Studies of cervical and breast cancer have been the “pioneers” of behavioral and social science research concerning cancer screening, and now a literature on colorectal cancer screening also is growing steadily. Our article is intended to complement the introductory report 1 in this current supplement to Cancer and adopts that article's perspective on the breadth of areas that comprise intervention-related, cancer screening research.
The first section offers several lessons learned to date during the evolution of the behavioral and social science literature. These lessons also may be important for anticipating research priorities as new areas of cancer screening emerge in the coming years (i.e., new technologies for an existing domain of screening, and technologies for new domains of screening). The comments for each lesson learned are brief but are intended to direct attention to the theme (see Table 1 for a list of the lessons learned).
Table 1. Lessons Learned Regarding Perspectives on Behavioral and Social Science Research on Cancer Screening
|Lesson 1: The findings from epidemiologic surveillance of morbidity and mortality are a strong basis for identifying populations to target for interventions.|
|Lesson 2: Progress in cancer control requires detailed information about individual cancer screening behaviors.|
|Lesson 3: The priorities of behavioral and social science research in cancer screening are closely tied to the identification of groups with low utilization of screening, as defined by sociodemographic variables and psychosocial factors.|
|Lesson 4: Monitoring and responding to the diversification of personal screening histories are needed.|
|Lesson 5: Behavioral and social science researchers should be involved with the evaluation of the use of new technologies.|
|Lesson 6: Behavioral and social science research on cancer screening has become a field of study with its own particular features.|
|Lesson 7: Multilevel theoretical and conceptual approaches are needed to understand the full context of cancer screening.|
The second section cites several emerging topics that will be important for the future, but about which we have not yet learned enough to draw lessons. We faced the same challenge as other authors in the current issue, namely, extracting lessons from a literature that is still evolving. Our comments, therefore, reflect summary impressions, rather than a historic review. We believe that our article will generate similar efforts, by others, to fill gaps and offer additional perspectives. 2
BEHAVIORAL AND SOCIAL SCIENCE RESEARCH ON CANCER SCREENING: LESSONS LEARNED
Lesson 1: The Findings from Epidemiologic Surveillance of Morbidity and Mortality are a Strong Basis for Identifying Populations to Target for Interventions
The connection between trends shown in registry-based surveillance data and the priorities of intervention research is so natural that it can almost be taken for granted. Registry-based surveillance provides unique opportunities to determine the magnitude of the cancer problem, to identify trends, and to establish priorities for intervention-related research. Among the more prominent examples are disparities across sociodemographic groups, such as stage at diagnosis and treatments, 3–5 and breast cancer mortality monitoring that consistently demonstrates higher mortality rates among African-American women despite a lower incidence rate.6–8 Population-based data from the late 1980s into the later 1990s also documented lower screening rates among African Americans and other women of color.9–15 Recent years have seen an apparent closing of the screening gap between African Americans and non-Hispanic white women in population-level surveys.16 The finding that mortality rates have not equalized means that more attention should be directed to areas that can explain why disparities persist. These likely will be areas to which behavioral and social science can contribute (e.g., validity of the data on reported screening, types of treatments prescribed, follow-through with the full course of treatments, and macrolevel barriers that still work against women of color and against women from other underserved groups). The lesson learned is the powerful potential that comes from the integration of epidemiology with behavioral and social science.17 This integration has been a hallmark of applied cancer control research and should characterize work related to future screening modalities.
Lesson 2: Progress in Cancer Control Requires Detailed Information Regarding Individual Cancer Screening Behaviors
The contribution of epidemiologic surveillance to the priorities of intervention is important, but it is still only one piece of the larger context of behavioral and social science research. Registry-based mortality and morbidity statistics illustrate the disparities, but these statistics per se rarely give specific direction regarding how to proceed. Surveillance must also track adoption of the screening procedure itself, as well as the potential correlates of being and not being screened.
The National Health Interview Survey (NHIS) and the Behavioral Risk Factor Surveillance System (BRFSS) have become the primary sources for population-level surveillance of the prevalence of cancer control practices. Questions to assess “ever-had” screening status and “most recent” receipt of a procedure are the most obvious to ask, and they have been asked almost exclusively. To our knowledge, very few assessments have obtained rates of consecutive or repeat on-schedule examination. For example, the Year 2000 NHIS Cancer Control Module requested information about the total number of mammograms in the past 6 years, but to our knowledge this was the first instance of a survey requesting information about more than 1 mammogram since the 1990 NHIS Disease Prevention and Health Promotion Supplement. The 2001 California Health Interview Survey, like the Year 2000 NHIS, asked for the number of mammograms over 6 years. The BRFSS has not asked about repeat mammography in its core survey component, although the 2003 Women's Health module, optional for that year, does ask about the woman's next most recent mammogram. The 2002–2003 Health Information and National Trends Survey (HINTS), sponsored by the National Cancer Institute, assessed the timing of the two most recent examinations, with questions worded similarly to the 2003 BRFSS. The HINTS also asked about next most recent Papanicolaou (Pap) testing and colorectal examinations. These are all recent surveys, but they promise increased attention to repeat screening.
