Two-thirds of all annual deaths in Canada are caused by 4 noncommunicable chronic diseases: cancer, cardiovascular disease, diabetes, and chronic respiratory disease [1]. Approximately 12% of Canadians 65 years or older have lived with 2 or more of these chronic diseases, also known as multimorbidity [2]. Health behaviors such as physical inactivity, unhealthy eating, smoking tobacco, and alcohol use are leading risk factors for chronic diseases and play a central role in health status and quality of life [3]. The prevalence of risky health behaviors is high, with approximately 4 in 5 Canadian adults reporting at least one of these modifiable risk factors for noncommunicable chronic diseases [2].

Although the risk factors and consequences of multimorbidity have been studied extensively [4-6], much less attention has been paid to understanding how different combinations of multiple behaviors influence individuals’ life satisfaction and health. Our everyday lives are defined by multiple co-occurring health, social, family, personal, and work-related behaviors, each vying for the limited time, energy, and motivation available [7]. In spite of this, health behaviors tend to be studied and promoted largely separately from each other, resulting in guidelines and interventions for healthy aging siloed by specific behaviors. For example, Canada has distinct sets of guidelines for physical activity and sleep [8] and alcohol consumption [9]. In addition to focusing on single health behaviors, guidelines often do not consider the interconnectedness—or co-occurrence—of the various other family, social, work, hobby, and other behaviors that characterize daily life. Understanding how behaviors cluster and how such clusters are associated with physical and mental health, life satisfaction, and health care utilization may provide new opportunities to leverage this co-occurrence to promote multiple health behavior change interventions tailored to which behaviors co-occur for whom, thus better reflecting the real-world complexity of health care for aging Canadians.

International population data support the co-occurrence and clustering of health behaviors. For instance, Irish national data collected in the 2007 National Survey of Lifestyle, Attitudes, and Nutrition investigated the co-occurrence of smoking, alcohol use, physical inactivity, and unhealthy eating in adults aged 18 years and older [10]. In this cross-sectional analysis, the authors identified 6 clusters of behaviors to describe the population, which were labeled as (1) healthy lifestyle (characterized by people who had never smoked, high physical activity, highest healthy eating, and moderate alcohol use), (2) former smokers (former smokers who reported high physical activity, moderate alcohol use, and healthy eating), (3) temperate (moderately active and moderate drinkers who had never smoked), (4) physically inactive (people with low levels of physical activity, poor eating habits, and who reporting some smoking and high alcohol use), (5) mixed lifestyle (those who had never smoked and reported moderate physical activity and variable alcohol consumption), and (6) multiple risk factor (moderate physical activity, moderate to high alcohol use, and variable healthy eating). Although the cluster labels do not directly convey the co-occurrence of the behaviors, definitions of each label describe which behaviors co-occur most in each group in terms of the extent of the performance of each behavior. Beyond defining how behaviors cluster, they also showed that membership in each cluster was associated with demographic factors and health and well-being outcomes: those in the healthy lifestyle cluster were more likely to be women 65 years or older, be in the highest socioeconomic status (SES) group, and exhibit low psychological distress; those in the former smokers cluster were more likely than the healthy lifestyle cluster to be men and be in a lower SES group; the temperate cluster was associated with being male and being in a lower SES group as well as being younger; the physically inactive cluster was associated with being male aged 18 to 29 years and being in a lower SES group and having higher psychological distress; the multiple risk factor cluster had lowest reported energy and vitality, highest distress, lowest self-reported health, and lowest quality of life; and the mixed lifestyle cluster was associated with being male and younger and being in a lower SES group, with highest distress and lower energy and vitality. These findings show how the population can be segmented by the multiple health behaviors that characterize their lives and that these segmented clusters are socially patterned and associated with different health outcomes. It remains to be seen how such clusters may hold when focusing on adults aged 45 to 85 years and extending the behaviors under consideration beyond health behaviors. Identifying these clusters and their associations with demographic factors, general health, and health care utilization can inform the tailoring of future interventions targeting multiple behavior change.

