging health problems. New health problems could be predicted based on current health trends, changes in risk factor prevalence, and prevention and control strategies. For example, if smoking prevalence increased, several tobacco-related diseases would be expected to rise. In this regard, the technique of disease modelling is potentially very useful. Another benefit of a comprehensive surveillance system is to develop and evaluate intervention strategies. If intervention strategies are effective, they should reduce subsequent health problems and risk factor prevalence. For example, if a smoking reduction program (intervention) is successful, it will be accompanied by a reduction in smoking prevalence (risk) and tobacco-related diseases (health). A comprehensive surveillance system should be based mainly on existing routine data collection, rather than creating a new system from scratch. In other words, planning, prioritization and co-ordination could put existing resources to better use. By linking various existing databases on health, risk and intervention variables, it is hoped that the full potential (early warning and program development) of a comprehensive surveillance system can be realized. Many of the current surveillance systems collect data only on health outcomes, risk factors or intervention practices, thereby limiting their uses. An efficient comprehensive surveillance system would not collect millions and millions of pieces of information. A systematic, evidence-based process would narrow down all the information to a list of indicator variables. For example, using the indicator approach, one does not need to measure the concentrations of all the gases in the air to ascertain air quality. Instead, one or two indicator gases, such as carbon dioxide and sulphur dioxide, may suffice. The process of narrowing down and selecting indicators must be systematic and evidence-based, not hypothesis-driven. The initial steps outlined in this paper are methods for such a systematic process. Many of the current surveillance systems collect data based on the recommendations of those people who are in charge and/or those clients who have the money to "buy in," both unsystematic approaches that can distort the information base of the system. The long-term nature of a comprehensive surveillance system should be stressed, since this enables the detection of trends. Ground rules and working definitions for a long-term data collection system must be developed. For example, the definition of a smoker should not be changed from year to year. One relevant ground rule could be that, if new knowledge dictates the need for a new definition, the old definition must be continued side by side in the database for a number of years, so as to provide a smooth transition. A surveillance system for the 21st century should be population-based or based on representative samples of the population. Such widespread data collection would help to reduce the false positive research cycle of sample-based studies described in this paper. A comprehensive system should evolve and improve with time, especially with respect to data accuracy. For example, record linkage and capture-recapture methodology may be used to improve data quality and to estimate the extent of missing information. Other methodological issues are also suggested in this paper for further work. This paper is intended to raise questions and stimulate debate on the important topic of epidemiologic surveillance in the 21st century. For example, is additional effort in population-based surveillance needed? Can surveillance tell us as much about etiology as sample-based studies? To answer the many problems we face will require collection of a tremendous amount of data. Can population-based surveillance collect detailed data efficiently? A reversion to studies of whole populations would obviously eliminate chance error and some types of selection bias (and possibly the discipline called "statistics"), but would it resolve confounding or information bias? Should ecologic correlation studies be used as early models for a future surveillance system? Will all associations with the same strength really have an equal chance of being detected? Since the number of persons exposed and affected will vary, the power to detect associations will vary as well. On the other hand, tests of statistical significance may not be needed, on the grounds that the whole population will be studied. How will we deal with the vast number of weak associations that will be uncovered by a population-based surveillance system? How would this widespread surveillance data be used? What of the ethical and legal problems posed by the need for privacy of individuals? Finally, who would or should fund the operations of such a comprehensive surveillance system? Current databases may be inaccurate, problematic and far from able to provide satisfactory information for a comprehensive surveillance system. Scientists must take that very first step now, so that a reasonably useful comprehensive system will be in place by the 21st century.
This paper is based on an invited presentation at the International Scientific Conference on Epidemiology held in Tianjin, China, September 16, 1997.
