Introduction
It has been long debated whether depression and anxiety increase the risk of cancer. Previous meta-analytic evidence of population-based studies has shown mixed results (Ahn, Bae, Ahn, & Hwang, Reference Ahn, Bae, Ahn and Hwang2016; Jia et al., Reference Jia, Li, Liu, Zhao, Leng and Chen2017; Oerlemans, van den Akker, Schuurman, Kellen, & Buntinx, Reference Oerlemans, van den Akker, Schuurman, Kellen and Buntinx2007; Sun et al., Reference Sun, Dong, Cong, Gan, Deng, Cao and Lu2015; Wang et al., Reference Wang, Li, Shi, Que, Liu, Lappin and Bao2020). However, these meta-analyses had several limitations, such as variations across studies included in the assessments of depression and anxiety, cancer diagnosis and assessment, and the covariates considered. To address these limitations, we previously performed individual participant data (IPD) meta-analyses of 18 cohort studies (N = 319 613, cancer incidence = 25 803) and found that depression and anxiety were associated with increased risk of lung and smoking-related cancers independent of demographic factors. This association was attenuated after further adjustment for health behaviors, such as smoking and physical activity. In contrast, depression and anxiety were not associated with the risk of overall cancer, or breast, prostate, colorectal, or alcohol-related cancers (van Tuijl et al., Reference van Tuijl, Basten, Pan, Vermeulen, Portengen, de Graeff and Ranchor2023).
Health behaviors may explain part of the association among depression, anxiety, and cancer risk. Previous evidence indicates that individuals with a diagnosis of depression or anxiety are more likely to smoke cigarettes, to drink alcohol heavily, to have a higher body mass index (BMI), and be more physically inactive compared with those without such a diagnosis (Penninx, Reference Penninx2017; Strine et al., Reference Strine, Mokdad, Dube, Balluz, Gonzalez, Berry and Kroenke2008). These health behaviors and health-related factors (e.g. BMI, here collectively described as health behaviors) are known risk factors for various types of cancer. Smoking is a risk factor for lung, breast, and colorectal cancer, while high alcohol consumption and a high BMI have been linked to breast and colorectal cancer (Dekker, Tanis, Vleugels, Kasi, & Wallace, Reference Dekker, Tanis, Vleugels, Kasi and Wallace2019; National Center for Chronic Disease Prevention and Health Promotion (US) Office on Smoking and Health, 2014; Rojas & Stuckey, Reference Rojas and Stuckey2016). Physical inactivity is a risk factor for breast, colorectal, and lung cancer (Dekker et al., Reference Dekker, Tanis, Vleugels, Kasi and Wallace2019; Friedenreich, Ryder-Burbidge, & McNeil, Reference Friedenreich, Ryder-Burbidge and McNeil2021; Kerr, Anderson, & Lippman, Reference Kerr, Anderson and Lippman2017; Rojas & Stuckey, Reference Rojas and Stuckey2016). Physical inactivity has also been related to prostate cancer, although the evidence for this is weaker (Friedenreich et al., Reference Friedenreich, Ryder-Burbidge and McNeil2021; Kerr et al., Reference Kerr, Anderson and Lippman2017).
In addition, depression and anxiety disorders have been linked to sedentary behavior (Hiles, Lamers, Milaneschi, & Penninx, Reference Hiles, Lamers, Milaneschi and Penninx2017), abnormal sleep duration, and poor sleep quality (van Mill, Hoogendijk, Vogelzangs, van Dyck, & Penninx, Reference van Mill, Hoogendijk, Vogelzangs, van Dyck and Penninx2010). These health behaviors may also be related to a higher risk of cancer. For example, sedentary behavior, independent of physical inactivity, has been shown to increase the risk of colon and lung cancer (Friedenreich et al., Reference Friedenreich, Ryder-Burbidge and McNeil2021; Kerr et al., Reference Kerr, Anderson and Lippman2017; Schmid & Leitzmann, Reference Schmid and Leitzmann2014). There is meta-analytical evidence linking a long sleep duration to a higher risk of colorectal cancer but not with overall cancer (Chen et al., Reference Chen, Tan, Wei, Li, Lyu, Feng and Li2018), while a large epidemiological cohort study in elderly found poor sleep quality to be linked with a higher risk of overall cancer (Song et al., Reference Song, Zhang, Wang, Fu, Song, Dou and Wang2021). These findings suggest that both sedentary behavior and sleep are potential mediators of the association among depression, anxiety, and cancer.
