Needing More to Understand Breast Cancer Screening Adherence: Multilevel Analysis

Recommendations for breast cancer screening date back to the 1970s and early 1980s, yet nonadherence to those recommendations continues to be a challenge as new women age into the screening age group or already eligible women are not screened as recommended.^1,2 Although the overall impact of screening on breast cancer mortality is less than hoped, it is clear that breast cancer mortality rates are falling, and analysis suggests that screening mammography accounts for half of the decline.^3,4 The U.S. Preventive Services Task Force recommends mammography every 2 years for women aged 50–74 years.² Examination of cohorts of women in Canada regularly offered screening suggests that only ∼31% are continually adherent over time.⁵ In the United States, where regular referral to screening is not guaranteed, adherence may be less.

Concern about nonadherence has fostered hundreds of studies of factors that affect women's choices to seek mammography on their own or when it is offered by a physician. Most of these studies focus on cognitive factors such as women's beliefs about the outcomes and benefits of screening, their intentions, and their perceptions of what others want them to do.⁶ In this issue of Journal of Women's Health, Beaber et al.⁷ again look at individual factors but raise the bar on adherence studies and our conceptualization of the problem by also looking at factors associated with the screening facility.

Beaber et al.⁷ note that 15.5% of women seeking care are nonadherent to a plan of getting their next mammogram within 2 years of a negative examination. The women sought care through a large network of screening facilities where a large variation (10%–26.5%) in subsequent screens exists. The network is open, so people can come and go. Therefore, the reported rate of nonadherence over time may be an overestimation of nonadherence if some of the women sought mammograms elsewhere. Despite this limitation, they showed that non-Hispanic Asian/Pacific Islanders, heavier women, and women aged 50–59 years were more likely to be nonadherent to getting a subsequent mammogram within 2 years.

Beaber et al.⁷ cannot explain the screening variation with facility characteristics they measured and acknowledge that their measurement was limited. Although it is not clear that all women who did not return to the facility were not screened, they are raising an important point that women level characteristics alone are not sufficient to explain the variation in cancer screening among centers. To further contribute to mortality reductions, one question becomes what other factors need to be examined to measure, explain, and encourage healthy behavior. Beaber et al.⁷ suggest considering facility, practice, and health system characteristics, but what exactly are those?

Their suggestion is consistent with a push to go beyond the individual to consider how context affects behavior.^8,9 Burke et al.,⁸ Pasick and Taplin,⁹ and Taplin et al.¹⁰ have raised the issue of measuring context for many years and recognize both sociocultural and ecological considerations. The former includes consideration of the multiple dimensions of the social and cultural phenomena of daily life including history, politics, family, organizations, and community.⁹ Considering measures of family, organizations, and community encourages thinking consistent with ecological models.¹⁰

These models consider how family dynamics, social networks, health care provider team roles, culture, and organizational structure may interact and influence individual behavior.¹⁰ They have the advantage of separating the way we can think about interventions since working with individuals, provider teams, and organizations require different approaches.¹⁰

But the theory behind these contextual and sociocultural models is a patchwork of concepts because no current theory is completely explanatory. Damschroder et al.'s¹¹ consolidated framework for implementation has made an important step forward and can guide practitioners, but it is more of a heuristic than a theoretical model. She suggests that work to date relevant to implementation can be summarized in five domains (intervention characteristics, outer setting, inner setting, characteristics of the individual, and the process of implementation).¹¹

Ecological models are also thinly affiliated with theory.¹² These models suggest that there are interactions between levels of increasing human aggregation (individuals, providers, organizations, communities, states, and nations). Although this is an intuitively appealing way to think about behavior, it is not specified how “levels” of a multilevel model operate to affect each other and individuals.⁸ To build interventions and promote behavior consistent with health, we must measure and intervene upon the right variables. Without more complete and tested theory, it cannot be clear whether Beaber et al.⁷ measured the right variables, much less whether they were measured in the right way.

Beaber et al.⁷ could only study individual and facility factors they had available in their observational database: age, median household income of the woman's zip code, race/ethnicity, and body mass index. Missing are the individual factors such as intention, beliefs, and perceptions of what others want (social norm).⁶ This study also notes variation by radiology facility that persists after accounting for the measured individual characteristics. They measured communication practices of the facilities (sending reminders to women and/or their providers) and did not have other characteristics. Whether their model would explain more variance in screening behavior if it included a broader set of individual and contextual characteristics cannot be determined here. Instead this study demonstrates that we need more inclusive conceptualization and measurement to explain behavior.

The challenge is that inclusive conceptualizations are complicated and thinly grounded in theory.^8,12 Behavioral scientists have been struggling to explain how social context influences behavior for years.¹³ Most studies focus on the individual level characteristics and have limited, if any, contextual measures.⁸ The result is there is much less known about which contextual factors affect screening adherence.^6,8,12 This is important because contextual issues such as health care system culture, insurance incentive programs for prevention, and physical environment are factors that can be influenced to achieve healthy behaviors. To decide which of these to focus upon first, we need more inclusive theories and much better measures of salient contextual influences.

For researchers, health care practitioners, and designers of health care systems, the need to develop theory, specify constructs, and clarify measures is humbling and hopefully exciting. Although work is underway to develop multilevel interventions, we have a long way to go to make evidence-based decisions that guide the design of our health care context in a way that makes it easy to achieve healthy behaviors.^14,15 To do that work, Beaber et al.⁷ show us we need to improve our theories and collect a richer set of individual and contextual factors that help us explain behavior.