Introduction

The estimated 26 million cancer survivors and who will be living in the United States by 2040 (Tonorezos et al., 2024) are at increased risk for late effects of disease (i.e., cancer recurrence, secondary cancers, and chronic comorbidities) and early mortality (Leach et al., 2015; Sogaard et al., 2013). “Cancer survivors” includes everyone with a history of cancer from the time of diagnosis through end of life (Denlinger et al., 2014). Engaging in healthy lifestyle behaviors such as eating a high-quality diet, being physically active, and getting sufficient sleep have been independently associated with reduced all-cause and disease-specific mortality (del Pozo Cruz et al., 2020; Xiao et al., 2014). The American Cancer Society (ACS) has published nutrition and physical activity recommendations for adult survivors of cancer to promote overall health (Rock et al., 2022). These include consuming a plant-based, high-fiber diet with a variety of colorful fruits and vegetables and engaging in regular moderate-to-vigorous physical activity. Although cancer-specific sleep recommendations have not been released, the American Academy of Sleep Medicine and Sleep Research Society recommend that adults sleep seven to nine hours per night for optimal health (Watson et al., 2015).

Importantly, these health behaviors are interrelated. Insufficient sleep is associated with decreased leptin and elevated ghrelin levels, which can cue increased food intake (Spiegel et al., 2004). Decreased sleep is also associated with greater intake of processed foods and high-sugar refined foods (Godos et al., 2021) and less physical activity (Mead et al., 2019; Semplonius & Willoughby, 2018). Moderate physical activity can benefit sleep quality (Wang & Boros, 2021), while sedentary behaviors are associated with lower dietary quality (Gillman et al., 2001). Similarly, dietary quality impacts the energy available for physical activity, which in turn can affect sleep (Dolezal et al., 2017). Although there is evidence supporting these interrelationships, (Pan et al., 2022; Sarma et al., 2019) the co-occurrence of these behaviors among individuals affected by cancer remains unclear.

Increasing attention has been paid to health behaviors and their collective impact on health outcomes among cancer survivors. Glasgow and colleagues (2022) categorized a sample of 856 survivors participating in a national survey as either adherent or non-adherent to six health behavior recommendations (i.e., aerobic physical activity, strength training, fruit and vegetable consumption, sleep duration and quality, and sedentary time). They found that adherent individuals reported better psychological health. This work identified which behaviors are relevant to psychological health in a nationally representative sample. However, the analytic approach, which assessed health behaviors independently or as a single composite using multiple ordinal regression, precluded identification of patterns of health behavior engagement.

To address this gap, research is moving toward person-centered analytic approaches such as latent class analysis (LCA) or latent profile analysis (LPA). Olson and colleagues (2023a) utilized LCA to evaluate lifestyle behaviors of 591 cancer survivors and identified three patterns based on physical activity, diet, smoking, and sleep. However, behaviors were dichotomized prior to conducting the LCA, much like Glasgow and colleagues (2022). Similarly, Fong and colleagues (2023) conducted an LCA based on a combination of categorical (i.e., BMI, physical activity, smoking status, alcohol consumption, time since last physician visit) and continuous (i.e., dietary intake, sun safety) behaviors among 661 participants with cancer. Notably, sleep was absent. In these studies, patterns of behaviors were associated with quality of life and sociodemographic characteristics such as education and biological sex in small regional samples. However, the categorization of health behaviors in these analyses may have sacrificed information that could have identified more nuanced patterns of engagement. A gap remains in generalizability to a broader population of cancer survivors and how dietary quality, physical activity, and sleep may co-occur along a continuum.

Although health behavior patterns have been explored among cancer survivors, informal cancer caregivers (e.g., family members, friends) have generally been excluded. Informal caregivers provide a significant amount of unpaid cancer care and experience documented adverse health outcomes (Perkins et al., 2013; Pinquart & Sörensen, 2003; van Ryn et al., 2011; Yun et al., 2005). Although evidence regarding the impact of cancer caregiving on health behavior engagement is mixed (Litzelman et al., 2018; Ross et al., 2013; Secinti et al., 2022), cancer survivors and caregivers demonstrate interrelated health behaviors (Badr et al., 2019; Kiecolt-Glaser & Wilson, 2017). It is possible that by considering both survivors and caregivers, additional information could emerge regarding the prevalence, composition, and impact of health behavior patterns in the cancer context.