Although behavioral mechanisms in the association among depression, anxiety, and cancer are plausible, few studies have empirically studied mediation by health behaviors. To the best of our knowledge, only one study to date has shown that smoking mediates the association between depressive symptoms and lung cancer in the Nurses' Health Study (Trudel-Fitzgerald, Zevon, Kawachi, Tucker-Seeley, & Kubzansky, Reference Trudel-Fitzgerald, Zevon, Kawachi, Tucker-Seeley and Kubzansky2022). This study (N = 42 913 women) identified 1009 cases of lung cancer over 24 years, finding that smoking partially mediated the association between depressive symptoms and lung cancer.
Our previous IPD meta-analyses found associations of depression and anxiety with the incidence of lung cancer and smoking-related cancers, but not with the incidence of overall cancer or breast, prostate, colorectal, or alcohol-related cancers (van Tuijl et al., Reference van Tuijl, Basten, Pan, Vermeulen, Portengen, de Graeff and Ranchor2023). In the present study, we also considered the cancer outcomes for which no significant main effect was observed. While the causal steps approach for mediation analysis requires a significant total effect of a predictor on the outcome, the counterfactual approach suggests that the absence of a main effect does not preclude mediation. This is because there may be inconsistent mediation, such that the direct and indirect effects may point in opposite direction and thus may cause the total effect to be close to zero (Fairchild & McDaniel, Reference Fairchild and McDaniel2017; Mackinnon & Fairchild, Reference Mackinnon and Fairchild2009).
Our aim was to evaluate if health behaviors (smoking, physical inactivity, alcohol use, a high BMI, sedentary behavior, sleep duration, and sleep quality) mediate the relationships between depression, anxiety, and risk of cancer (overall cancer, breast, prostate, lung, colorectal, smoking-related, and alcohol-related cancers). Based on the literature, our hypotheses were as follows (van Tuijl et al., Reference van Tuijl, Voogd, de Graeff, Hoogendoorn, Ranchor, Pan and Dekker2021): (1) smoking and physical inactivity mediate the association among depression, anxiety, and lung cancer; (2) smoking mediates the association among depression, anxiety, and smoking-related cancers; (3) smoking, physical inactivity, alcohol use, and a high BMI mediate the association among depression, anxiety, and overall cancer; (4) smoking, physical inactivity, alcohol use, and a high BMI mediate the association among depression, anxiety, and breast cancer; (5) physical inactivity mediates the association among depression, anxiety, and prostate cancer; (6) smoking, physical inactivity, alcohol use, and a high BMI mediate the association among depression, anxiety, and colorectal cancer; (7) alcohol use mediates the association among depression, anxiety, and alcohol-related cancers. Analyses other than these hypotheses were considered exploratory.
Methods
Study design
The Psychosocial Factors and Cancer Incidence (PSY-CA) study consists of 18 prospective cohort studies in the Netherlands, the UK, Norway, and Canada. Three cohorts included multiple subcohorts that were considered separately, resulting in 22 cohorts for analysis. A detailed description of the PSY-CA study can be found elsewhere (van Tuijl et al., Reference van Tuijl, Voogd, de Graeff, Hoogendoorn, Ranchor, Pan and Dekker2021, Reference van Tuijl, Basten, Pan, Vermeulen, Portengen, de Graeff and Ranchor2023). We undertook two-stage IPD meta-analyses. In the first stage, standardized analyses were performed on harmonized datasets of participating cohorts. In the second stage, meta-analyses were performed to pool cohort-specific effect estimates. The present study was pre-registered on PROSPERO (https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42020193716).
Study population
We used data from cohort studies that had information on depression (symptoms or diagnosis) or anxiety (symptoms or diagnosis), including 14 cohorts and 4 subcohorts. We excluded participants who had a history of cancer at baseline, except when the cancer was non-melanoma skin cancer. Participants with any cancer diagnosis in the first year of follow-up were excluded to reduce the risk of reverse causality.
Depression and anxiety
Symptoms of depression and anxiety were assessed using validated, self-report questionnaires. As various questionnaires were used across cohorts, continuous sum scores were converted to z-scores in each cohort. Depression diagnosis (including major depressive disorder and dysthymia) and anxiety diagnosis (generalized anxiety disorder, social anxiety, panic disorder, and agoraphobia) were based on clinical interviews or, if not available, on questionnaires using clinically validated cut-offs. Several cohorts used depression and anxiety symptom questionnaires as a screener to identify and invite participants to a clinical interview. Participants with scores below the screening cut-off were considered to not meet the diagnostic criteria for either depression or anxiety. Cohort-specific details on depression and anxiety variables are provided in online Supplementary Table S1.
