Cancer Causes Control DOI 10.1007/s10552-017-0872-4

ORIGINAL PAPER

Physical activity and lung cancer risk in men and women Vikki Ho1 · Marie‑Elise Parent2,3 · Javier Pintos3 · Michal Abrahamowicz4 · Coraline Danieli4 · Lesley Richardson3 · Robert Bourbonnais2 · Lise Gauvin1 · Jack Siemiatycki1 · Anita Koushik1 

Received: 10 August 2016 / Accepted: 14 February 2017 © Springer International Publishing Switzerland 2017

Abstract  Purpose  Although evidence has accumulated that recreational physical activities (PA) may reduce lung cancer risk, there is little evidence concerning the possible role of a potentially more important source of PA, namely occupational PA. We investigated both recreational and lifetime occupational PA in relation to lung cancer risk in a population-based case–control study in Montreal, Canada ­(NCASES = 727; ­NCONTROLS = 1,351). Methods  Unconditional logistic regression was used to estimate odds ratios (OR), separately for men and women, adjusting for smoking, exposure to occupational carcinogens, and sociodemographic and lifestyle factors. Results  In both sexes, increasing recreational PA was associated with a lower lung cancer risk ­(ORMEN = 0.66, 95% confidence interval (CI) 0.47–0.92; O ­ RWOMEN = 0.55, 95% CI 0.34–0.88, comparing the highest versus lowest tertiles). For occupational PA, no association was observed among women, while increasing occupational PA was Electronic supplementary material  The online version of this article (doi:10.1007/s10552-017-0872-4) contains supplementary material, which is available to authorized users. * Anita Koushik [email protected] 1

Department of Social and Preventive Medicine and CRCHUM (Centre de recherche du CHUM), Université de Montréal, 850 Saint‑Denis Street, Montréal, QC, Canada

2

INRS (Institut Armand-Frappier Research Center), Laval, QC, Canada

3

CRCHUM (Centre de recherche du CHUM), Montréal, QC, Canada

4

Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montréal, QC, Canada





associated with increased risk among men (­ORMEN = 1.96, 95% CI 1.27–3.01). ORs were not modified by occupational lung carcinogen exposure, body mass index, and smoking level; results were similar across lung cancer histological types. Conclusions  Our results support the previous findings for recreational PA and lung cancer risk. Unexpectedly, our findings suggest a positive association for occupational PA; this requires replication and more detailed investigation. Keywords  Case–control studies · Exercise · Histology · Lung neoplasms · Motor activity · Occupation · Recreation Abbreviations PA Physical activity OR Odds ratio CI Confidence interval MET Metabolic equivalent of tasks CSI Comprehensive smoking index SD Standard deviation DAG Directed acyclic graphs Smoking is the principal cause of lung cancer [1]; however, other factors influence risk as only 15% of smokers and also non-smokers develop this disease [2]. Evidence is accumulating that physical activity (PA) can influence cancer risk [3–5]. A recent meta-analysis indicates that regular recreational PA may be associated with a 24% reduced risk of lung cancer [6]. Several mechanisms have been proposed including reduced oxidative stress, decreased cell proliferation, modification of metabolic hormone levels via reduction of fat-produced estrogens, and reduced airway contact with carcinogens through increased ventilation [3, 5, 7].

13

Vol.:(0123456789)



Cancer Causes Control

Most studies investigating the PA–lung cancer relationship have focused on recreational PA. However, PA is also derived from occupation, household, and transportation-related activities [8], among which perhaps the most important is occupational PA, as most people spend many hours at work, and some jobs entail considerable PA [8]. Few studies have examined the role of occupational PA on lung cancer risk [9–17], and findings have been inconsistent and inconclusive. Counterintuitively, some studies have reported higher risks associated with higher occupational PA [9, 11, 15]. Previous studies typically utilized crude assessments of occupational PA, such as that based on selfreported cross-sectional assessments or job title-based categorizations [9–16]. As well, residual confounding by smoking is possible [15, 16, 18–20] and few considered potential confounding by occupational carcinogens [9, 12, 17]. The occupational PA–lung cancer relationship has primarily been investigated in men only [12, 13, 15–17], and most studies did not assess possible differences by histological types. We conducted one of the largest case–control studies of lung cancer among men and women that collected information on lifetime occupational history and participation in recreational activities. This investigation assessed the impact of these two sources of PA, independently and jointly, on the risk of lung cancer overall and its major histological types among men and women.

sparsely reported by proxy respondents. Similarly, reporting recreational PA 20 years prior may have been difficult for a proxy respondent. To minimize the potential for exposure misclassification, the present analysis was restricted to participants who self-responded to the interview ­(NCases = 442 men, 285 women; N ­ Controls = 805 men, 546 women).

