Volume 22, No. 2
2001

[Table of Contents]

Assessing the Surveillance Capability of Canada's National Health Surveys

Gary J Umphrey, Ora Kendall and Ian B MacNeill

Abstract

We assessed Canada's national health surveys as surveillance instruments, with emphasis on comparing the temporal structure of data sets with those generated by the US Behavioral Risk Factor Surveillance System (BRFSS). Only the Canadian Tobacco Use Monitoring Survey (CTUMS) has the BRFSS capability to generate continuous, uniform time series with monthly intervals. These time series can offer substantial extra value for retrospective analysis such as program evaluation in addition to surveillance. Expanding CTUMS is a simple option for providing an ongoing, uniform monthly survey instrument for non-tobacco variables. The Canadian Community Health Survey (CCHS) will generate monthly data, and could potentially generate useful continuous time series even though surveys at the health region and provincial levels will alternate annually. Reconfiguring the CCHS, or even implementing a provincial surveillance survey based on the BRFSS model are other viable options, but each option has associated tradeoffs or obstacles.

Key words: BRFSS; CCHS; CTUMS; health information; health surveys; surveillance time series; temporal change

Introduction

Improving surveillance has been a significant issue in recent discussions on upgrading Canada's health information infrastructure.^1-5 Specific actions to meet this demand include the introduction of two new national health surveys, the Canadian Tobacco Use Monitoring Survey (CTUMS) and the Canadian Community Health Survey (CCHS). While the new surveys will add greatly to our surveillance capability, we believe that there may still be scope for further cost-effective improvement. We will focus on data set temporal structure, since the way data are generated over time is a fundamental property of survey design configuration and has important implications for surveillance timeliness and statistical analysis.

National health surveys are not the only tools available for surveillance. Compelling cases can also be made, for example, for strengthening Canadian capability in the surveillance of policy implementation⁶ and for making better use of numerous potential sources of surveillance data at local levels.⁷ But by generating population-based data for a wide range of health-related variables, national health surveys contribute to surveillance by providing:

• point-in-time, demographic portraits of population health;
• a means for detecting temporal changes in health factors;
• information on associations among variables;
• a scientifically credible basis for cross-validating information from other surveillance instruments, especially those that generate "quick and cheap" data (e.g., sales records); and
• a basis of comparison for localized or specialized surveillance activities.

The national survey data sets tend to have lasting value. Longer time series with high, consistent data quality have greater analytic value, and this places a premium on establishing survey systems that will be maintained over time.

Definitions of public health surveillance typically emphasize that it must be ongoing, a prerequisite for detecting change over time, and it must be integrated with advancing public health goals.^3,8 At its root, proper surveillance involves "keeping a close watch", an idea that goes beyond simply monitoring for temporal change. How close the watch should be is dependent on what is being watched. Both incoming ballistic missiles and advancing glaciers can have enormous impacts, but it would be a waste of resources to conduct surveillance on the latter at the intensity required for the former.

Seismographs and smoke detectors are in one sense ideal surveillance instruments in that they operate continuously. Little would be saved by turning them off for intermittent periods brief enough that their function would not be compromised. Similar instruments for public health surveillance would, as noted by McQueen in reference to lifestyle factors in particular, "[collect] data continuously, producing a seamless flow of data that can detect subtle and long-term changes in the variables of interest at the population level".⁹ However, the significant cost associated with each interview and statistical considerations relating to desired precision and power mean that, for a fixed cost, there are real trade-offs between the quantity and quality of information that can be acquired for each point-in-time cross-sectional survey and survey frequency.

The US Behavioral Risk Factor Surveillance System (BRFSS) conducts ongoing monthly surveys that can generate continuous uniform time series. This is likely as close to the ideal of seamless data flow as is practical for a national health survey. The same design for generating data over time has been adopted by CTUMS. Some similar design elements suggest that the BRFSS also influenced the CCHS. Some key differences remain between the BRFSS and the Canadian surveys, however, so it is part of our methodology to use the BRFSS as a comparative model.