Cancer incidence
Seven cancer types were considered: overall cancer, breast cancer, colorectal cancer, lung cancer, prostate cancer, smoking-related cancers, and alcohol-related cancers. Online Supplementary Table S2 provides the ICD codes of these cancer outcomes. Cancer cases, including cancer type and date of diagnosis, were identified through linkage with national or regional registries in all cohorts. In two cohorts (CARTaGENE and Rotterdam study) data from hospital visits, insurance claims, and general practitioner records were also taken into account.
Health behaviors
All health behaviors were assessed at baseline. The four main health behaviors included in the hypotheses were: number of cigarettes per week (or equivalent of other tobacco smoking), hours of physical activity per week, number of alcoholic drinks per week, and assessed or self-reported BMI.
The three health behaviors in exploratory analyses were: hours of sedentary behavior per week (or hours of TV watching per week), sleep quality, and night-time sleep duration. Cohort-specific details on the availability and assessment of health behaviors are provided in online Supplementary Table S3.
We operationalized most health behaviors as continuous variables in the main analyses and as categorical variables in sensitivity analyses. Continuous variables were converted to z-scores in each cohort. Sleep duration was only used as a categorical variable with the categories short (<7 h), normal (≥7 to <9 h), and long (≥9 h) sleep (Chen et al., Reference Chen, Tan, Wei, Li, Lyu, Feng and Li2018). Smoking status was dichotomized as current smoker and non-current smoker. Excessive alcohol use was defined as more than seven drinks per week based on recommendations of the Dutch Health Council (Kromhout, Spaaij, de Goede, & Weggemans, Reference Kromhout, Spaaij, de Goede and Weggemans2016). Overweight is defined as a BMI of 25 or higher. Physical activity, sedentary behavior, and sleep quality were categorized into tertiles.
Covariates
Sociodemographic covariates were available in all cohorts, including birth year, age, sex, country of origin (whether the participant and his or her parents were born in the country in which the study was carried out), and educational level (low, medium, and high). For HUNT3, instead of educational level, profession was used as an alternative socioeconomic indicator.
Availability of additional covariates differed across cohorts, which is summarized in online Supplementary Table S4. Current or a history of anti-depressant use was self-reported. Depending on the cancer outcome and data availability in each cohort, self-reported family history of (overall, breast, prostate, lung, and colorectal) cancer referred to the cancer history of the participant's parents, siblings, and/or children. In analyses with breast cancer as the outcome, the following covariates were additionally included if available: parity (nulliparity, 1–2 pregnancies, and ≥3 pregnancies), menarche age, menopausal status, and oral contraceptive use.
Statistical analysis
We performed a meta-analysis of causal mediation analysis following procedures outlined by Zhu, Centorrino, Jackson, Fitzmaurice, & Valeri, (Reference Zhu, Centorrino, Jackson, Fitzmaurice and Valeri2021). The meta-analysis was carried out in two stages: (1) local analysis in each cohort to retrieve estimates of pathways in mediation models (Fig. 1) and (2) meta-analysis to pool pathways estimated in the first stage and perform causal mediation analysis. Mediation effects were estimated for both single mediator models (Fig. 1b) and parallel multiple mediator models using the four hypothesized health behaviors (Fig. 1c). In the parallel multiple mediator models, we included 15 cohorts where all four health behaviors were available. We explored a parallel multiple mediator model that included all seven health behaviors which were available in four cohorts (online Supplementary Fig. S1).
Figure 1. Theoretical framework of mediation analysis.
Stage one: local analysis
First, we estimated the association between depression/anxiety and the incidence of cancer using Cox models in which no health behavior was entered (i.e. path c; Figure 1a), and entry age (age at baseline) and exit age (age at diagnosis, death, or drop-out/study end) were used as the underlying time scale (van Tuijl et al., Reference van Tuijl, Basten, Pan, Vermeulen, Portengen, de Graeff and Ranchor2023). Next, we estimated the association between depression/anxiety and each health behavior (i.e. path a; Figure 1b) using linear regression models for continuous health behaviors or (multinomial) logistic regression models for categorical health behaviors. For single mediator models, we estimated the associations between a single health behavior and cancer (i.e. path b; Figure 1b) and the associations between depression/anxiety and cancer while controlling for a single health behavior (i.e. path c′; Figure 1b) using Cox models. For the parallel multiple mediator models, we estimated Cox models that contained the associations between all health behaviors and cancer (i.e. paths b; Figure 1c; Online Supplementary Fig. S1) and the associations between depression/anxiety and cancer while controlling for all health behaviors simultaneously (i.e. path c′; Figure 1c; Online Supplementary Fig. S1). Intercepts, coefficients, and variance–covariance matrices of regression models were extracted for stage two.