The time constraints faced by population-based surveys, like those noted earlier, are a realistic consideration. Assessing repeat screening in any one domain requires extra questions beyond those used for ever-had and most-recent status, and the number of questions becomes even larger if there are multiple ways to meet the guidelines (as for colorectal screening). There is, however, also a difference between surveillance analyses of “utilization” alone versus analyses that are intended to include variables reflecting the factors involved in personal decision-making to obtain screening. The assessment of these potential correlates presents additional challenges, both because of the need for extra questions and because many measures used in the behavioral and social sciences are lengthy. For population-level surveys, therefore, it is important to have indicators of psychosocial constructs that use a limited but efficient set of items, sometimes called “short forms.” One benefit of a population-level survey is that its representative sample helps to convey confidence that a statistically significant association between a variable and cancer screening has broad generalizability. The 1990 NHIS Disease Prevention and Health Promotion Supplement, as well as the 1992 NHIS Cancer Control Supplement, included extra questions—beyond those normally asked—to assess covariates of screening. 18–20 These datasets are now rather old, however, relative to trends in the diffusion and adoption of mammography since the early 1990s. Population-level surveys should periodically include questions that assess key behavioral and social science variables, such as perceived risk, intention for future screening, social network support, satisfaction with provider interaction, life stressors, mood, affect, and information-seeking. Follow-up surveys in local samples, perhaps even purposely drawn to investigate specific hypotheses, can then employ longer scales to examine the constructs in more refined ways.
An impetus for tracking cancer screening practices in population-level surveys is the inclusion of screening in national objectives and priorities for disease prevention/health promotion, such as the series of “Healthy People” initiatives and the “challenge goals” of the American Cancer Society (ACS) for morbidity and mortality reduction by 2015. 21–24 This inclusion is a significant step, because all objectives require the use of one or more population-level databases to monitor progress and make projections. Cervical, breast, and colorectal screening have been among the objectives for Healthy People 2000 and 2010, as well as for the ACS challenge goals. However, to our knowledge to date, the objectives have been to achieve target percentages of recent screening. Therefore, there has been no mandate to have questions regarding repeat screening included in population-level datasets such as the NHIS or the BRFSS. In the absence of such data, behavioral and social science research is hindered both in targeting groups for repeat screening interventions and in identifying factors that influence repeat screening rates. Consequently, a next important step is for repeat screening to be incorporated into future Healthy People agendas, beginning with the objectives for 2020. This inclusion would also prompt individual investigators to add indicators of repeat screening to their own surveys.
Another marker of the acceptance of a cancer screening procedure is the procedure becoming a service covered by health insurance. 25–28 Research on barriers to obtaining screening has repeatedly shown that insurance coverage and ability to pay are important factors. Perhaps an even stronger indicator of integration into the health care industry is when a screening procedure becomes a quality-of-care indicator, such as for the Health Plan Employer Data Information Set (HEDIS) used as accreditation guidelines for health maintenance organizations (available from URL: http://www.ncqa.org/Programs/HEDIS [accessed July 15, 2004]). Insurance coverage and inclusion as a quality indicator also have implications for research on understanding and promoting screening use. Individuals and providers ask why they should obtain or recommend a screening test. Although data on morbidity/mortality reduction from clinical trials can be used as a justification, it is also a very practical fact to know that the procedure is covered by insurance and that the procedure is one of the ways that insurance plans are ranked for quality of health care. At the level of persons and providers, insurance coverage does not guarantee a rapid uptake of screening use, but status as a covered service contributes to people's and providers' decision-making and thereby influences behavioral and social science research on cancer screening. Similarly, but at another level, persons responsible for planning population-level surveys need a basis on which to include content areas and their specific items. Again, results of clinical trials are important, but a cancer screening test's status as a covered service and quality indicator is an additional rationale.
In summary, this lesson stresses the importance of what may at first seem like a small thing—having depth of content on at least one recognized and respected population-level data set. In practice, however, this requires deliberate efforts to include the necessary questions as well as rationales for that inclusion.