In operation for over a quarter century, the BRFSS has been refined with experience and has built a substantial, well-documented record of use. The BRFSS has served as a model for surveillance surveys in other countries, such as Australia and China. In addition to being ongoing, the BRFSS has the flexibility to meet federal, state and local needs. There has always been a need in Canada for more focused provincial and local information as it is in the provincial and local mandates to develop their own policies and deliver public health services.

We take a broad view of surveillance. Differentiating the function of surveillance from other potential uses of data, such as research, is a useful point of distinction for assessing surveillance adequacy,^8,10 but the cost-benefit analysis for evaluating a surveillance instrument should include all potential benefits, whether or not they strictly constitute surveillance per se. In addition to the "hard" benefits arising from the data generated, the instrument can produce "soft" benefits by motivating information acquisition, dissemination and use.

Potential changes in a health information infrastructure or a component such as a survey can be assessed in terms of marginal utility. We propose that an optimal surveillance system would be one for which increasing or reallocating resources could not yield cost-effective gains in information value, nor could resources be reduced without a loss of information value that outweighs the savings achieved. In the cost-benefit analysis this entails, it is typically easier to quantify costs than benefits. Reductions in mortality and morbidity and program savings are quantifiable measures for validating change. Other benefits may be more difficult to quantify, especially in economic terms, yet these may be the most important justification for a program. These include:

• better knowledge of the awareness and impact of policies and programs;
• more effective delivery of health services that may not save money, but do improve client health, comfort, or convenience;
• better understanding of the factors affecting human health;
• development of ancillary information that places targeted research into a broader context and assists in prioritizing further research, surveillance, and policy initiatives;
• assurance that newly emerging health issues will be identified and acted upon in a timely manner;
• assurance that current programs are still appropriate and adequate; and
• greater ability to address the health needs of special subgroups.

Such diverse considerations ensure that "optimization" is not a simple algorithmic exercise, but more of a balancing act. Formulating a coherent surveillance strategy should hasten the evolutionary process of refining the surveillance system and ultimately reduce development costs.

Background and Overviews

US Behavioral Risk Factor Surveillance System (BRFSS)

The BRFSS is a state-based, ongoing survey that conducts a number of random-digit-dialed telephone interviews each month (see the BRFSS web site for extensive documentation).¹¹ Formally established in 1984 by the Centers for Disease Control and Prevention (CDC) and 15 participating states, the BRFSS grew to include all 50 states, the District of Columbia, and three territories by 1997. The target population consists of those 18+ years old living in households. (The US has a separate school-based Youth Risk Behavior Surveillance System, not treated here.)¹² State sample sizes in 1999 ranged from 1,248 to 7,543 per year, with a median of 2,939 (D. Nelson, personal communication). The CDC attempted to fund about 2,500 interviews per state in 1999; state discretion in spending CDC funds and state-funded additional interviews account for the variation in sample size.

Key BRFSS attributes are the nature of the federal-state collaboration, the practice of conducting surveys at monthly intervals, and flexibility. The CDC coordinates the BRFSS, designs a substantial part of the questionnaire, provides logistical support, and supplies core funding. The questionnaire has three parts: a core component, common for all states and prepared by the CDC, consisting of fixed, rotating, and emerging core questions; optional modules prepared by the CDC but selected for inclusion by each state; and questions that are developed or acquired by each state. State-added questions are rated highly for utility by BRFSS program directors.¹³ Each state conducts its own survey. Implementation differs somewhat among states; for example, some use in-house units to conduct the interviews, while others contract them out. The state-based survey system is believed to make it easier to address local issues, such as the impact of a new state program.