The estimated models were adjusted for two confounder sets: (1) a minimally adjusted model included sociodemographic covariates available across all cohorts: birth year, sex, educational attainment, and country of origin; (2) a maximally adjusted model included other potential confounders depending on cancer outcome and availability within the cohort. Online Supplementary Table S4 gives an overview of covariates added to each model in each cohort.
Considering the heterogeneous characteristic of alcohol non-drinkers (Rosansky & Rosenberg, Reference Rosansky and Rosenberg2020), we additionally performed a subgroup analysis focusing on alcohol use among those who consumed at least one alcoholic drink a week.
To test exposure–mediator interaction, we estimated the associations among depression/anxiety, health behavior, and their product term with each cancer outcome in Cox models. As summarized in another study of PSY-CA (Basten et al., Reference Basten, Pan, van Tuijl, de Graeff, Dekker, Hoogendoorn and Geerlings2024), the exposure–mediator interaction was generally not statistically significant and therefore omitted from the mediation analyses.
The inclusion of cohort-level results from stage one into stage two was based on two criteria to avoid the inclusion of unreliable estimates due to low number of cases: (1) each model should include at least ten cancer cases and (2) the a priori expected number of cancer cases among individuals with a depression/anxiety diagnosis should be at least five. The expected number of cases was calculated by multiplying the proportion of depression/anxiety diagnosis by the number of cases in the cohort.
Stage two: meta-analysis
The estimated intercept, path a, path b, and path c′, as well as the corresponding variances and covariances, from each cohort were entered into a random-effects multivariate meta-analysis. We estimated between-cohort heterogeneity in path a, b and c′ using I 2 for each path (Higgins & Thompson, Reference Higgins and Thompson2002), with the restricted maximum likelihood (REML) estimator.
Based on the pooled coefficients, we calculated natural direct effects, natural indirect effects (i.e. mediating effects), and total effects in hazard ratios (HRs) (VanderWeele, Reference VanderWeele2011). We obtained 95% Monte-Carlo confidence intervals (CIs) of these effect estimates based on the pooled variance–covariance matrices. Effects were statistically significant when a HR of 1 was not included in the CI. In multiple mediator models with categorical variables, we used the pooled path c to represent total effect, as little guidance is available on the computation of total effect for this scenario. We obtained the pooled path c using random-effects univariate meta-analysis.
Sensitivity analysis
Models were rerun with continuous health behaviors converted to categorical variables (see health behaviors above).
Results
Cohort characteristics
Table 1 shows characteristics of each cohort included in the current study, and Table 2 shows follow-up duration and number of cancer outcomes of each cohort. In total, there were 319 613 participants involved in the study, including 25 803 cancer diagnoses and 3 254 714 person years of follow-up. Mean age at baseline per cohort ranged between 27.6 and 75.7 years, and 24.8–100% of participants were female.
Table 1. Baseline characteristics of each cohort
ALSPAC, Avon Longitudinal Study of Parents and Children (mothers cohort); Atlantic PATH, Atlantic Partnership for Tomorrow’s Health; BMI, body mass index; ELSA, English Longitudinal Study of Ageing; HELIUS, Healthy Life in an Urban Setting; HUNT, Nord-Trøndelag Health Study; LASA, Longitudinal Aging Study Amsterdam; NESDA, Netherlands Study of Depression and Anxiety; OHS, Ontario Health Study; RS, Rotterdam Study; UCC-SMART-2, Utrecht Cardiovascular Cohort – Second Manifestations of Arterial Disease 2; UHP, Utrecht Health Project.
a Only contributed to analyses related to symptoms of depression and anxiety.
Table 2. Follow-up duration and number of cancer events per type and cohort
ALSPAC, Avon Longitudinal Study of Parents and Children (mothers cohort); Atlantic PATH, Atlantic Partnership for Tomorrow’s Health; ELSA, English Longitudinal Study of Ageing; HELIUS, Healthy Life in an Urban Setting; HUNT, Nord-Trøndelag Health Study; LASA, Longitudinal Aging Study Amsterdam; NESDA, Netherlands Study of Depression and Anxiety; OHS, Ontario Health Study; RS, Rotterdam Study; UCC-SMART-2, Utrecht Cardiovascular Cohort – Second Manifestations of Arterial Disease 2; UHP, Utrecht Health Project.