Lesson 3: The Priorities of Behavioral and Social Science Research in Cancer Screening are Closely Tied to the Identification of Groups with Low Utilization of Screening, as Defined by Sociodemographic Variables and Psychosocial Factors
For all current screening tests, two priorities have been to track the prevalence of cancer screening across subgroups of the population and to identify groups at risk of lower utilization. An emphasis on those at risk of underutilization and being underserved will very likely also characterize research on future screening modalities.
Identifying barriers, facilitators, correlates, predictors, and determinants of being and not being screened
Most likely, the most ubiquitous literature concerning cancer screening are studies conducted under the rubrics of identifying the “barriers and facilitators,” “correlates,” “predictors,” “risk factors,” and “determinants” of being or not being screened. 29–45 In a way, both identifying these factors and dealing with them in interventions are at the heart of research on cancer screening that takes a behavioral and social science perspective. Some variables, often those that are sociodemographic in nature, are not directly modifiable but are the basis for selecting population groups that should be the focus of interventions. Other variables can be modified and become the direct targets of intervention materials and activities. The subsequent translation of these variables into causal chains that can guide intervention development is an ultimate task facing behavioral and social science.
In the creation of causal chains, there are important differences among terms denoting a barrier, a facilitator, a correlate, a predictor, and a determinant. Based on how they are used in the literature, the words barrier and facilitator most likely are the most flexible of the above terms, but therefore also harder to classify precisely with regard to their implications. Sometimes discussions of barriers or facilitators are presented solely in univariate frequency tables that are based on self-reports of the factors that deter people from being screened or influence them to be screened. In this format, even an n = 1 for a barrier might be considered important for denoting a person's at-risk status for not being screened. At other times, status as a barrier or facilitator results from multivariable analyses that adjust for other covariates. In these instances, at-risk status is conveyed probabilistically on the basis of group membership. Barriers and facilitators are often examined from the standpoint of having or not having tangible resources (e.g., income, available transportation, health insurance, regular source of care). It is also possible, however, to conceptualize even a psychosocial variable (e.g., self-efficacy, depression and affect, social norms, pros and cons) in the language of being a barrier or facilitator.
Important nuances exist, however, in the terminology used to identify variables that place persons at risk of lower utilization, even when multivariable analyses are used. Research design is closely related to these differences and to the translation of results into causal chains for intervention planning. Cross-sectional, observational studies identify barriers and facilitators in the form of correlates. Because correlates are assessed concurrently with the report of screening status, and the screening practice is assessed as occurring or not occurring in a time interval before the cross-sectional assessment (e.g., 2 years), the result is the well known potential confound between cause and effect.
Longitudinal, prospective studies are widely viewed as being preferable to cross-sectional studies because they identify barriers and facilitators at the level of being predictors of screening behavior. Even so, the ability of a variable to predict a screening practice in longitudinal analyses does not fully confirm that variable's exact place in the causal chain leading to the behavior. In observational, prospective research there is still the possibility of a third-variable association, commonly referred to as confounders, a concern that also can affect cross-sectional studies. Determinants are therefore the ideal variables to identify, because a determinant is a type of predictor that has been shown to be in the direct causal chain of influence on the health practice and to have the least likelihood of confound by a third variable. Of course, status as a determinant can be very difficult to confirm, because doing so may involve having to conduct a manipulation of the variable to demonstrate its causal association on the health practice or on a mediator variable. Finally, some studies are presented as reporting risk factors for not being screened. Depending on the design of a study, a variable labeled as a risk factor may be a correlate, a predictor, or a determinant.
A challenge for cancer screening is to place variables with this variety of designations into a causal chain that indicates priorities among the variables, in regard to their hypothesized distal versus proximate influence on screening decision-making and behavior. A corollary lesson learned in this process is the additional challenge of integrating results from what are very often cross-sectional surveys, with different sets of variables included in the analyses. A key consideration is the range of variables that are included in studies, because each variable denotes a characteristic that may be associated with underutilization. We need to be able to explain and influence enough variance of a health practice to make a difference, even if not all variance can be explained and not all barriers or facilitators can be influenced. Drawing variables from any one domain of assessment (e.g., personal attitudes and beliefs, family and social network, sociodemographics, access variables) will very likely explain only a limited percent of variance and, therefore, not optimally inform intervention. The development of hierarchical methods of statistical analysis now allows more sophisticated examination of factors that influence screening utilization.