Spreading the annual sample evenly over 12 months keeps the BRFSS ongoing, which in turn benefits data quality, data analysis, and the capacity to react to emerging issues. Employees can be kept on a continuing basis, and interviewers can be highly trained and can cost-effectively gain experience and consistency over time. It is also BRFSS experience that a bias shift (such as a change in the proportion underreporting a particular behaviour) is more readily detected with continuous monthly data if a question or methodology is modified (D. McQueen, personal communication).

The data sets generated potentially offer substantial added value for statistical analysis in comparison to annually aggregated data sets. The continuous time series with monthly intervals generated for fixed core questions have the following attributes:

• the data are better suited for detecting temporal changes;
• data are collected at frequent enough intervals that the effect of a program or policy can be tracked over time, a valuable property when the effect occurs quickly;
• data can be flexibly aggregated to larger time periods (e.g., annually, semi-annually, quarterly), or to make before-and-after comparisons when a new program is initiated or other potential impacts occur;
• seasonality components can be examined, both for their own sake and to de-seasonalize the monitoring of temporal trends; and
• statistical procedures analogous to quality control techniques (e.g., p-charts, CUSUM charts⁵) would allow fast response to statistically significant changes in time.

Rotating core questions generate 12 months of data cycling with a 12-month gap. Typically, such data are pooled by year and treated as equivalent to biennial surveys, but seasonality components can be examined with the monthly data. Rotating core modules are made available as optional modules in off-core years, so individual states can maintain ongoing surveillance on these variables. Illinois splits the state sample and runs dual questionnaires (current year plus previous year), so that all "rotating core" modules are run continuously (B. Steiner, personal communication).

In practice, most data are rolled up annually for reporting and analysis.¹⁴ Examples of monthly data use include an examination of seasonal patterns of leisure-time physical activity¹⁵ and a time series analysis of trends in perceived cost as a barrier to medical care.¹⁶

Two other lessons have emerged from BRFSS experience. First, the use of surveillance data builds over time.¹³ Second, concentrating resources on "front end" data collection at the expense of "back end" analysis can result in sub-optimal information yield.⁹

Canadian National Health Surveys to 1998

The Canadian Sickness Survey in 1950/51 initiated what is now a half-century tradition of national population-based health surveys.¹⁷ These surveys have provided a wealth of health information. Repeating surveys typically provide the best basis for detecting temporal change. Current national surveys that are intended to be repeated on an ongoing basis include the National Population Health Survey (NPHS), General Social Survey (GSS), Tracking Nutrition Trends Survey, Health of Canada's Youth Survey (HCYS), National Longitudinal Survey of Children and Youth (NLSCY), Physical Activity Monitor, and National Vaccine Coverage Survey.¹⁷

Information on the same variable has also been collected over time under surveys of different names. Results from an eclectic collection of surveys can be compared if the surveyed population and the wording of the questions are closely comparable, and if survey methodologies do not introduce significant biases or if the biases are at least consistent.

Papers on tobacco use by Stephens^18-19 and Gilmore²⁰ illustrate various strengths and weaknesses in the set of national health surveys for monitoring temporal change prior to 1998. Stephens'¹⁸ 1988 review of tobacco use, attitudes and knowledge used four different survey sources: the Labour Force Survey (LFS) for 11 different years between 1965 and 1986, the Canada Health Survey for July 1978 to March 1979, the Health Promotion Survey for June 1985, and the General Social Survey (GSS) for October 1985. Factors enhancing integration of the multiple survey sources were:

• similar sampling designs;
• almost identical wording for relevant questions;
• Statistics Canada collected all data; and
• well documented methods allowed informed judgments on variations in approach.

Consistency was hindered, however, by variations in methodologies (e.g., both household visits and telephone interviews were used) and particularly by the acceptance of proxy data in the LFS. Estimates of tobacco use in youth by proxy data were shown to have a strong downward bias, and proxy report data could not be separated from self-report data prior to 1981.