Lung cancer and smoking-related cancers
We report results from maximally adjusted models only because they were largely similar to those from minimally adjusted models. For lung cancer, in single-mediator models, number of cigarettes smoked mediated the associations among depression (symptoms and diagnosis), anxiety (symptoms and diagnosis), and lung cancer (HRs range 1.04–1.10). Physical inactivity mediated the associations among diagnoses of depression, anxiety, and lung cancer, with small mediating effects (HRs 1.02 and 1.01, respectively). In exploratory analyses, individuals with more depression symptoms or a depression diagnosis had a higher BMI, which in turn was associated with a lower risk of lung cancer (HRs 0.99 and 0.97, respectively). Sedentary behavior mediated the associations among depression (symptoms and diagnosis), anxiety (symptoms and diagnosis), and lung cancer (HRs range 1.01–1.02). These results were similar in multiple mediator models (Table 3 and online Supplementary Table S5). Alcohol use (among drinkers) mediated the associations among depression (symptoms and diagnosis), anxiety (symptoms and diagnosis), and lung cancer in single mediator models, but not in multiple mediator models (Table 3).
Table 3. Mediation of smoking, physical inactivity, alcohol use, and BMI in the associations of depression and anxiety with lung cancer
Indirect effect is calculated based on path a and path b; direct effect refers to path c′; total effect is calculated based on indirect and direct effects. These effects are Hazard Ratios.
Multiple mediator model includes cohorts with the four main health behaviors (smoking, physical inactivity, alcohol use, and BMI).
For smoking-related cancers, in single mediator models, smoking (HRs range 1.03–1.06) mediated the associations among depression (symptoms and diagnosis), anxiety (symptoms and diagnosis), and smoking-related cancers. In exploratory analyses, physical inactivity (HRs range 1.001–1.01), alcohol use (among drinkers; HRs range 1.004–1.02), and sedentary behavior (HRs range 1.002–1.01) also mediated the associations. Again, these results were similar in multiple mediator models (Table 4 and online Supplementary Table S6).
Table 4. Mediation of smoking, physical inactivity, alcohol use, and BMI in the associations of depression and anxiety with smoking-related cancers
Indirect effect is calculated based on path a and path b; direct effect refers to path c′; total effect is calculated based on indirect and direct effects. These effects are Hazard Ratios.
Multiple mediator model includes cohorts with the four main health behaviors (smoking, physical inactivity, alcohol use, and BMI).
Overall cancer and other types of cancer
In line with our hypotheses, smoking, physical inactivity, alcohol use, and a higher BMI mediated the associations among depression, anxiety, and overall cancer and colorectal cancer, and alcohol use mediated the associations among depression, anxiety, and alcohol-related cancers. However, except for smoking, the mediating effects of health behaviors were generally small (HRs below 1.01). Contrary to our hypotheses, health behaviors did not mediate the associations among depression, anxiety, and breast cancer and prostate cancer. Detailed results for these cancer outcomes are reported in online Supplementary text and Tables S7–S26.
Between-cohort heterogeneity
Between-cohort heterogeneity was high in the associations between depression/anxiety and health behaviors, with I 2 generally above 75% (substantial heterogeneity). Between-cohort heterogeneity was relatively low in the association between depression/anxiety and cancer and the association between health behaviors and cancer, with I 2 generally below 50% (moderate heterogeneity) (Tables 3 and 4; Online Supplementary Tables S5–S28).
Sensitivity analysis
Mediating effects of health behaviors operationalized as categorical variables were generally consistent with those of continuous variables reported above (online Supplementary Tables S5–S28).
Discussion
In this pooled analysis of 18 prospective cohort studies we found that (1) cigarette smoking mediated the associations among depression, anxiety, and lung cancer and smoking-related cancers; (2) physical inactivity and sedentary behavior also mediated these associations, but to a much lesser degree than smoking; (3) smoking, physical inactivity, alcohol use, and a higher BMI mediated the associations among depression, anxiety and overall cancer, colorectal cancer, and alcohol-related cancers, but except for smoking, the mediating effects of health behaviors were generally very small; (4) health behaviors did not seem to mediate the associations among depression, anxiety, and breast cancer and prostate cancer.