Adopting the objective of eliminating disparities in utilization
A second prominent perspective in behavioral and social science literature on underserved populations has been the objective of redressing inequalities of access and disparities in utilization. This theme is related to Lesson 1 and to the search for barriers to screening. Sociodemographic variables (such as age, race or ethnicity, income, education, regular source of care, and insurance status) commonly have been used to define at-risk population subgroups. Many other areas of health care use and health status show disparities along these fundamental personal characteristics. Thus, it is natural that behavioral and social science research on cancer screening has also been inextricably connected to the objective that underserved populations should have access to health care. In fact, behavioral and social science research on cancer screening might be viewed as having an inherently applied orientation. An argument could be made that there is no reason to determine the factors that influence screening decision-making and use unless there is an intention to translate that information into interventions. There is every reason to expect that future screening modalities will have the same applied emphasis that has occurred for breast, cervical, and colorectal cancer screening. For example, a major impetus behind prostate cancer research and intervention is the disproportionate disease burden among African-American men. 8
Lesson 4: Monitoring and Responding to the Diversification of Personal Screening Histories are Needed
Three basic indicators of screening typically used as dependent or outcome variables are ever having the examination, timing of the most recent examination, and obtaining repeat on-schedule examination. 46 At first, each of these indicators seems straightforward. The longer a screening technology is available, however, the more any one individual can develop a history with the screening procedure, and individuals considered at a group level have more opportunities to develop a variety of experiences with the procedure. Personal screening histories result from a combination of several variables, including ever having the examination, timing of the most recent examination, total number of examinations over a time period, usual interval between examinations, the reason for each examination, and the need for follow-up procedures. Intention for future screening can be added as an additional dimension, because intention (or absence of it) has been shown to be a predictor of obtaining or not obtaining future screening.47–49
The potential for diversity among personal screening histories is evident by a straightforward computational example. Assume a cohort of women who are turning age 40 years and have never undergone a mammogram. When they are age 44 years, an initial survey is conducted, with three possible baseline statuses: never had the screening procedure, most recent screening is on schedule, and most recent screening is off schedule. Also assume four follow-up surveillance surveys over a period of years, each with the possible classifications of being on schedule or off schedule (including remaining status of never had) when assessed. These options yield a total of 48 possible patterns (i.e., 3 × 24), not taking into account permutations caused by possible callbacks, false-positive results, or unsatisfactory experiences with the examination setting and/or staff. It is important not to overstate the complexity. Not all possible patterns are necessarily so distinct that they require being studied separately. Assessing repeat screening does, however, present more options and complications than assessing and analyzing ever had status and recent examinations.
A prospective study of repeat screening, whether observational or intervention, can last 6–8 years from start to finish if the criterion is 3 on-schedule, every-other-year examinations. The challenge is even greater as the time interval increases to > 2 years, such as for sigmoidoscopy and colonoscopy. Clinical trials to establish the effectiveness of a screening procedure for mortality reduction routinely have a ≥ 8-year follow-up period, and tracking often continues well beyond the original time frame. For behavioral and social science research, how easily can studies lasting 6–8 years be justified to monitor naturalistic trends in screening adoption or to test strategies for achieving high levels of repeat screening? The lesson learned is that monitoring repeat screening and personal histories takes time, but the core issue is how behavioral and social science research can accomplish the sort of extended data collection and research designs that are necessary.
A potential advantage for future screening research is the ability to link databases. For example, psychosocial interview data at the beginning of a longitudinal survey could be linked with screening history data across the clinical settings that participants used. In the U.S., the recently enacted privacy policies under the Health Insurance Privacy and Accountability Act will have to be satisfied for many of these studies to be done. A premium will likely be placed on anticipating at the outset of a study the full scope of research data to be used in a project, as well as potential future uses or extensions of the data. If that challenge can be met, and necessary consents can be obtained, the ability to examine the use of a screening test in broader contexts than those studied to date should be greatly enhanced.