In another study, Buck and Frosini [11] used data from the Health Survey for England to investigate the clustering of smoking, excessive alcohol use, unhealthy eating, and physical inactivity among adults aged 16 to 74 years between 2003 and 2008. They showed that in 2008, only 7% of people engaged in none of the behaviors and only 5% engaged in all 4; 63% of people engaged in 1 or 2, with the remaining 25% engaging in 3 or more behaviors. The most common co-occurring behaviors included either unhealthy eating and physical inactivity (more prevalent in women) or alcohol use, unhealthy eating, and inactivity (more prevalent in men). Men were more likely than women to have 3 co-occurring unhealthy behaviors, and those 65 years and older were more likely to have 2 co-occurring unhealthy behaviors than any other age group (16-24, 25-44, 45-64, and ≥65 years).

Although the importance of co-occurring health behaviors is clear, the role of nonhealth behaviors should not be overlooked. Using Australian population survey data collected in 2007 and 2009 in the Household Income and Labour Dynamics of Australia survey, we showed that other behaviors co-occurred with healthy eating and to some extent differed between men and women aged 18-65 years [12]. For Australian women, some family behaviors (caring for young children) and work behaviors (being employed in a managerial position) were associated with healthy eating, whereas not working was negatively associated with healthy eating. For men, the co-occurrence of work-related behaviors with healthy eating depended on the type of job category (positive for managers and negative for laborers). The findings highlight that nonhealth behaviors also co-occur with health behaviors and underscore the connectedness of the multiple behaviors in daily life.

The examples discussed so far have examined co-occurring health behaviors in the general population. However, some studies have focused on adults aged 50 years and older. For example, one study specifically focusing on older adults involved cross-sectional data collected in Germany in 2006, involving 982 men and 1020 women aged 50 to 70 years [13]. The authors investigated how 4 health behaviors (tobacco use, alcohol use, unhealthy eating, and physical inactivity) clustered in the population of adults in one German state and reliably grouped individuals by the unhealthy behaviors that they engaged in: 25% were defined by the lack of unhealthy behaviors, 21% as inactive, 18% as low fruit and vegetable eaters, 13% as smokers with other risk behaviors, and 23% as drinkers with other risk behaviors [13]. Although some adults were only described using one unhealthy behavior, over one-third had multiple unhealthy behaviors that clustered together. Furthermore, membership in each cluster was also patterned in terms of sociodemographic factors: those with multiple co-occurring unhealthy behaviors tended to be men, living alone, and of a lower SES. This study focused on a limited set of unhealthy behaviors, and it remains unclear which other behaviors characterizing daily life may also co-occur.

A different methodological approach was taken by Shaw and Agahi [14], who used baseline data from the 1998 Health and Retirement Study [15], which collected data from adults older than 50 years, to form health behavior profiles. In total, 12 health behavior profiles were constructed based on combinations of smokers versus nonsmokers, physically active versus inactive, and those who reported no versus moderate versus heavy alcohol consumption. Profiles varied widely in the percentage of participants captured in each profile. The 6 most prevalent profiles included the following: (1) physically inactive, nondrinkers, who do not smoke (6702/19,662, 34.1%); (2) physically active, nondrinkers, who do not smoke (4662/19,662, 23.7%); (3) physically active, moderate drinkers, who do not smoke (1986/19,662, 10.1%); (4) physically inactive, moderate drinkers, who do not smoke (1684/19,662, 8.6%); (5) physically inactive, nondrinkers, who smoke (1274/19,662, 6.5%); and (6) physically active, nondrinkers, who smoke (820/19,662, 4.2%).