Stephens¹⁹ reported on the results of a workshop held to examine trends in smoking prevalence from 1991 to 1994. A landmark event during this period was the federal action on February 8, 1994 (with concordant actions by five provinces on or shortly thereafter) to sharply reduce tobacco taxes in order to combat cigarette smuggling. Key information on smoking prevalence was provided by the 1991 GSS and the 1994 Survey on Smoking in Canada, but no comparable information was available for smoking prevalence in 1992 and 1993. Instead, smoking prevalence for these two years had to be inferred from information gleaned from several smaller national and provincial surveys, including three commercial surveys with relatively small sample sizes but consistent time series, as well as industry consumption statistics.

Gilmore²⁰ analysed the statistical significance of changes in tobacco use prevalence during the period 1985-1999 using only Statistics Canada surveys. This study demonstrated the difficulties in detecting change with continually changing surveys. The GSS for 1991 and 1996, the NPHS for 1994/95 and 1996/97, and CTUMS for the first half of 1999 were judged to be comparable surveys. The National Alcohol and Drugs Survey of 1989, the Health Promotion Survey of 1990, the Canadian Alcohol and Drugs Survey of 1994, and the GSS for 1995 were judged as reasonably comparable to the other surveys for daily smoking rates, but current and non-daily rates were judged as not comparable due to question differences.

We say little about these surveys in the following sections for several reasons. The NPHS and NLSCY are longitudinal surveys, which have limitations as surveillance instruments.^5,21 The GSS now has limited health-related content. The topic of youth surveillance merits separate treatment, and the remaining surveys cover specialized topics. The role and design of these surveys for surveillance will need to be reappraised with a view to how they fit in with the surveillance capabilities of the CCHS and CTUMS. However, we expect the new surveys to have relatively little impact on most of the previously established surveys, which generally provide complementary health information.

Canadian Tobacco Use Monitoring Survey (CTUMS)

CTUMS conducts monthly computer-assisted telephone interview surveys.²² The target population consists of those aged 15+ years, with residents of the territories and full-time residents of institutions excluded. The annual sample size is targeted at 20,000, partitioned equally among provinces and between the two age groups of 15-24 and 25+. The first questionnaire contained 35 questions related to tobacco use plus demographic questions. Analysis of the first wave of surveys, covering February to June 1999, was released in mid-January 2000. Reporting is scheduled for each half calendar year.²²

Canadian Community Health Survey (CCHS)

The CCHS started in September 2000.²³ The design (as of November 1, 2000) consists of two surveys alternating annually in a two-year cycle (briefly summarized here, the design is discussed in greater detail by Béland et al.²⁴). The health region level survey in the first year of the cycle has a customized component to meet the individual priorities of the 136 health regions. The target population consists of household residents 12+ years old in all provinces and territories, excluding primarily those on native reserves, Canadian Forces bases and some remote areas. Youths (12-19) and seniors (65+) will be systematically oversampled. The aggregate sample of 130,750 will consist of 115,000 computer-assisted personal interviews (CAPI) and 15,750 computer-assisted telephone interviews (CATI); this breaks down into samples of 2,000-42,260 per province, 800-900 per territory and 280-3410 per health region.^23-24 The second-year provincial-level survey will consist of 30,000 computer-assisted personal interviews. The target population will differ somewhat from the health region-level survey; for example, the first provincial-level survey, focusing on mental health, will be restricted to those 15+ (B. Diverty, personal communication).

The CCHS broadly covers health determinants, health status and health system utilization. The 45-minute health region-level survey has 30 minutes of common content, 10 minutes of optional content selected from a set of modules, and five minutes of socioeconomic and demographic content. The provincial-level survey will contain additional common content and one focus content topic (an in-depth treatment of a topical issue);²³ the first is expected to run at least one hour (B. Diverty, personal communication). Quarterly release for high-level population health indicators is planned.²³

Comparing CTUMS and the CCHS to the BRFSS

CTUMS and the BRFSS fixed core component generate data sets with the same temporal structure: ongoing uniform time series at monthly intervals. CTUMS is restricted to tobacco use and has a richer tobacco content, but the BRFSS fixed core includes a broader range of health variables plus additional tobacco use content in the optional modules.