1. Graunt J. Natural and political observations made upon the Bills of Mortality. London: John Martin, James Allestry and Thomas Dicas, 1662. Republished Baltimore: The Johns Hopkins University Press, 1939. 2. Halperin W, Baker EL Jr. Public health surveillance. New York: Van Nostrand Reinhold, 1992. 3. Gordis L. Epidemiology. Philadelphia: WB Saunders Co, 1996. 4. Monson RR. Occupational epidemiology. Boca Raton: CRC Press, 1990. 5. Breslow NE, Day NE. Statistical methods in cancer research. Vol 1: The analysis of case-control studies. Lyon: International Agency for Research on Cancer, 1980. 6. Lilienfeld AM, Lilienfeld DE. Foundations of epidemiology. New York: Oxford University Press, 1980. 7. McNeil D. Epidemiological research methods. New York: John Wiley & Sons, 1996. 8. Health Canada. Report on the health of Canadians. Prepared by the Federal, Provincial, and Territorial Advisory Committee on Population Health. Ottawa: Health Canada, Communications and Consultation Directorate, 1996. 9. Health Canada. Public health intelligence reinvestment twelve-month progress report, 1996. Ottawa: Laboratory Centre for Disease Control, 1996. 10. Farmer R, Miller D, Lawrenson R. Lecture notes on epidemiology and public health medicine. Oxford: Blackwell Science Ltd, 1996. 11. Lomas J. Research and evidence-based decision making. Aust NZ J Public Health 1997;21:439-40. 12. Shah CP. Public health and preventive medicine in Canada. Toronto: University of Toronto Press, 1994. 13. Teutsch SM, Churchill RE. Principles and practice of public health surveillance. New York: Oxford University Press, 1994. 14. Sackett DL. Bias in analytic research. J Chronic Dis 1979;32:51-63. 15. Choi BCK, Noseworthy AL. Classification, direction, and prevention of bias in epidemiologic research. J Occup Med 1992;34:265-71. 16. Choi BCK, Pak AWP. Bias, overview. In: Armitage P, Colton T, eds. Encyclopaedia of biostatistics. John Wiley & Sons Ltd, 1998;1:331-8. 17. Kanarek MS. Indirect epidemiology studies in environmental hazard risk assessment. In: Richmond CR, Walsh PJ, Copenhaver ED, eds. Health risk analysis: proceedings of the Third Life Sciences Symposium. Philadelphia: Franklin Institute Press, 1981. 18. Murata M, Takayama K, Fukuma S, et al. A comparative epidemiologic study on geographic distributions of cancers of the lung and the large intestine in Japan. Jap J Cancer Res (Gann) 1988;79:1005-16. 19. Centers for Disease Control. CDC surveillance update. Atlanta (Georgia): CDC, 1988. 20. Monson RR. Editorial commentary: epidemiology and exposure to electromagnetic fields. Am J Epidemiol 1990;131:774-5. 21. Wilk MB. Health information for Canada: report of the National Task Force on Health Information. Final draft. Ottawa: Statistics Canada, 1991. 22. Jones J, Hunder D. Consensus methods for medical and health services research. Br Med J 1995;311:376-80. 23. Beech B. Studying the future: a Delphi survey of how multi-disciplinary clinical staff view the likely development of two community mental health centres over the course of the next two years. J Adv Nurs 1997;25:331-8. 24. Olmstead-Schafer M, Story M, Haughton B. Future training needs in public health nutrition: results of a national Delphi survey. J Am Diet Assoc 1996;96:282-3. 25. Petitti DB. Meta-analysis, decision analysis, and cost-effectiveness analysis: methods for quantitative synthesis in medicine. New York: Oxford University Press, 1994. 26. Choi BCK, Pak AWP, Leake JL. Epidemiologic design options for health risk studies on dental amalgam. Can J Comm Dent 1996;11:7-12. 27. Choi BCK. Recording, notification, compilation, and classification of statistics of occupational accidents and diseases: the Thai experience. J Occup Environ Med 1996;38:1151-60. 28. Peron Y, Strohmenger C. Demographic and health indicators: presentation and interpretation. Ottawa: Statistics Canada, Health Division, Research and Analysis Section, 1985; Cat 82-543E. 29. Choi BCK, Pak AWP, Purdham JT. Effects of mailing strategies on response rate, response time, and cost in a questionnaire study among nurses. Epidemiology 1990;1:72-4. 30. Choi BCK, Hanley AJG, Holowaty EJ, Dale D. Use of surnames to identify individuals of Chinese ancestry. Am J Epidemiol 1993;138:723-34. 31. Choi BCK. N-nitroso compound and human cancer: a molecular epidemiologic approach. Am J Epidemiol 1985;121:737-43. 32. Choi BCK, Connolly JG, Zhou RH. Application of urinary mutagen testing to detect workplace hazardous exposure and bladder cancer. Mutation Res 1995;341:207-16. 33. Nanan DJ, White F. Capture-recapture: reconnaissance of a demographic technique in epidemiology. Chronic Dis Can 1997;18(4):144-8. 34. Choi BCK, Robson L, Single E. Estimating the economic costs of the abuse of tobacco, alcohol and illicit drugs: a review of methodologies and Canadian data sources. Chronic Dis Can 1997;18(4):149-65. 35. Ying JYC, Abernathy T, Choi BCK. A comprehensive evaluation of the 1993 City of Toronto smoking by-laws. Can J Public Health 1995;86:32-6. 36. Choi BCK. Risk assessment in dentistry: health risks of dental amalgam revisited. J Can Dent Assoc 1996;62:328-32.
Author References |
![]() |
|||
![]() |
Last Updated: 2002-10-22 | ![]() |