Methods

Average recreational PA (MET-h/week) Month Session Hours ∑ ( Year )( Month )( Session )(METActivity )

Recreational and occupational PA conceptualization Participation in recreational activities was solicited for a one-time period 20 years prior to diagnosis/interview; we assumed that activity during this period characterized the relative PA level during a part of adult life that may be relevant to lung cancer initiation [22]. Twelve activities were assessed including walking, jogging or running, gardening, home exercises or exercise class, golf, racquet sports, bowling or curling, swimming or water exercises, skiing or skating, bicycling, social dancing or other strenuous exercise. Energy expenditure in units of metabolic equivalent of tasks (METs) was assigned to each activity using the Compendium of Physical Activities [23]. A MET is defined as the ratio of the metabolic rate associated with performing an activity as compared to the resting metabolic rate [24]. For each participant, we calculated average recreational PA in units of MET-hours per week (MET-h/week) using the following equation:

52 Week∕Year

Study overview A population-based case–control study was conducted in Montreal, Canada from 1996 to 2001 [21]. Histologically confirmed cases of lung cancer were identified at 18 Montreal-area hospitals; the participation rate among cases was 84% (N = 709 men, 422 women). Controls were randomly selected using the Provincial Electoral List, and were frequency-matched to cases on 5-year age group and sex; the participation rate among controls was 69% (N = 889 men, 564 women). In-person interviews were conducted in two parts. The first was a structured section assessing sociodemographic and lifestyle characteristics, including smoking, diet, alcohol intake, and recreational activities conducted 20 years prior. The second was a semi-structured section assessing job history, during which participants indicated their main and subsidiary tasks in each job. For participants that had died, were too ill, or had difficulties with communication, interviews were conducted with their next of kin. Occupational PA was determined using job-related task information (see below), which was

13

=

Total # of Activities where month/year for a seasonal activity was given a value of three/season; session/month represented the number of sessions/month; h/session represented the number of h/session; and ­METActivity represented the MET value for each activity. For example, a person who walked for exercise year-round for 10 sessions/month at 1 h/session, at a MET level of 3.4 (7.84 MET-h/week accumulated), and gardened for 9 month/year for 10 sessions/month at 1 h/session, at a MET level of 3.98 (6.89 MET-h/week accumulated), would have an average recreational PA of 7.37 MET-h/week ((7.84 MET-h/week + 6.89 MET-h/week)/2 activities). To determine occupational PA, participants’ job histories were evaluated by “experts,” specifically an industrial hygienist and an exercise physiologist. For each participant, the experts evaluated each main job-related task and assigned energy expenditure in METs using the Compendium as a reference [23]. The overall MET value for a given job (­METJob) was then calculated by weighting the MET values of each job-related task by the percentage of day spent in that task. Examples of occupations with low

Cancer Causes Control

­METJob values (<1.5) included drivers, accountants, and occupations related to management and administration, whereas occupations with high M ­ ETJob values (>10.0) included loggers, foremen, and firefighters. For each participant, we then calculated the weighted average lifetime occupational PA in units of MET-h/week using the following equation:

summary categorical variable that defined participants as ‘Never exposed,’ ‘Not Substantially Exposed,’ and ‘Substantially Exposed’ to any of the listed lung carcinogens. Family income, based on the median family income level for the census tract the participant lived in at the time of interview, was represented by tertiles. Categorical variables were used to represent ethnicity and years of schooling. �

4.33Weeks ((YearFinish −YearStart )+1) ( Months Year ) ( Month )