The present study had two aims. The first was to explore how dietary quality, physical activity, and sleep co-occur among a sample of cancer survivors and informal cancer caregivers. To achieve this, we used a person-centered analytic approach (LPA) to identify naturally occurring patterns of health behavior engagement. The second was to determine the association of sociodemographic and clinical variables to health behavior patterns to facilitate identification of individuals who may particularly benefit from health promotion intervention. Given the exploratory nature of this work, there were no a priori hypotheses.

Methods

Study Sample

Data were derived from the National Cancer Institute (NCI) Health Information National Trends Survey (HINTS) database. HINTS is a nationally representative, cross-sectional survey of the US non-institutionalized adult population (Finney Rutten et al., 2020; Nelson et al., 2004). Participants who completed the HINTS 5 Cycle 3 survey via mail or web (January-May 2019; response rate: 30.3%) in English or Spanish were included in the present analysis. Datasets and additional methodological documentation for HINTS 5 Cycle 3 are publicly available (U.S. Department of Health & Human Services, 2020; Westat, 2021). Although more recent data have been fielded for HINTS, the 2019 survey is the most recent survey that included information about diet, physical activity, and sleep in a singular dataset. The analytical sample was limited to individuals with a self-reported history of cancer (“Have you ever been diagnosed with having cancer?”; n = 856) or those who reported providing unpaid care to an individual with cancer (“Are you currently caring for or making health care decisions for someone with a medical, behavioral, disability, or other condition?” Caregiving condition: Cancer; n = 75). Fifteen respondents reported both having a history of cancer and being a current informal caregiver and were categorized as “Dual Role” (weighted N = 711,464). Those who reported a history of cancer with no informal caregiving responsibilities were categorized as “Cancer Survivor Role” (n = 841, weighted N = 22,754,490) and those who reported current informal cancer caregiving without a cancer diagnosis were categorized as “Caregiver Role” (n = 60, weighted N = 2,895,557). This study did not require Institutional Review Board approval because it used publicly available deidentified data.

Health Behaviors

Fruit and Vegetable Intake. Fruit and vegetable intake (FVI) were assessed using the following questions: “About how many cups of fruit (including 100% pure fruit juice) do you eat or drink each day?” and “About how many cups of vegetables (including 100% pure vegetable juice) do you eat or drink each day?” Survey respondents were provided with approximate serving sizes and could select “None”, “1/2 cup or less”, “1/2 cup to 1 cup”, “1 to 2 cups”, “2 to 3 cups”, “3 to 4 cups”, or “4 or more cups”. FVI was recoded (Supplemental Table A), summed, and analyzed continuously as an estimate of total daily cups of fruit and vegetables.

Physical Activity. Weekly minutes of moderate-to-vigorous physical activity (MVPA) was calculated from responses to two questions: “In a typical week, how many days do you do any physical activity or exercise of at least moderate intensity, such as brisk walking, bicycling at a regular pace, and swimming at a regular pace (do not include weightlifting)?” and “On the days that you do any physical activity or exercise of at least moderate intensity, how long do you typically do these activities?”. Daily minutes of physical activity were calculated by dividing the weekly minutes by seven. The resulting values were log transformed for normality prior to analysis.

Sleep Duration. Average nightly hours of sleep over the previous week were self-reported in response to the question “During the past 7 days, how many hours of sleep did you get on average per night?”.

