The CSM is the most comprehensive and evidenced-based theoretical approach applied to FCR (Fardell et al., 2016). Originally developed to encompass the cognitive, behavioral, and emotional responses to various illnesses (Leventhal et al., 1992), Lee-Jones et al. (1997) applied the CSM components to the context of cancer in their FCR theoretical formulation. The CSM components have since been empirically validated in cancer survivors (Fardell et al., 2016; Simonelli et al., 2017). According to this theoretical formulation, when an illness threat (triggers, i.e., aches and pains) is perceived, it activates the cancer survivor’s illness representation informing the selection of coping response, which will ultimately influence the illness and emotional outcomes, including FCR (Leventhal et al., 1992; Lee-Jones et al., 1997).

The illness representation is comprised of five illness attributes: illness identity – refers to the illness label (cancer) and related symptoms (e.g., fatigue); consequences – refers to the perceived impact of cancer on an individual’s life, including social, psychological, and physical consequences (e.g., impact on family); control – refers to the perceived level of control over cancer or curability by oneself or others (e.g., incurable, recurrence preventable); timeline – refers to the perceived time frame of cancer growth, illness course, and recovery (e.g., acute, chronic, or cyclical); and causes – refers to the perceived cause of cancer (e.g., stress, unhealthy lifestyle, or family history; Leventhal et al., 1992).

These attributes will inform the coping response chosen by the patient to manage emotional (typically with emotion focused coping) and/or illness outcomes (typically with problem-focused coping; Hagger et al., 2017). In samples of chronically ill patients, illness identity, consequences, and timeline are often correlated with emotion-focused coping, while control is more related to problem-focused coping (Richardson et al., 2016; Hagger et al., 2017). Health behaviors are generally conceptualized as problem-focused coping to manage illness outcomes (Richardson et al., 2016; Hagger et al., 2017); however, they have been shown to help manage emotional outcomes such as depression and quality of life (Aguiñaga et al., 2018). Therefore, from a theoretical standpoint, it is possible that health behaviors can help manage FCR, an emotional outcome.

Based on the CSM, if health behaviors are a coping response to FCR, cancer survivors with an illness representation that is indicative of a more severe illness (e.g., those who perceive cancer to be chronic or its consequences to be more significant) are expected to engage in more physical activity (PA) and fruit and vegetable intake (FVI). Subsequently, if these coping strategies are appraised as effective, FCR should be reduced. The few studies using the CSM framework to test the relationship between illness representation, PA and FVI, have yielded mixed results (Mullens et al., 2004; Costanzo et al., 2011; Burris et al., 2012; McGinty et al., 2012; Green et al., 2014), suggesting the need for further investigation.

Additionally, previous studies of FCR and health behaviors using the CSM were restricted to one or two disease sites and confined to early survivorship (i.e., 2 years post active treatment; Stanton et al., 2015) limiting their generalizability. Additional studies are required to clarify the contradictory findings using a large sample of survivors with a range of cancer diagnoses.

In the context of cancer survivorship, self-efficacy is defined as the perceived confidence in handling problems related to one’s health (Bandura, 1997). In one study, the inclusion of self-efficacy improved the fit of the CSM in predicting personal control over illness in older adults with 10 different chronic diseases (Schüz et al., 2012). Additionally, several studies have found that self-efficacy played a mediating role in predicting FCR. For example, improvements in self-efficacy were found to mediate the effects of a brief communication intervention on FCR (Shields et al., 2010), between FCR vulnerability factors such as trait anxiety and cancer reminders and FCR in women with breast cancer (Ziner et al., 2012), and between physical symptoms and FCR in men with prostate cancer (Torbit et al., 2015). In the present study, health self-efficacy was conceptualized as the appraisal of the coping response (i.e., the health behaviors) and was expected to mediate the relationship between the coping and the emotional outcome, FCR.