Taken together, population survey data worldwide are converging on the idea that health behaviors cluster differently across the population and that the resulting clusters have distinct sociodemographic and health outcome patterns. However, to our knowledge, this has only been explored to a limited extent in Canadian data. Canadian National Population Survey data have shown that physical inactivity, alcohol use, and smoking co-occur [16,17]. Understanding how behaviors co-occur at a population level may provide novel ways of developing support and guidance for patients and clinicians and preventing noncommunicable chronic diseases in the general population using a multiple behavior approach.

To this end, we aim to leverage cross-sectional baseline data from the Canadian Longitudinal Study on Aging (CLSA) [18]. By 2031, 1 in 4 Canadians will be aged 65 years or older, and the CLSA aims to understand the determinants of health and wellness as people age [19]. For our purposes, we will use CLSA baseline data to address the following objectives: (1) describe how health behaviors cluster, (2) describe how other behaviors of everyday life (family, work, hobby, and community behaviors) cluster, (3) identify sociodemographic factors (sex, age group, marital status, income, country of birth, and social support availability) associated with cluster membership, (4) examine whether life satisfaction and health differ across clusters, (5) examine whether and which clusters of behavior are associated with health care utilization, and (6) examine whether clusters of health behaviors are associated with nonhealth behaviors.

Baseline data for the CLSA were collected from 51,338 participants aged between 45 and 85 years. Data were collected between 2010 and 2015. Secondary data analysis for this project was approved by the Ottawa Health Science Network Research Ethics Board (protocol ID #20190506-01H).

The scope, size, and rigor of the CLSA data set will provide us with an unprecedented opportunity to investigate how behaviors cluster. The findings will allow us to assess alignment with findings from other countries, while extending findings in novel ways by investigating how social, family, and work behaviors cluster alongside health behaviors typically investigated. Perhaps most importantly, this study will help to inform the development of novel health behavior change interventions tailored to subpopulations of adults defined by the behaviors that cluster within them. If behaviors co-occur, intervening on one may impact—or be impacted by—the others [62]. Interventions that target only one behavior may thus be undermined by the impact of a conflicting co-occurring behavior or miss an opportunity to leverage the enabling nature of a positively co-occurring behavior. In addition, targeting multiple co-occurring behaviors simultaneously has potential practical benefits, such as reduced expenses for intervention providers and reduced time commitments for those receiving the intervention. Understanding which behaviors co-occur and for whom is an important first step toward developing health behavior change interventions that address people’s actual challenges. In the context of a behavioral intervention development and testing framework [63], the proposed analysis will inform the foundational basic behavioral science—regarding patterns of co-occurring risky health behaviors and their associated outcomes—which precedes early phase behavioral trials.

This study will ultimately help to develop and evaluate more targeted interventions to support healthy aging and well-being in adulthood by identifying how clusters of co-occurring health behaviors are associated with sociodemographic factors, general health, and health care utilization. However, the proposed analyses are not without limitations. For example, many of the items selected for planned analysis are self-report, which have inherent strengths and weaknesses [64]. In addition, there are several points in the proposed analysis that require subjective decision making on behalf of the research team (eg, selecting variables for inclusion in models, data preprocessing decisions, selecting missing data procedures, and interpreting and selecting clustering solutions). We have sought to document these here to provide a clear explanation of decision processes, while acknowledging where and how researcher degrees of freedom are used to ensure transparency. Finally, we recognize that the analyses proposed are, to some extent, limited by their cross-sectional nature. Nevertheless, given the longitudinal nature of CLSA and planned future data releases, the proposed analyses have a number of novel implications and set the stage for planned future longitudinal analyses that extend the research questions to investigate changes in behavior clusters as a function of time both between and within individuals. Thus, this proposed study will establish the foundation for future analyses. More broadly, the methodological approaches proposed for this analysis lend themselves to replication in other similar data sets internationally, and we hope that the sharing of R code will help to enable this.

Finally, we emphasize that our choices of analytic methods are hypothesis generating, not hypothesis testing. However, if replicated, such findings may ultimately find their way into clinical practice guidance and public health guidance.