Sociodemographic and clinical covariates

Sociodemographic and clinical information was self-reported. Sociodemographic characteristics available in HINTS included age, sex, race/ethnicity, geographic area (rural vs. urban designation), annual household income, household size, relationship status, and educational attainment. Cancer history included age at diagnosis, years since diagnosis (calculated from age at survey completion and age at diagnosis), family history of cancer, and cancer type. Other health characteristics included perceived general health (excellent to poor), health insurance coverage, comorbid conditions (diabetes, high blood pressure, heart disease, lung disease, or depression), body mass index (BMI; calculated as weight (kg)/height (m2), and BMI category (underweight- < 18.5 kg/m2, normal weight- 18.5–24.9 kg/m2, overweight- 25.0–29.9 kg/m2, and obese ≥ 30.0 kg/m2) (Weir & Jan, 2020). Smoking status (current, former, never) and alcohol use (none, moderate- ≥ 1 drinks/week, heavy- ≥ 8 drinks/week for females or ≥ 15 drinks/week for males) (Rock et al., 2020) were also included.

Other covariates

Self-efficacy and psychological distress, behavioral covariates in HINTS previously associated with health behaviors (Glasgow et al., 2022; Skiba et al., 2021), were also included. Self-efficacy was assessed using a single item “Overall, how confident are you about your ability to take good care of your health?” with responses provided on a 5-point Likert scale. Consistent with previous analyses, responses were dichotomized into low and high (Fleary et al., 2019; Niederdeppe & Levy, 2007; Skiba et al., 2021). Patient Health Questionnaire-4 (PHQ-4) scores were used to categorize psychological distress as: No Distress (0–2), Mild (3–5), Moderate (6–8) and Severe (9–12), consistent with previous analyses (Glasgow et al., 2022; Kroenke et al., 2009).

Statistical analysis

Data were managed according to NCI guidance (Moser et al., 2013). Full sample survey weights provided by HINTS were applied using the Taylor series variance estimation method to provide population level estimates for all analyses, as recommended. No significant differences were found between response mode (i.e., mail, web) for indicator variables. Differences in sociodemographic characteristics, cancer history, and health characteristics stratified to compare cancer survivors and caregivers were assessed using Chi-square tests for categorical variables and t-tests for continuous variables.

We conducted an exploratory latent profile analysis (LPA) to identify groups of individuals with similar co-occurring health behavior patterns. We fit an LPA model with one categorical latent variable and three observed continuous variables (FVI, MVPA, and sleep) via maximum likelihood and following best practices for model selection and interpretation (Masyn, 2013). Profiles were added iteratively to determine the best model fit. The optimal number of profiles was determined by interpretation of Akaike’s information criterion (AIC) and Bayesian information criterion (BIC), wherein smaller values indicated better fit, and global entropy, wherein larger values indicated better fit. (Nylund et al., 2007). Profiles were further evaluated for interpretability, clinical relevance, and sample size, and those containing < 5% of the sample were deemed spurious (Roesch et al., 2010). Profile membership was assigned based on highest probability of belonging. LPA models were not stratified based on role (survivor/caregiver), which was instead evaluated as a predictor of profile membership. Standardized profile means (z-scores) of the indicators were calculated to visualize differences among profiles.

Final profiles were used as a categorical outcome variable in subsequent multinomial logistic regression analyses to predict profile membership based on sociodemographic and health characteristics. Current age (< 65 vs. ≥ 65), age at diagnosis (≥ 40 vs. ≤ 39), years since diagnosis (≤ 10 vs. ≥ 11), and number of comorbid conditions (0 vs. ≥ 1) were dichotomized prior to regression analysis to facilitate interpretability. Analytical samples were restricted to complete cases, and individuals with missing responses for any covariate were excluded. An alpha level of 5% was considered statistically significant. All analyses were completed in STATA 18.0 (StataCorp LLC, College Station, TX, USA).