Lesson 5: Behavioral and Social Science Researchers Should be Involved with the Evaluation of the Use of New Technologies
Lesson 2 might be interpreted to suggest a sequential process by which a possible technology is first validated and then incorporated into surveillance databases. That type of orderly process is not necessarily the case. Tests and procedures can be, and have been, made available before general consensus on their benefits has been reached. Examples include ovarian cancer screening, prostate-specific antigen testing for prostate cancer, and spiral computed tomography (CT) scans for lung cancer. Mammography for women ages 40–49 years has also been a topic of debate. Even Pap testing, mammography, fecal occult blood testing, sigmoidoscopy, and colonoscopy were considered new technologies at some point in their histories. A strict use of terminology, therefore, might distinguish between “a technology being used for testing,” and an accepted evidence and consensus-based “screening technology” that follows the principles discussed in more detail by Meissner et al. 1 in the current issue and by other reports in the literature.50
The broader question is when and how behavioral and social science should become involved with the technology early in its extension into the health care sector. Too often, research begins after the technology is established. For several reasons, behavioral and social science research should begin early in the use of a new technology. First, new technologies provide a context for studying personal and health professionals' decision-making before clinical trials and other empirical evidence are considered definitive. 51 Primary care providers can benefit from knowing the attitudes and concerns of patients when a procedure does not yet have a consensus of support. More than one clinical trial is necessary to arrive at a consensus of evidence. Almost always, therefore, a procedure is available for an extended time but is still considered to be under investigation. Early involvement in the development of a technology may identify, from user and provider perspectives, problems that can be corrected if necessary or incorporated into plans to facilitate dissemination. It may even identify unexpected positives or benefits that can be used to promote utilization.
Second, the early phase of the adoption of a procedure is an opportunity to study its use in the context of what most likely is a limited distribution of facilities, modest diffusion of information among providers and the general public, sporadic insurance coverage, and a limited history of participants' experiences with the technology. The context of personal and providers' decision-making changes as the number of facilities increases, insurance coverage expands, the procedure becomes a public health priority, providers become familiar with the procedure, and the procedure becomes more common among one's family and social network.
Third, even with more than one clinical trial, uncertainty is possible regarding when a procedure has been sufficiently justified empirically. Controversies may exist concerning the interpretation of clinical trials data. 52, 53 As noted in this special issue,1 and reviewed in detail in another publication,50 the seminal empirical data justifying the procedure are likely to be reexamined periodically.54–58 In addition, discrepancies may exist regarding the guidelines of different professional organizations (e.g., age ranges, periodicity between examinations), even if the clinical trials data support the procedure. The debate concerning mammography for women ages 40–49 years is a prime example.59–63 Behavioral and social science cannot wait for virtually unanimous consensus to become involved in the study of a procedure, nor can researchers anticipate when reexaminations of data may occur or what the conclusions may be.
Along with reexaminations of clinical trials data, discrepancies in guidelines endorsed by professional and governmental entities can be confusing to both individuals and their health care providers. An informed decision-making perspective in such circumstances is a reasonable strategy for behavioral and social science research to adopt. 64 In addition, it is possible that evidence from clinical trials will not provide guidance for every circumstance that could be posed. For example, setting guidelines for mammography among women age > 75 years has posed a challenge because none of the seminal trials had women in this age group at baseline.65
Lesson 6: Behavioral and Social Science Research on Cancer Screening has Become a Field of Study with its Own Particular Features
Cancer screening is not a single discipline, but a domain of health care and related research, a public health objective, a focus of health policy, and an area of advocacy. As such, it is often characterized by utilization targets that are pursued by the initiatives of several disciplines. In addition, behavioral and social science research in cancer screening is still relatively new compared with many traditionally recognized disciplines.
It may be helpful to consider why investigators are drawn into the field and desire to establish a long-term program of cancer screening research. The interval nature of screening creates a context with different complexities than those found in many daily health practices, such as smoking cessation, alcohol and illegal substance use, physical activity/exercise, diet, weight loss, safe sex, human immunodeficiency virus infection/acquired immunodeficiency syndrome risk reduction, and even wearing seat belts. The interval nature of screening presents a challenge of knowing where deliberations about the behavior “fit” in a person's life between examinations, and how decision-making for screening occurs in comparison to that for more frequent practices. Interaction with health professionals, the health care system, and medical technology is required, which also adds dimensions that differ from many daily health practices. In addition, cancer screening deals with the situations of false-positive results and follow-up after abnormal results. 66 To better compare screening with other health-related behaviors,67, 68 it is important to examine the ways health practices differ in the “performance demands” placed on people for successful, sustained implementation.