In line with our hypotheses, smoking mediated the associations among depression, anxiety, and lung cancer and smoking-related cancers. Specifically, depression/anxiety was associated with a 3–10% increased risk of these two outcomes, and this was through smoking. Our data also confirmed the hypotheses that physical inactivity mediated the associations among depression, anxiety, and lung cancer, although its mediating effects were much smaller than smoking. In fact, the mediating effects of smoking for lung cancer and smoking-related cancers were the largest among all health behaviors and cancer outcomes studied, which highlights its important mediating role in the associations.
Furthermore, exploratory analyses showed that physical inactivity and sedentary behavior simultaneously mediated the associations among depression, anxiety, lung cancer, and smoking-related cancers, suggesting that different mechanisms in the associations may exist between time spent sitting and not engaging in physical activity. Indeed, the pooled correlation between physical inactivity and sedentary behavior was low in our study (r = 0.17, 95% CI 0.05–0.27). Physical inactivity may increase the risk of lung cancer and smoking-related cancers through reduced pulmonary function, forced expiratory volume, and forced vital capacity which likely increase the duration of exposure to carcinogenic agents in the lungs (Cannioto et al., Reference Cannioto, Etter, LaMonte, Ray, Joseph, Al Qassim and Moysich2018; Garcia-Aymerich, Lange, Benet, Schnohr, & Antó, Reference Garcia-Aymerich, Lange, Benet, Schnohr and Antó2007). Although the underlying mechanisms are unclear, sedentary behavior reduces the activity of weight-bearing skeletal muscles, which may alter anti-cancer responses of myokines in the skeletal muscles and activate inflammatory pathways that are important for cancer development (Aoi et al., Reference Aoi, Naito, Takagi, Tanimura, Takanami, Kawai and Yoshikawa2013; Hojman et al., Reference Hojman, Dethlefsen, Brandt, Hansen, Pedersen and Pedersen2011).
We observed inconsistent mediation patterns regarding depression, BMI, and lung cancer in exploratory analyses, where a higher BMI was related to a lower risk of lung cancer, even after controlling for smoking. This negative association between BMI and lung cancer was also observed in previous meta-analyses (Duan et al., Reference Duan, Hu, Quan, Yi, Zhou, Yuan and Yang2015; Renehan, Tyson, Egger, Heller, & Zwahlen, Reference Renehan, Tyson, Egger, Heller and Zwahlen2008). Several mechanisms have been proposed, such as smoking being an explanatory factor or weight loss representing a preclinical event prior to lung cancer; however, these mechanisms were proven unaccountable for the inverse association between BMI and lung cancer (Abdel-Rahman, Reference Abdel-Rahman2019). A recent review summarized that several studies showed increased central adiposity associated with a higher risk of lung cancer and argued that body compositions assessed using anthropometric indicators or image-based techniques should be considered when estimating the risk of lung cancer (Vedire et al., Reference Vedire, Kalvapudi and Yendamuri2023). While few cohorts involved in our study have such information, further studies are warranted to investigate the mediating role of body compositions in the association between depression and lung cancer.
As mentioned before, we did not find an association of depression and anxiety with the risk of overall cancer, or colorectal or alcohol-related cancers in the previous study (van Tuijl et al., Reference van Tuijl, Basten, Pan, Vermeulen, Portengen, de Graeff and Ranchor2023), but nevertheless examined mediation for these outcomes for reasons outlined in the introduction. Our results generally confirmed the hypotheses for these outcomes. Exploratory analyses also revealed the mediating role of sedentary behavior in the associations among depression, anxiety and overall cancer and colorectal cancer. Notably, mediating effects of these health behaviors, except for smoking, were very small, and thus the clinical implications of these findings are likely to be limited.
Contrary to our hypothesis, we did not find consistent mediating effects of health behaviors in the associations among depression, anxiety, and breast cancer. Our findings may suggest that behavioral mechanisms play a lesser role in the associations among depression, anxiety, and breast cancer. On the other hand, it has been suggested that depression and anxiety may be related to lower estrogen levels (Wharton, Gleason, Olson, Carlsson, & Asthana, Reference Wharton, Gleason, Olson, Carlsson and Asthana2012), which may decrease the risk of breast cancer (Clemons & Goss, Reference Clemons and Goss2001). While testing the mediation by hormone in the associations among depression, anxiety and breast cancer was not planned in PSY-CA, we hypothesized that menopausal status may moderate the associations. However, our analyses based on PSY-CA study did not find menopausal status moderating the associations.