Sensitivity Analysis. A sensitivity analysis was conducted to facilitate interpretation in comparison to prior research (Fong et al., 2023; Glasgow et al., 2022; Jenny L Olson et al., 2023a, 2023b). To determine the proportion of respondents in each profile meeting established health behavior recommendations, responses were dichotomized as meeting or not meeting recommendations according to the following cut-points: 1) FVI: ≥ 2 cups of fruit and ≥ 3 cups of vegetables per day (Rock et al., 2020); 2) MVPA: ≥ 150 min per week (Rock et al., 2020); and 3) Sleep: 7 to 9 h per night (Hirshkowitz et al., 2015). Adherence to established guidelines by profile were compared using Pearson’s chi-squared tests of independence with Bonferroni adjustment for multiple comparisons.

Results

Descriptive statistics

Sample characteristics are in Table 1. All responses included in the present analysis were completed using the English HINTS survey; no respondents in the selected study sample completed the survey in Spanish. Cancer survivors were on average older than cancer caregivers (63.5 ± 15.9 and 55.5 ± 11.3 years, respectively). The most common cancer diagnosis was cutaneous (29%). Most cancer caregivers reported providing care to either a parent (37.9%) or partner (18.2%). Most respondents identified as non-Hispanic (84%) and resided in urban areas (83%). About half reported never smoking (54%) and no current alcohol use (46%), while 71% had overweight or obesity and reported one or more chronic comorbid health conditions.

Table 1 Weighted Sample Characteristics of Cancer Survivors and Informal Cancer Caregivers Participating in HINTS 5 Cycle 3, 2019
Full size table

Identified health behavior profiles

Four health behavior profiles were identified (Fig. 1): Least Engaged–No MVPA, Least Engaged–Low MVPA, Moderately Engaged, and Highly Engaged. See Supplemental Table B for model fit information. The Least Engaged–No MVPA profile was distinguished by no reported MVPA. Comparatively, the Least Engaged–Low MVPA profile was characterized by reported MVPA below the sample mean. These two profiles had similar FVI and sleep duration, which were below the sample mean. The Moderately Engaged profile was characterized by health behaviors resemblant to the sample mean. The Highly Engaged profile was characterized by engagement across health behaviors greater than the sample mean. Profiles were most distinguished by mean MVPA (Table 2). The largest profile group was Least Engaged–No MVPA with 37% of the sample.

Fig. 1
figure 1

Standardized Means (z-score) in the Indicators of Health Behaviors Among Four Health Behavior Profiles

Full size image
Table 2 Weighted Summaries of Health Behaviors by Health Behavior Profile in HINTS 5 Cycle 3, 2019
Full size table

Associations of profile membership and sociodemographic and health characteristics

Profile membership was significantly associated with sociodemographic and health characteristics (Table 3). All models used the Least Engaged–No MVPA profile as the reference group. Respondents in the Least Engaged–Low MVPA profile were significantly more likely to be a caregiver but less likely to be aged ≥ 65 years, identify as Hispanic, and have greater than a high school education. Respondents in the Moderately Engaged profile were significantly more likely to have higher annual household income, moderately use alcohol, and have high health self-efficacy. They were also significantly less likely to report an “other” cancer diagnosis, be partnered, have greater than a high school education, live in a rural setting, report fair-to-poor general health, have obesity, and report mild psychological distress. Respondents in the Highly Engaged profile were significantly more likely to be a caregiver, moderately use alcohol, and have high health self-efficacy. They were also significantly less likely to be aged ≥ 65 years, report a gynecologic cancer diagnosis, have greater than a high school education, and have obesity. No other sociodemographic or health characteristics were significantly associated with profile membership.

Table 3 Sociodemographic and Clinical Characteristics Associations with Health Behavior Profile Membership
Full size table

Sensitivity analysis—profile adherence to recommendations

Information about adherence to recommendations can be found in Supplemental Materials. Across profiles, the least adhered to behavior was FVI (6.9%) followed by MVPA (34.4%), while sleep had the greatest observed levels of adherence (56.1%). Post-hoc analyses indicated that percent adherence to FVI and MVPA recommendations were significantly different across profiles.