This cross-sectional study aimed to explore the relationships between constructs from the CSM and self-efficacy theory, and health behaviors (PA and FVI) and FCR in a population-based sample of survivors of 10 cancers [tobacco use was not included in the current analyses as its relationship with FCR was the focus of a separate study by Westmaas et al. (2019)]. We used data from wave 3 of the American Cancer Society’s Study of Cancer Survivors-I (SCS-I) because only the wave 3 survey included assessment of three constructs from the CSM (described below) in addition to health behaviors, self-efficacy, and FCR. The following relationships were expected (see Figure 1):

The current cross-sectional study is part of a larger longitudinal study examining FCR and health behaviors (Séguin Leclair et al., 2019) using data from the American Cancer Society’s SCS-I, a national prospective longitudinal study of American cancer survivors with data collected in three waves beginning in 2000, T1, M = 1.3 years (SD = 0.32), T2, M = 2.2 years (SD = 0.34), and T3, M = 8.8 years (SD = 0.63) post cancer diagnosis. Participant eligibility criteria were the following: diagnosed with one of the 10 most highly incident cancers [prostate, breast, lung, colorectal, bladder, non-Hodgkin lymphoma (NHL), skin melanoma, kidney, ovarian, and uterine], over 18 years old at diagnosis, residing in one of the target states at the time of diagnosis, and diagnosed with a local, regional, or distant SEER Summary Stage cancer. Survivors were ineligible for the study if they were unable to complete the survey due to mental incompetence, unable to communicate in English or Spanish, or had terminal illness (Smith et al., 2007). The studies were approved by the Institutional Review Board of Emory University (Atlanta, GA, United States), for each state, including the Connecticut Department of Public Health Human Investigation Committee, and the University of Ottawa Research and Ethics Board (Ottawa, Ontario). Additional details on recruitment and methodology are available elsewhere (Smith et al., 2007).

Data were screened and cleaned using IBM SPSS 25. See Séguin Leclair et al. (2019) for detailed description of sample selection. Means, SD, and bivariate correlations were computed for all model variables. Statistical assumptions for regression analysis were verified.

Path analysis was conducted to test the hypothesized model (see Figure 1) using IBM AMOS at a level of significance p < 0.05. Path coefficients were standardized to facilitate comparison and interpretation of data. Bootstrapping with 2,000 samples and 95% CIs was used to calculate indirect effects. Model fit was established using the following goodness-of-fit indices with corresponding criteria: a small and non-significant chi-square likelihood ratio statistic (χ²), standardized root mean square residual (SRMR) ≤ 0.08, comparative fit index (CFI) ≥ 0.95, and root mean square error of approximation (RMSEA) ≤ 0.06 (Hayduk et al., 2007). Using the modification indices proposed by AMOS, additional regression weights were sequentially added to the model until the goodness-of-fit indices reached previously mentioned criteria. Concurrently, theoretical meaning was considered before the addition of parameters in the model. Adequate sample size was reached for the total sample (with 19 model parameters the minimum sample size required is n = 190), based on the suggested 10 participants/parameter (Kline, 2015). Given that variables met the normality assumption, the maximum likelihood estimation method was used.

The 2,337 participants in this sample were mostly Caucasian (89.9%) women (60.4%) with college education or more (68%) and with a mean age of 56 at diagnosis. The most common cancer sites were breast (28.6%), prostate (21%), and colorectal (13.6%). See Table 1 for all participants’ socio-demographic and medical characteristics.

Table 2 displays the means, SDs, and bivariate correlations for the seven variables (illness consequences, control, timeline, physical activity, FVI, self-efficacy, and FCR) in the hypothesized model. Overall, survivors reported an average score of M = 11.4 (SD = 7.1) on the FCRI-SF, with 32.9% scoring above the clinical cut-off >13. The average number of days participants consumed five servings of fruits and vegetables per week was 3.31 days (SD = 2.17). Participants reported engaging in 97.59 (SD = 151.28) min of moderate to vigorous physical activity per week.

The hypothesized model was tested but yielded poor goodness-of-fit indices, χ² (9, N = 2,337) = 1718.72, p < 0.001; SRMR = 0.16; CFI = 0.49; RMSEA = 0.29. By adding four additional paths as suggested by the modification indices, the model fit improved [χ² (5, N = 2,337) = 38.12, p < 0.001; SRMR = 0.02; CFI = 0.99; RMSEA = 0.05; see Figure 2]. The chi-square likelihood ratio statistic remained significant, but the model was deemed acceptable given its large sample size.