CCHS data sets generated for questions unique to either the health region-level survey or provincial-level survey that are repeated in subsequent cycles will have the same temporal structure as those generated by the BRFSS's rotating core: 12 monthly estimates alternating with 12 months without data. Some common content is expected to be carried over from the health region-level survey to the provincial-level survey. Since health region-level data can be rolled up to provide provincial-level estimates, continuous monthly-interval time series can be generated for common content questions. These time series would have the peculiar property of being generated by biennially rotating surveys. Results otherwise obtainable through time series analysis could be obscured by differing methodologies, but options exist to minimize or eliminate such differences. For example, by excluding CATI data, CAPI-only estimates could be generated for health region-level survey years, essentially matching the methodology of the CAPI-only provincial-level survey years (G. Catlin, personal communication).

We believe that the impact of the 1994 tax reduction on cigarettes would have been more readily and clearly discerned if a monthly surveillance survey had been in place. With CTUMS now operational, it can be argued that the need has been met for generating uniform time series on tobacco use at national and provincial levels. But the case for maintaining a similar level of surveillance on certain other health variables is reasonable, and it would be prudent to have an instrument in place for close surveillance if other needs emerge, such as tracking the temporal and demographic dynamics of use and exposure to the drug Ecstasy. In the next section we briefly outline some scenarios for enhancing surveillance.

Options for Enhancing Surveillance

Adapt the BRFSS Model for Canada

It is useful to consider what a Canadian adaptation of the decentralized BRFSS survey model might look like. Under various allocation schemes, a national sample of roughly 20,000-30,000 would match information quality at the province-state level. Following the current BRFSS standard of 2,500 per state would result in a national annual sample of 25,000 for the provinces (the current design of CTUMS is similar to this); additional but likely smaller samples would be required to cover the territories. Since Ontario and Quebec carry such heavy weight in the national roll-up (much more so than any pair of states in the US), an allocation scheme designed to balance reliability requirements between national and provincial or territorial levels may be preferable. A scheme employed by the 1994 NPHS resulted in a core national sample of 22,000, with 1,200 (a set minimum) to 1,996 sampled per province or territory except for Ontario (4,817) and Quebec (3,584).²⁵ Some provinces or territories could choose to share a single survey unit yet administer separate questionnaires.

Advantages of this option include the close involvement of the provincial health systems with the national surveillance system, the flexibility that individual provinces and territories would have on questionnaire content, and BRFSS experience and expertise. We would expect separate Canadian-designed questionnaires, but any questions duplicated with the US survey would allow nation-to-nation or province-to-state comparisons. These comparisons could be of interest given the very different US health care model.

Reconfigure the CCHS

As noted previously, there is at least the potential for the CCHS to generate continuous time series, but issues of data uniformity and content capacity remain to be resolved. If the CCHS had been designed so that the health region-level survey was spread evenly over 24 rather than 12 months, an instrument would already exist for conducting ongoing uniform surveys at monthly intervals for a wide range of health variables. The annual national sample of 65,000 would generally provide information for provincial roll-ups comparable to or exceeding the usual BRFSS state standards; annual samples would exceed 2,000 for all provinces except PEI (1,000). Even the larger health regions would have quite reasonable annual samples. Idaho is one of the few states that stratifies BRFSS sampling by health district, with each targeted for 700 interviews annually (J. Aydelotte, personal communication).

While it may have been operationally simpler to spread the very large health region survey over two years, the current CCHS design appears to have been driven by a requirement to deliver health region-level information quickly. Concentrating the smaller samples for the sparsely populated health regions also has analytic benefits; for example, there were concerns about the validity of estimates for smaller health regions if sampling was spread over 24 months. However, if the health region-level survey was operated continuously, moving averages may have advantages as a way of reporting health region information.