In addition, against our hypothesis, we found that individuals with depression or anxiety were more physically inactive, which in turn was related to a lower risk of prostate cancer in single mediator models, but not in multiple mediator models. However, we found that depressed individuals smoked more, which was related to a lower risk of prostate cancer in both single and multiple mediator models. Besides unmeasured confounding, a possible explanation could be that non-smokers are more likely to receive prostate-specific antigen screening compared with smokers. A similar explanation was posted by the authors of a previous meta-analysis, finding that lower physical activity was related to a lower risk of prostate cancer (Moore et al., Reference Moore, Lee, Weiderpass, Campbell, Sampson, Kitahara and Patel2016). The authors suggested that this positive association may be biased by screening behavior: physically active men are more likely to receive prostate-specific antigen screening than inactive men, which may increase the likelihood of diagnosing indolent prostate cancers (Moore et al., Reference Moore, Lee, Weiderpass, Campbell, Sampson, Kitahara and Patel2016). Another meta-analysis showed that smokers appeared to have an increased risk of developing aggressive prostate cancer than non-smokers (Foerster et al., Reference Foerster, Pozo, Abufaraj, Mari, Kimura, D'Andrea and Shariat2018). We were unable to further distinguish these mediating effects by prostate cancer stage or behavior due to the limited number of cohorts with such information.
Methodological strengths of the present study include the use of validated measures of depression or anxiety, the harmonization of data to reach conceptually similar variables, the use of the same statistical procedure across all cohorts, and the control of key confounders. In addition, the present study is among the largest to date which provided sufficient statistical power to extensively investigate mediation by several health behaviors in the associations among depression, anxiety, and cancer. A few limitations nevertheless need to be acknowledged. First, we used information on depression/anxiety and health behaviors collected at the same time, making their temporal order unclear. Indeed, bidirectional associations between depression/anxiety and health behaviors have been shown in previous studies (Azevedo Da Silva et al., Reference Azevedo Da Silva, Singh-Manoux, Brunner, Kaffashian, Shipley, Kivimäki and Nabi2012; Hiles et al., Reference Hiles, Lamers, Milaneschi and Penninx2017). While lagged data may have provided a more accurate representation of the association between depression/anxiety and health behaviors, such data were available in only a limited number of cohorts. Second, we did not consider different types (e.g. major depressive disorder or dysthymia in depression; generalized anxiety disorder or social anxiety in anxiety) or remission of depression and anxiety because most cohorts included did not have such information. Third, we were unable to directly assess between-cohort heterogeneity in indirect effects due to the limited methodology available for the meta-analysis of causal mediation analysis. We instead assessed path-specific heterogeneity and found high heterogeneity in the association between depression/anxiety and health behaviors. Fourth, availability of health behaviors, especially sedentary behavior and sleep, and the number of cancer cases differed across cohorts. As a result, the pooled indirect effect sizes may not be directly comparable because they may stem from meta-analyses of different numbers of cohorts. Fifth, the results of single and multiple mediator models were based on complete-case analyses as participants with missing values on health behaviors were excluded from the models. This means that there was reduction in the analytical sample from single to multiple mediator models. Although multiple imputation was considered to deal with missing values under the missing not at random assumption, developing cohort-specific multiple imputation models and running long scripts for 22 cohorts was considered unfeasible. However, since effect estimates were generally the same across single and multiple mediator models, it is unlikely that the reduction of sample size in multiple mediator models due to missing data in health behaviors accounted for the results. Finally, although we included a number of potential confounders, including cancer-specific ones, there might be unmeasured confounders that bias the mediation effect estimates reported in this paper.
In conclusion, smoking constitutes a prominent mediating pathway linking depression and anxiety to the risk of lung cancer and smoking-related cancers. Our findings underline the importance of smoking cessation interventions in clinical practice for persons with depression or anxiety (Gierisch, Bastian, Calhoun, McDuffie, & Williams, Reference Gierisch, Bastian, Calhoun, McDuffie and Williams2012; Rüther et al., Reference Rüther, Bobes, De Hert, Svensson, Mann, Batra and Möller2014; Taylor et al., Reference Taylor, Baker, Fox, Kessler, Aveyard and Munafò2021).
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291724000850.
Data availability statement
The data that support the findings of this study are owned by participating cohort studies. Data are not publicly available but may be shared upon reasonable request at each cohort depending on cohort-specific regulations.