The final model indicated that both perceiving fewer illness consequences, and greater control over one’s health, were directly associated with higher PA (β = 0.15 and −0.24, p < 0.01, respectively) but also higher health self-efficacy (β = 0.24, −0.38, p < 0.01, respectively). Timeline was not directly associated with PA or FVI but was associated with lower health self-efficacy (β = −0.15, p < 0.01) and higher FCR (β = 0.51, p < 0.01). Both greater PA and FVI were directly associated with higher health self-efficacy (β = 0.10 and 0.11, p < 0.01, respectively), which in turn was directly associated with lower levels of FCR (β = −0.15, p < 0.01).

Small indirect effects were also observed between illness perception variables and health self-efficacy and FCR: timeline had an indirect effect on FCR (β = 0.06, p < 0.01), control had indirect effects on health self-efficacy (β = −0.01, p < 0.01) and FCR (β = 0.10, p < 0.01), and illness consequences also had indirect effects on health self-efficacy (β = 0.001, p < 0.01) and FCR (β = −0.02, p < −0.01). Last, PA and FVI had small indirect effects on FCR (β = −0.001 and −0.05, p < 0.01, respectively).

Although the American Cancer Society had developed a protocol to obtain an optimal sample of American cancer survivors, individuals who completed the questionnaire packages have specific characteristics, such as being female, White, and higher education (Smith et al., 2007). This limits the generalizability of findings. In addition to its cross-sectional design, this study captured FCR, CSM, self-efficacy, and health behaviors later in the cancer survivorship trajectory. While FCR severity was found to be stable across the three waves of data of the SCS-I (Séguin Leclair et al., 2019), factors in the CSM model have been shown to fluctuate over time (Leventhal et al., 2016). Furthermore, information of disease recurrence/progression was not available. Future studies should monitor health behaviors, including changes in these behaviors from pre-diagnosis, and CSM factors periodically throughout the survivorship trajectory, controlling for possible recurrence/progression (Leventhal et al., 2016). While the questionnaires used to assess illness representation components in this study were adequate measures of the constructs, the Revised Illness Perceptions Questionnaire was not used in this study (Moss-Morris et al., 2002). Given that this measure is commonly used in other CSM studies, this limits generalizability across studies. Moreover, the perceived cause of cancer recurrence, another component of illness representation in the CSM model, was not included in the original SCS-I survey questionnaire. As previous studies have consistently shown, causal attribution of cancer to poor diet or lack of exercise predicts adherence to health behaviors (Mullens et al., 2004; Costanzo et al., 2011; Burris et al., 2012); the absence of this measure might limit the understanding of factors predicting health behaviors. This study only focused on one possible emotional outcome outlined in CSM, FCR. Future studies could include additional outcomes that are associated with illness representation such as psychological distress (Llewellyn et al., 2007).

It would be important to replicate the present study using a longitudinal design that would allow taking into account FCR trajectories, given the emerging empirical evidence that cancer survivors can be classified into three FCR severity sub-groups: low, moderate, and high, which have distinct survivorship profiles and patient characteristics (Simard and Savard, 2009; Simonelli et al., 2017; Séguin Leclair et al., 2019). As part of the larger study (Séguin Leclair et al., 2019), we found three stable FCR trajectories (low, moderate, and high); furthermore, cancer survivors in the high FCR trajectory group engaged in less health behaviors than other survivors. Therefore, it is possible that health behaviors may play less of a role in modulating FCR for those with persistently high FCR. It would also be interesting to see if these relations we observed in the present sample differ by sex, ethnicity, or cancer stage.

The results of this study offer preliminary evidence that engaging in PA or consuming fruits and vegetables may increase health self-efficacy, which in turn may be beneficial to manage FCR in long-term survivors. In a previously published paper using the three waves of data from the SCS-I, we also found that survivors who quit smoking at T1 reported significant reductions in FCR at T3 compared to those who continued smoking (Westmaas et al., 2019). Together, these results support stepped care models that propose that all survivors would receive educational programs on health behaviors to help manage FCR (Stanton, 2012). The CSM showed a good fit across participants, further supporting its use in current FCR conceptualizations (Fardell et al., 2016; Simonelli et al., 2017) and interventions (Maheu et al., 2016; Butow et al., 2017). Timeline (i.e., perceiving cancer to be chronic) stood out as the strongest correlate of FCR in the model; hence interventions targeting the perceived chronicity of cancer may help cancer patients manage their FCR. Furthermore, improving cancer survivors’ self-efficacy to manage their health could be an FCR intervention target.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.