Cancer screening sometimes also has indefinite population guidelines and varying recommendations with regard to intervals between examinations. 69–72 Cancer screening has to allow for “shades of gray” that are part of the contexts within which persons and their health care providers make decisions. For example, emphasis is growing in cancer screening research on the topic of informed decision-making—a different type of dependent variable because it does not require obtaining the screening test.64
Other areas that attract interest in cancer screening research include tailored interventions, risk perceptions, health communications, message framing, cultural relevance, first-degree relatives, lay health advisors, and telephone counseling. These areas are not unique to cancer screening, nor do they have to be. They have become important topics for screening interventions because they reflect the attention being given to the process of implementing a screening practice, along with increasingly sophisticated research designs, setting-specific intervention strategies, and population groups. Behavioral and social science research on cancer screening very likely will continue to have a dual emphasis. First, until screening rates equalize across all groups in the population, emphasis will continue on themes of redressing disparities, access to services, and interventions that help to equalize rates of use. Second, research also will be directed to the processes of obtaining cancer screening, as a health practice per se, and to better understanding the nature and determinants of screening compared with other health habits, each of which has its own characteristics. 73–77 One aspect of this lesson learned, therefore, is that applied cancer screening should be a venue for the synergistic integration of these two domains of research.
Lesson 7: Multilevel Theoretical and Conceptual Approaches are Needed to Understand the Full Context of Cancer Screening
Nearly 10 years ago, Curry and Emmons 78 reviewed the theories and models that guided cancer screening research. They emphasized the need for integrated frameworks that include variables from several domains or levels. That priority has been emphasized by others4, 79, 80 and is still important. The challenge is not so much in identifying and assessing individual variables because several theories, models, and variables can be drawn upon,80–92 and two articles in this supplement provide multilevel perspectives.66, 93 Behavioral and social scientists are quite good at defining and measuring new variables. The attention directed to multilevel or ecologic approaches also promises to expand the pool of screening-related variables even further.94, 95 The larger challenge now is to assemble variables into causal chains that reflect the complex context of a specific behavior change. However, there is no step-by-step guide for applying an existing theory to a new health behavior. A previous study discussed the process of adapting the transtheoretical model to mammography,96 but even that commentary could not address all situations and was not a multilevel integration.
Intervention can be viewed as a situation that requires a matching of resources. These resources (broadly defined to include material, financial, attitudinal, and social) are those needed to change and sustain a health practice, compared with the resources possessed by the individual. Before the intervention, therefore, the resources needed to achieve the behavior change should be assessed. Before an intervention is initiated, a complementary assessment should be made of the resources available to an individual relative to the resources needed to achieve the behavioral objective. Intervention should then provide elements that fill any observed gaps. The individualization of tailored interventions, a prime example of resource matching, still requires knowing which variables should be tailored. Ideally, theory informs the selection of potential covariates that comprise the surveys and the subsequent analyses used to identify the factors that influence the decision to obtain screening. It therefore is necessary to have theories and models that sufficiently address the full context of the health practice. In addition, as noted by Curry and Emmons, 78 differences may exist between theories that explain current behavioral status (i.e., a cross-sectional perspective) and theories that are useful for explaining behavior change (i.e., a prospective, longitudinal perspective).
A task facing behavioral and social science research on cancer screening is defining the context of a study—whether it is a study to identify correlates, predictors, and determinants or an intervention project to change a screening practice. That is, where does a particular project fit into the larger mosaic of research relevant to improving screening rates? The concept of a “focal point” has recently been proposed as one way to approach defining the context of an intervention or a survey study to identify the correlates or predictors of a health practice. 67, 68 A focal point is defined by the simultaneous combination of the target population, the health practice, the intervention setting, and the eventual setting for implementing the behavior as a regular practice in daily life. This context-based perspective of an intervention, integrated with the view of intervention as a process of resource matching that is as individually specific as possible, leads to the conclusion that individuals must be assessed not only on traditional demographic and psychosocial variables (e.g., self-efficacy, health beliefs, social support, regular source of health care, insurance coverage) but also on any characteristics of the health practice (“performance demands”) as well as the settings (“setting demands”) that must be met to conduct the practice (e.g., ability to delay gratification or tolerate discomfort, self-monitoring, time constraints, peer pressure, or availability of cues for action).