Discussion

We observed four distinct patterns of health behavior co-occurrence among a nationally representative sample of cancer survivors and informal cancer caregivers. In general, respondents who reported greater FVI also reported greater MVPA and sleep duration. Likewise, participants reporting lower FVI had lower MVPA and sleep duration. Approximately 20% of the sample was in the Highly Engaged profile, indicating that more than 80% warrant some degree of intervention to support dietary quality, physical activity, and/or sleep behaviors. We also identified that role, cancer type, self-efficacy, age, alcohol use, and educational attainment were associated with profile membership. This information could help identify for whom health behavior interventions could be of greatest benefit.

Our results were similar to Olson and colleagues’ (2023a) finding that most survivors had difficulties with physical activity and diet, with or without sleep difficulties, and that those with these challenges were less likely to report high health-related quality of life. Our results are also consistent with their finding that the most engaged were younger and had a lower BMI. However, unlike these authors, we found that MVPA had the greatest variance across profiles while they identified sleep as the key differentiating behavior. This difference may be due to their categorization of variables prior to analysis, or because their profiles were also indicated by smoking status. In another study, Olson and colleagues also found greater health behavior engagement benefiting quality of life among rural cancer survivors (2023b). We found that rural-dwelling respondents were least likely to be in the Moderately Engaged profile. This is consistent with the broader cancer survivorship literature, suggesting that greater efforts may be required to promote health behavior engagement among rural-dwelling cancer survivors (Werts et al., 2023).

Our study evaluated health behavior patterns in cancer survivors and caregivers from a singular survey. Caregivers in our analysis were more likely to be members of the Least Engaged – Low MVPA or Highly Engaged profile. Available data in HINTS were not dyadic in design (i.e., information linking individual survivors to their caregiver(s) was not collected), so we were not able to evaluate behavior patterns between cancer survivor-caregiver dyads in the present study. Such investigation would be important, as qualitative evidence has suggested that including an informal caregiver in interventions may help survivors engage in healthy behaviors (Skiba et al., 2022). However, dyadic health behavior interventions developed specifically with the well-being of both the cancer survivor and the informal caregiver remain limited (Bisht et al., 2023). Given the demonstrated interrelationship between survivor-caregiver health behaviors (Badr et al., 2019; Kiecolt-Glaser & Wilson, 2017), research focusing on multiple health behaviors in diverse cancer survivor-caregiving dyads is warranted.

Our use of a person-centered analysis that considered health behaviors on a continuous scale allowed for the evaluation of the interrelationship among behaviors through identification of profiles where health behavior engagement converged. Previous studies have identified patterns and correlates of health behaviors respective to current recommendations (Fong et al., 2023; Glasgow et al., 2022; Olson et al., 2023a, 2023b). Such an approach is important, as higher adherence to health behavior guidelines is associated with lower cancer incidence and mortality (Kohler et al., 2016). However, simply categorizing behaviors as adherent or non-adherent runs the risk of missing individuals who are close to the cut-off but may need just a little more support to be adherent. For example, we found that those in the Moderately Engaged profile on average reported eight fewer minutes of daily MVPA than recommended. Although these individuals may benefit from health behavior intervention, they are likely to need different components than those in either of the Least Engaged profiles. Had we dichotomized behaviors prior to analysis, we would not have identified this pattern, and in turn our conclusions may have encouraged inefficient application of health behavior support. Of the behaviors analyzed, we found that sleep had the lowest variance across profiles. Nonetheless, this corresponded to a 33 min per night difference in sleep duration between the profile with the lowest average sleep (Moderately Engaged) compared to highest (Highly Engaged). This difference is clinically relevant, as just a few minutes of additional sleep per night is related to improvements in resting heart rate and decreased cardiometabolic risk (Rezaei & Grandner, 2021). Adult survivors of cancer have significantly heightened risk for cardiovascular disease relative to those with no history of cancer (Florido et al., 2022), and findings from the National Health Interview Survey have shown that survivors have lower odds of obtaining adequate sleep compared to cancer-free controls (Boyd et al., 2020). Therefore, despite the lack of cancer-specific sleep recommendations, our results highlight the criticality of including sleep when developing multiple health behavior interventions for cancer survivors.