Acknowledgements
The Trøndelag Health Study (HUNT) is a collaboration between HUNT Research Centre (Faculty of Medicine and Health Sciences, Norwegian University of Science and Technology NTNU), Trøndelag County Council, Central Norway Regional Health Authority, and the Norwegian Institute of Public Health. This research has been conducted using Atlantic PATH data under application 2019-103. The data used in this research were made available by the Atlantic Partnership for Tomorrow's Health (Atlantic PATH) study, which is the Atlantic Canada regional component of the Canadian Partnership for Tomorrow's Health funded by the Canadian Partnership Against Cancer and Health Canada. The views expressed herein represent the views of the authors and do not necessarily represent the views of Health Canada. The HELIUS study is conducted by the Amsterdam University Medical Centers, location AMC and the Public Health Service of Amsterdam. Both organizations provided core support for HELIUS. The HELIUS study is also funded by the Dutch Heart Foundation, the Netherlands Organization for Health Research and Development (ZonMw), the European Union (FP-7), and the European Fund for the Integration of non-EU immigrants (EIF). The Rotterdam Study is funded by Erasmus Medical Center and Erasmus University, Rotterdam, Netherlands Organization for the Health Research and Development (ZonMw), the Research Institute for Diseases in the Elderly (RIDE), the Ministry of Education, Culture and Science, the Ministry for Health, Welfare and Sports, the European Commission (DG XII), and the Municipality of Rotterdam. The authors are grateful to the study participants, the staff from the Rotterdam Study and the participating general practitioners and pharmacists. We thank the whole CARTaGENE team (https://cartagene.qc.ca/en/about), represented by authors NN and YP, for their contribution. Lifelines is a multi-disciplinary prospective population-based cohort study examining in a unique three-generation design the health and health-related behaviors of 167 729 persons living in the North of the Netherlands. It employs a broad range of investigative procedures in assessing the biomedical, socio-demographic, behavioral, physical, and psychological factors that contribute to the health and disease of the general population, with a special focus on multi-morbidity and complex genetics. The Lifelines initiative has been made possible by subsidy from the Dutch Ministry of Health, Welfare and Sport, the Dutch Ministry of Economic Affairs, the University Medical Center Groningen (UMCG), Groningen University and the Provinces in the North of the Netherlands (Drenthe, Friesland, Groningen). We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses. We gratefully acknowledge the contribution of the UCC-SMART research nurses; R. van Petersen (data-manager); A. Vandersteen (study manager) and the members of the Utrecht Cardiovascular Cohort-Second Manifestations of ARTerial disease-Studygroup (UCC-SMART-Study group): F.W. Asselbergs and H.M. Nathoe, Department of Cardiology; G.J. de Borst, Department of Vascular Surgery; M.L. Bots and M.I. Geerlings, Julius Center for Health Sciences and Primary Care; M.H. Emmelot-Vonk, Department of Geriatrics; P.A. de Jong and T. Leiner, Department of Radiology; A.T. Lely, Department of Gynecology and Obstetrics; N.P. van der Kaaij, Department of Cardiothoracic Surgery; L.J. Kappelle and Y.M. Ruigrok, Department of Neurology & Hypertension; M.C. Verhaar, Department of Nephrology & Hypertension, F.L.J. Visseren (chair), Department of Vascular Medicine, University Medical Center Utrecht and Utrecht University. The authors thank the participants of the OMEGA study, without whom this study would not have been possible. The authors thank the medical registries of the participating clinics for making patient selection possible, and all participating physicians for providing access to their patients' medical records.
Funding statement
KWF Dutch Cancer Society, Grant Number: VU2017-8288. The UK Medical Research Council and Wellcome Trust (ref: 217065/Z/19/Z) and the University of Bristol provide core support for ALSPAC. The linkage of ALSPAC to the cancer register was funded by Wellcome (ref:086118). The infrastructure of the NESDA study (www.nesda.nl) is funded through the Geestkracht program of the Netherlands Organization for Health Research and Development (Grant No. 10-000-1002) and financial contributions by participating universities and mental health care organizations (VU University Medical Center, GGZ inGeest, Leiden University Medical Center, Leiden University, GGZ Rivierduinen, University Medical Center Groningen, University of Groningen, Lentis, GGZ Friesland, GGZ Drenthe, Dimence, Rob Giel Onderzoekscentrum). ELSA is funded by the National Institute on Aging (R01AG017644) and the National Institute for Health and Care Research (198/1074-02).
Role of the funder/sponsor
The funding organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Disclaimer
The study has used data from the Cancer Registry of Norway. The interpretation and reporting of these data are the sole responsibility of the authors, and no endorsement by the Cancer Registry of Norway is intended nor should be inferred.
Competing interests
None.
Open access