The focal point concept is amenable to multilevel approaches because it places priority on identifying the relevant variables for explaining and changing behavior in a particular context. A focal point perspective is not a theory. Rather, it is a framework to help organize the search for variables that should go into the causal chain for the particular context defined by the researcher or interventionist, consistent with Lesson 3. A focal point is intended to be context specific and to allow the accommodation of a diverse set of variables. Research and theory development are constrained by the statistical methods that exist to analyze data and investigate hypotheses. Over the past few years, the rapid development of geographic and spatial analysis, as well as statistical techniques, has allowed increasingly sophisticated hierarchical, nested, or clustered analyses. 97, 98 These advances have allowed analyses that were not possible in the 1980s and 1990s. For such analyses to be conducted, however, it is also necessary that variables across multiple levels be included in databases that track the adoption of screening modalities. A primary asset of a multilevel approach is the attention it draws to the overall context of a health practice or health outcome. Multilevel approaches are well suited for application in specifically defined contexts and offer great promise for the central purpose of behavioral and social science applications to cancer screening—not only explaining the variation in personal health practices but also influencing that variation.97, 98
The number and diversity of variables encompassed by multilevel models present a challenge for intervention. The potential interactions of factors across levels are much greater than the already substantial combinations that can exist within a single domain (e.g., individual-level attitudes and beliefs, social network influences). Individually tailored interventions became more feasible in the 1990s with the development of computer technologies that allowed standardized management of large databases, processing of multiple variables, and rapid turnaround from input data to the production of the tailored materials or messages. 99 From a historical perspective, tailored interventions are still in an early phase of development and will certainly continue to be tested in behavioral and social science research related to any new screening technology. This comment does not assume that tailored interventions will always be superior to other strategies. Rather, it simply suggests that computer-assisted tailoring is one strategy to address the complexities inherent in multilevel perspectives on behavior and behavior change.
EMERGING AREAS OF BEHAVIORAL AND SOCIAL SCIENCE
This section offers ideas about areas of cancer control that are currently developing and will be a base of information from which to learn additional lessons.
It is Necessary to Determine the Best Questions to Ask to Track Repeat Screening and Personal Screening Histories
Surveillance studies should monitor the full range of indicators of screening (i.e., including questions to assess regular/repeat screening, not only ever had and recent screening), as well as variables that reflect personal experiences with previous screening, insofar as those experiences may influence future screenings (e.g., false-positive results, painful examination, follow-up after an abnormal screen). We currently have little experience with the nuances of asking women and men about their long-term experiences with screening technologies. For example, is asking for the number of examinations in the past “X” years a good strategy in retrospective surveys? Is it preferable to ask separately about each examination in a timeline follow-back procedure? How many screening intervals should be covered to establish a person's pattern? Publicly sponsored surveys (e.g., the NHIS, the BRFSS, and the HINTS) should not be expected to bear the entire burden of being information resources, but they should include some questions to assess repeat screening and to allow identifying personal histories. State health surveys are an additional resource.
Few data are available on patterns of screening within a cohort followed prospectively. Current data are predominantly cross-sectional. Following people prospectively, especially over multiple screening intervals, introduces the consideration of individuals moving out of, and perhaps back into, calculations of percentages, depending on completion of follow-ups and on how the reason for having an examination is treated. If persons who have an examination because of a possible problem are removed from the denominator because of a focus solely on screening, do they become eligible again at another follow-up if their examination revealed no problem? It will be important for investigators to state clearly their inclusion/exclusion criteria for each calculation.
Strategies Should be Investigated to Combine Individual Correlates and Predictors to Inform Identifying Intervention Populations and Selecting Intervention Strategies
Standard data analysis methods, such as linear and logistic regression, are generally reported in a straightforward main-effects manner and yield statistically independent correlates and predictors of cancer control practices. It is unlikely, however, that these variables actually do operate independently in the day-to-day context of the factors that influence the receipt of a screening test.
The literature appears to offer little guidance concerning how to take a set of individual variables and combine them into a causal chain that reflects the context of the health practice under investigation. This consideration is beyond the distinctions of terminology discussed in Lesson 3. Contributing to the challenge are several factors including the typical main-effects analysis model that is used, the lack of variables in population-level datasets that assess decision-making processes, the difficulty of testing fully interacted models in surveys with relatively small sample sizes, and the absence of prospective naturalistic data to identify predictors and determinants of screening utilization. The translation of results from correlate and predictor studies into interventions can benefit from more integration with approaches developed specifically for intervention planning, such as the precede–proceed model 100, 101 and the intervention mapping technique.102
It Will be Important to Understand Better the Dynamics of Cancer Screening among Men
Pap testing and breast cancer screening, the pioneer modalities, are domains of women's health. By extension, therefore, most intervention-related research has been performed with women. Colorectal cancer screening affects both men and women, and prostate cancer is specifically a men's health issue. Prostate cancer research so far has emphasized informed decision-making rather than interventions that actively promote utilization. The conventional view is that men are somewhat more likely than women to avoid seeking health care. However, this generalization is clearly the tip of a substantially larger issue, and cancer screening by men is a topic worthy of separate study. It is important to build a knowledge base regarding men's cancer screening, and doing so will require a deliberate effort, 103 with as much attention given to identifying barriers and facilitators as has been given to breast and cervical screening.