An important consideration when examining co-occurrence of health behaviors is that these behaviors are mutually time exclusive. Dietary intake, physical activity, and sleep happen rhythmically throughout each 24-h period. In 2022, the average US adult spent an estimated 1.23 h eating or drinking, 0.65 h preparing food and cleaning up, 0.29 h participating in recreational physical activity, and 9.02 h sleeping in each 24-h period (U.S. Bureau of Labor Statistics, 2023). Measurement of activity and sleep combined with eating times in a 24-h period may improve future understanding and promotion of co-occurring health behaviors in cancer survivorship (Kirkham et al., 2022; Shirazipour et al., 2023). Future interventions may consider the balance of multiple health behaviors across a 24-h period to promote greater behavior change.

Strengths and limitations

This study provides early evidence of the co-occurrence of multiple synergistic health protective behaviors in a sample of cancer survivors and informal cancer caregivers. Unique to this analysis, we pooled together cancer survivors and informal cancer caregivers. Indicator variables were evaluated on a continuous scale, providing greater granularity. The utilization of HINTS data provided a nationally representative sample, increasing the generalizability of our findings to cancer survivors and informal cancer caregivers across the United States. Moreover, the methodology employed by HINTS reduced the risk of sampling bias through random sampling and nonresponse bias through applying weights (Maitland et al., 2017).

There is a potential for recall bias, as HINTS data are self-reported health behaviors and the selected behaviors come from singular survey items. There may also be measurement error in reporting or potential overestimation of FVI due to the response scale on which the original data were collected and our subsequent recoding to enable evaluation on a continuum. The evaluation of sleep duration as a continuous variable did not account for the documented curvilinear relationship between sleep and health (Hoogland et al., 2019), with both short and long sleep linked to poor outcomes. Weighted ranges in our sample demonstrated that most long-sleepers (i.e., ≥ 9 h per night) were in the Least Engaged–No MVPA profile, further supporting the potential utility of health behavior intervention for this group. Data included in this analysis were collected prior to the COVID-19 pandemic, which may have changed population engagement in health behaviors, although this remains unclear (Cole et al., 2023; Donzella et al., 2021; Ye & Ren, 2022). Finally, the temporality of the relationships examined and changes over time cannot be determined in this cross-sectional study.

Implications and future research

In 2021, HINTS fielded a cancer survivor specific survey that did not capture sleep (Blake et al., 2023). Given the co-occurrence of behaviors we observed in the present analysis and the clinical relevance of sleep for cancer outcomes, it is important that future surveys measuring health behaviors include assessment of sleep. Developing interventions that focus on promoting multiple health behaviors, rather than a singular behavior, remains a priority to address the deleterious effects cancer treatment and cancer caregiving can have on health outcomes (Carroll et al., 2022; Guida et al., 2021). Multiple studies promote diet and physical activity together and sleep separately in cancer survivors (Fox et al., 2022; Rodrigues et al., 2023; Squires et al., 2022; Thomson et al., 2023), but little work has championed all three simultaneously. Theoretically informed behavior change interventions can support strategic selection of techniques (Abraham & Michie, 2008). Additionally, due to the concordant tendencies of behaviors within survivor-caregiver dyads, future research should consider dyadic dynamics in the design, assessment, and interpretation of health behavior promotion interventions (Badr et al., 2019).

Conclusions

This analysis of a nationally representative sample of cancer survivors and cancer caregivers found that dietary quality, physical activity, and sleep duration generally co-occurred. Profiles were most distinguishable by MVPA relative to FVI and sleep duration. Profiles also differed by numerous sociodemographic and clinical variables, including survivor/caregiver role, current age, relationship status, education, income, rurality, alcohol use, self-efficacy, psychological distress, BMI, and cancer type. These characteristics, combined with profile membership, may identify cancer survivors and informal cancer caregivers who are most likely to engage in multiple health behaviors, thus not needing intervention, and those for whom health behavior intervention may be warranted.