It is hard to fault the design tradeoffs made - the CCHS can deliver health data at an unprecedented level of geographic resolution with the same temporal structure as the BRFSS rotating core, and there is flexibility for selecting content to meet provincial and health region priorities. It may very well be that little extra value is to be gained from closer surveillance on most variables. But we do not believe this is true for tobacco use, and it cannot be assumed true of all other variables. How desirable or necessary it is to reconfigure the CCHS to meet these needs will depend, in part, on the yet-to-be-determined surveillance capability of the common content component, and possibly through insights gained during the first survey cycle. Even if the CCHS is capable of meeting needs for generating continuous monthly time series, the question of whether it is the most efficient instrument for doing so will need to be addressed.

Expand CTUMS to Include Other Variables

Since CTUMS is already operating as an ongoing monthly telephone survey, it would seem relatively simple to expand its scope to include some other variables. Oversampling of youth and young adults is a useful design feature for surveillance of other risk factors. Even if the survey was renamed, the value of existing data would remain intact.

Certainly there is room to add questions while keeping the questionnaire size reasonable. By way of illustration, adding the entire BRFSS fixed core of health-related non-tobacco questions to the CTUMS questionnaire would still leave it with under 90 questions.

This option is likely to be resisted by CTUMS stakeholders if they perceive that expanding the scope of the survey would result in reduced tobacco use content and/or in a more cumbersome survey that delivers less timely information. But we see no reason why these concerns cannot be met, and the assurance of stable funding and possibly increased sample sizes and extended demographic coverage would be powerful inducements to change.

Discussion

A health information system will continually evolve as it responds to changes in demographics, spending priorities, medical technologies, health care practices, and information technologies. A surveillance system needs both flexibility, to respond to new information and priorities, and stability, to facilitate the detection and evaluation of temporal change. Canada already has several national health surveys in place that primarily require stable funding to continue to act as surveillance instruments.

Populations, needs, resources, institutions and traditions differ between countries. We see both considerable merit and substantial obstacles to adapting the BRFSS model. The decentralized survey structure of the BRFSS generates significant benefits by motivating interest in developing and using health information at state and regional levels. The argument that Statistics Canada should be able to achieve more uniform methodology for better comparisons among provinces and territories seems sound, yet numerous studies have rated BRFSS data highly for reliability and validity.¹⁶ The specialized expertise required to extract greater value from surveillance data with more sophisticated statistical analyses is likely more efficiently provided at the national level.

Among the options presented for generating uniform time series at monthly intervals, the simplest is to expand CTUMS to include other variables. If the CCHS retains its current design, one of the two surveys would be able to generate a data set with the same temporal structure as any BRFSS variable. The common content component of the CCHS could potentially generate satisfactory, if not completely uniform, continuous time series. The CCHS could potentially subsume CTUMS if matters of content depth and timeliness were resolved, but this may not be a realistic alternative. An expanded CTUMS would take some content pressure off the CCHS, which appears to be already saturated for questionnaire content. We note here that the BRFSS, despite having a somewhat narrower focus than the CCHS, may implement dual questionnaires in order to cope with demands for increased content (D. Nelson, personal communication).

Ferrence and Stephens⁶ see CTUMS and the CCHS as being complementary rather than competing surveys for the surveillance of tobacco use. CTUMS is expected to be more timely (8-9 months estimated total turnaround time versus 2-2.5+ years for the CCHS; main difference is in time to introduce new topics), has richer tobacco use content, monitors tobacco use only, and generates continuous uniform time series at monthly intervals, while the CCHS has lower age coverage (12+ years versus 15+ years for CTUMS) and superior small area coverage in the health region level survey. This complementary relationship should remain intact if close surveillance needs for non-tobacco variables are met with an expanded CTUMS or through the CCHS common content core.