Attention Should be Given to Processes and Events that Occur between Examinations and that can Influence Obtaining Screening
Because screening is obtained at intervals, gaps of time occur during which various factors can work for or against obtaining the examination due next. Lesson 6 noted that this is a unique feature of screening practices. However, the literature regarding cancer screening predominantly has adopted only an interval perspective and has treated screening as an event that does or does not happen as a discrete occurrence each time. To our knowledge, little insight is available regarding the presence of cancer screening tests in people's day-to-day thinking. Questions still must be answered about how screening tests are considered or not considered between due dates. For example, do persons have “decision windows,” during which they decide to schedule or not schedule a screening? If they do, do they or do they not revisit the question until a later time? Another important question is how do persons who are past due for screening or who have never had an examination decide to have the procedure?
It is Important to Conduct Process and Process-to-Outcome Evaluations
It is important to monitor how well interventions are implemented and to examine how program implementation is associated with achieving outcomes. 104 It is also important to identify whether any aspects of an intervention were more influential than others in producing an effect on use of screening. Therefore, “fidelity” and “dose” of an intervention should be analyzed by process and process-to-outcome evaluation. However, process evaluation indicators have not been as prominent as might be desired in screening intervention studies. Even when relevant process data have been collected, however, there is a subtle barrier to conducting these analyses. When an intervention does have a significant effect, process and process-to-outcome evaluation can be expected to find some associations in the data. In effect, because an overall benefit was found, the task is to determine which subgroups of persons benefited the most. However, when the overall results are not significant, process-related analyses may be less likely to occur because of a lesser prospect of finding useful results or a concern that null results would not be publishable.
Behavioral and Social Science Research Has to Deal with the Development of New Screening Technologies for Existing Modalities
The introduction of each new screening technology signals the beginning of a search to develop the next generation of methodologies that will have one or more of the following characteristics: less invasive, more acceptable to the target populations, less prone to side effects, more cost-effective, and/or more able to detect pathologies at earlier stages when they are more curable. 105, 106 For example, early breast cancer screening used xerography, thermography, and even traditional chest X-ray machines, before the introduction of dedicated mammography units. Colorectal screening has progressed from digital rectal examination to double-contrast barium enema, fecal occult blood testing, sigmoidoscopy, and colonoscopy. Attempts continue to refine these screening modalities, resulting in new techniques such as ultrasound, digital mammography, and magnetic resonance imaging for breast cancer,107 and CT colography (virtual colonoscopy), among others.108, 109
Some refinements in technology are relatively invisible to the general population and therefore do not affect the typical questions used in surveys to monitor trends in cancer screening. For example, using different methods of Pap smear preparation and automated reading of cervical smears, rather than visual examination, are refinements at the level of laboratory processing 110 and do not affect questions used to assess receipt of the test per se. Such refinements most likely have few ramifications for studies concerning the factors that affect receipt of screening.
Other changes to technology may affect the target population, health care providers, and behavioral and social science research. If new technologies or procedures are more sensitive, they can increase the risk of false-positive results or even of identifying abnormalities that are so early in development there is no certainty that they will become cancerous. There are two possible implications for research. First, public awareness of these points has implications in the area of informed decision-making (e.g., the “watchful waiting” option for prostate cancer). Second, a larger number of available procedures complicates assessment by requiring more questions in surveys and may also complicate interventions because more options have to be addressed when planning messages and strategies.
It is important to recall that when Pap testing and mammography were developed before the explosion of biotechnology, even a single procedure was a significant step forward. That previous context was deceptively simple (in a relative sense, in any case) compared with the procedures currently being developed and others that will be explored for new screening modalities. To achieve the objectives of reducing mortality and morbidity, the option a provider recommends or a person selects may not matter. However, for behavioral and social science research concerning decision-making and factors that account for screening use, more options translate into more complexity.
The current article has discussed several lessons learned from the literature regarding cancer screening. The future of cancer screening research has a solid foundation. The seminal years of research to date have produced a large group of investigators who have expertise in different screening modalities, special populations, particular settings, and varying theoretical perspectives. We have also witnessed the progressive diversification of attention to specific population groups, such as first-degree relatives, persons with genetically based risk, and cancer survivors. Many more training programs in cancer control exist now than in the mid-1980s and early 1990s. Over the past 20–30 years, therefore, the foundation has been laid for a cadre of behavioral and social science investigators to work with future screening procedures and population groups.