The question of which variables merit closer surveillance requires further consideration. The dynamic nature of tobacco use and its premier status as a preventable health risk likely makes it the best choice,²⁶ but it is unlikely to be the only good one. The 2000 BRFSS fixed core was composed of four questions on health status, seven on health care access, one on diabetes, five on tobacco use, 11 on women's health, and 11 on HIV/AIDS. Is this set necessary and sufficient? In planning for the next decade, proposed criteria for selecting BRFSS core measures are based on public health impact, scientific validity, data utility, and implementation considerations.²⁷ However, placement between the fixed and rotating core is still an open question. Theoretically, close surveillance of any variable is warranted if sufficient extra value is expected. We suggest that factors that can be heavily impacted by government policy, especially through legislation or taxation, merit inclusion. For such variables the greatest value of surveillance data may come through a clearer picture of the impact (or lack of impact) by retrospective analysis of the time series overlapping the time of change.

Viewed from the perspective of data set temporal structure, the set of Canadian national health surveys is close to matching all BRFSS capabilities. The BRFSS still has some flexibility that the Canadian national health surveys do not, and in a coordinating role the CDC is perhaps better situated than Statistics Canada for ensuring that survey content is best designed to meet public health information needs. Statistics Canada held a broad consultation process with users of health information to determine content for the CCHS,²⁸ but we consider the onus to be on the Canadian health community to ensure content quality. The CCHS survey provides considerable flexibility and an additional level of geographic resolution, but we believe it is best to ensure the capability of conducting close surveillance on variables other than tobacco use. Key issues and questions to address in the further evolution of the national health surveys relate to:

• what content to include and what levels of temporal and demographic resolution are in some sense optimal for each variable;
• adequacy of coverage for youth and other special populations;
• extracting greater value from data through increased application of specialized statistical methods;
• integrating information from survey and non-survey sources;
• ensuring appropriate support and flexibility in assisting provinces/territories and health regions to meet their information needs; and
• obtaining maximal information transferral to health policy and practice.

Acknowledgements

We wish to thank Michael Ackland (Victorian Department of Human Services, Melbourne, Australia), James Aydelotte (Idaho Department of Health and Welfare), Gary Catlin (Statistics Canada), Margaret de Groh (Health Canada), Brent Diverty (Statistics Canada), Richard Mathias (University of British Columbia), David McQueen (CDC), David Nelson (CDC), Bruce Steiner (Illinois Department of Health) and two anonymous reviewers for their assistance in the preparation of this paper.

References

1. Mowat D. Guest editorial: health surveillance in Canada. Chronic Dis Can 1998;9(4): 143-144.

2. Choi BCK. Perspectives on epidemiologic surveillance in the 21st century. Chronic Dis Can 1998;19(4):145-151.

3. Health Canada. A discussion paper on an integrated national health surveillance network for Canada. Federal/Provincial/Territorial Surveillance Integration Design Team. Ottawa, September 1998.

4. Wigle D, Mowat D. Health surveillance: changing needs, constant function. Can J Public Health 1999;90(3):149-150.

5. MacNeill IB, Umphrey GJ. Concept paper on Canadian health surveys and risk factor surveillance. Prepared for the Laboratory Centre for Disease Control, Health Canada, Ottawa. March 11, 1998.

6. Ference R, Stephens T. Monitoring tobacco use in Canada: the need for a surveillance strategy. Chronic Dis Can 2000;21(2):50-53.

7. Sherman GJ, Mathias RG. Surveillance support for population health decision making. Can J Public Health 1996;87(4):221-222.

8. Thacker SB, Berkelman RL. Public health surveillance in the United States. Epidemiol Rev 1988;10:164-189.

9. McQueen DV. A world behaving badly: the global challenge for behavioral surveillance. Am J Public Health 1999;89(9):1312-1314.

10. Teutsch SM, Thacker SB. Planning a public health surveillance system. Epidemiol Bull 1995;16(1):1-6.

11. Centers for Disease Control and Prevention. Behavioral Risk Factor Surveillance System. Web site at: http://www.cdc.gov/nccdphp/brfss/index.htm. Accessed June 16, 2001.

12. Centers for Disease Control and Prevention. Youth Risk Behavior Surveillance System. Web site at: http://www.cdc.gov/nccdphp/dash/yrbs/index.htm. Accessed June 16, 2001.

13. Figgs LW, Bloom Y, Dugbatey K, Stanwyck CA, Nelson DE, Brownson RC. Uses of Behavioral Risk Factor Surveillance System data, 1993-1997. Am J Public Health 2000;90(5):774-776.

14. Centers for Disease Control and Prevention. Behavioral Risk Factor Surveillance System. Reprints from the MMWR 1990-1998. Atlanta, Georgia: BRFSS.

15. Centers for Disease Control and Prevention. Monthly estimates of leisure-time physical inactivity - United States, 1994. MMWR 1997;46(18):393-397.

16. Nelson DE, Thompson BL, Bland SD, Rubinson R. Trends in perceived cost as a barrier to medical care, 1991-1996. Am J Public Health 1999;89(9):1410-1413.

17. Kendall O, Lipskie T, MacEachern S. Canadian health surveys, 1950-1997. Chronic Dis Can 1997;18(2):70-94.

18. Stephens T. A critical review of Canadian survey data on tobacco use, attitudes and knowledge. Report prepared for the Tobacco Programs Unit, Health Promotion Directorate, Health and Welfare Canada, Ottawa 1988.

19. Stephens T. Workshop report: trends in the prevalence of smoking, 1991-1994. Chronic Dis Can 1995;16(1):27-32.

20. Gilmore J. Report on smoking prevalence in Canada, 1985 to 1999. Statistics Canada, Catalogue 82F0077XIE, 2000.

21. Kalton G. 1992. Panel surveys: adding the fourth dimension. Proceedings of the Statistics Canada Symposium 92. Design and Analysis of Longitudinal Surveys, 1992; pp. 7-18.

22. Health Canada. Canadian Tobacco Use Monitoring Survey. Web site at: http://www.gosmokefree.ca/ctums. Accessed June 16, 2001.

23. Statistics Canada. The Canadian Community Health Survey (CCHS). Web site at: http://www.statcan.ca/english/concepts/health/ccshinfo.htm. Accessed June 16, 2001.

24. Béland Y, Bailie L, Catlin G, Singh MP. CCHS and NPHS - An improved health survey program at Statistics Canada. American Statistical Association Proceedings 2000; Survey Research Section. To be published.

25. Singh MP, Tambay JL, Krawchuk S. The National Population Health Survey: design and issues. American Statistical Association Proceedings 1994; Survey Research Section, p. 803-808.

26. Ellison LF, Morrison HI, de Groh M, Villeneuve PJ. Health consequences of smoking among Canadian smokers: an update. Chronic Dis Can 1999;20(1):36-39.

27. Centers for Disease Control and Prevention. Proposed criteria for inclusion of measures in the BRFSS core instrument. Draft, October 5, 1999.

28. Statistics Canada. First steps: report on consultations for the proposed Canadian Community Health Survey (CCHS). Fall 1999 (draft).

Author References

Gary J Umphrey, Department of Mathematics and Statistics, University of Guelph, Guelph, Ontario

Ora Kendall, Quantitative Analysis and Research Division, Public Health Agency of Canada, Health Canada, Ottawa, Ontario

Ian B MacNeill, Department of Statistical and Actuarial Sciences, The University of Western Ontario, London, Ontario

Correspondence: Dr Ora Kendall, Quantitative Analysis and Research Division, Public Health Agency of Canada, Health Canada, Tunney's Pasture AL: 1907A3, Ottawa, Ontario K1A 0B4; Fax: (613) 952-0844; E-mail: Ora_Kendall@hc-sc.gc.ca

[Previous] [Table of Contents] [Next]