Weighted average lifetime occupational PA (MET-h/week) =

where ­YearStart and Y ­ earFinish indicated the year each job began and ended; months/year for a seasonal job was given a value of three for each season worked; weeks/month was assigned as a constant of 4.33; days/week represented the number of days/week worked; h/day represented the number of h/day worked; and M ­ ETJob represented the average MET value for each job based on the evaluation by the experts. For example, a person working year-round as a kitchen helper ­(METJob = 3) for 3 years at 5 days/week and 8.5  h/day (382.21 MET-h/week accumulated) and then year-round as a garment presser ­(METJob=3) for 40 years at 5 days/week and 8 h/day (4796.31 MET-h/week accumulated) would have an average lifetime occupational PA of 120.43 MET-h/week. Occupational PA before 18 years of age was excluded as jobs held during this period were not consistently reported; as well, occupational PA during the 2 years prior to diagnosis (or interview date for controls) was excluded to minimize the risk of reverse causality bias. For occupations listed as ‘homemaker,’ ‘student,’ ‘volunteer,’ and ‘short job,’ tasks were not assessed and MET values were not assigned; these job periods were excluded from the calculation of average lifetime occupational PA. Variable definitions Recreational and occupational PA were parameterized as both categorical and continuous variables. Categorical PA variables were created using sex-specific tertiles among controls. Age was represented by a continuous variable. Lifetime smoking history was represented by the comprehensive smoking index (CSI), a continuous measure that incorporates smoking status, duration, time since cessation, and intensity into one aggregate measure, and has previously been adapted for use in this study [25]. Exposure to any of the 10 occupational lung carcinogens (namely, crystalline silica, asbestos, nickel, soot, chromium VI, cadmium, diesel engine emissions, benzo[a]pyrene, inorganic acids, and coal combustion products [26]), assessed using the expert-based approach [21], was represented as one

∑

�

Days Week

��

Hours Day

�

� (METJob)

52 Weeks∕Year

Total # of Working Years Vegetable and fruit intake, assessed using a semi-quantitative food frequency questionnaire, were represented as continuous variables. Statistical analysis Multivariable unconditional logistic regression was used to estimate adjusted odds ratios (OR) with 95% confidence intervals (CI) for the associations of recreational and occupational PA with lung cancer risk, separately for men and women. Minimally adjusted models included age and the continuous CSI as covariates. Fully adjusted models additionally included a priori selected determinants of lung cancer in our study population, namely, ethnicity, years of schooling, family income, and occupational lung carcinogen exposure (for men only, since few women were exposed). We also evaluated other potential confounders using a novel approach that combines directed acyclic graphs (DAGs) and the change-in-estimate procedure [27] (Appendix  1). For men, no additional confounders were identified; for women, fruit intake and vegetable intake were identified as an additional confounder in the recreational PA and occupational PA models, respectively, and were included in the final multivariable models. Recreational PA was assessed 20 years prior to diagnosis/interview, while occupational PA was assessed throughout life, which precluded our ability to combine them into a single variable. To assess how the estimated lung cancer risk varied depending on the joint distribution of recreational and occupational PA, we estimated the ORs for different combinations of the two categorical PA measures. For categorical PA variables, the p-value for trend across categories was obtained by modeling the median value of categories as a continuous variable and assessing the Wald Chi-Square test statistic for this variable. For continuous PA variables, we first assessed whether the associations were consistent with the linearity assumption using the flexible multivariable fractional polynomials approach by Royston and Sauerbrei [28]. A linear representation was used if the non-linear effects were statistically

13



Cancer Causes Control

non-significant (p > 0.05 for the 2-degree of freedom likelihood ratio test) [29]. We evaluated modification of the PA–lung cancer ORs by occupational lung carcinogen exposure, body mass index (BMI), and smoking status, by testing two-way multiplicative interactions, separately in the regression models. In these analyses, occupational lung carcinogen exposure was dichotomized to ‘Never Exposed’ versus ‘Exposed,’ self-reported BMI two years prior to the interview was dichotomized at 24.9  kg/m2 [30], and smoking status was categorized as Never-smokers, ‘Lighter’ smokers, and ‘Heavier’ smokers, where categories of lighter versus heavier smokers were based on dichotomization at the median CSI value among all smokers. Associations by lung cancer histology (adenocarcinoma, squamous cell carcinoma, and small cell carcinoma) were examined using polytomous logistic regression. In sensitivity analyses, we assessed the impact of excluding proxy respondents. We also adjusted for additional smoking variables (CSI, smoking status, average amount of cigarettes smoked/day, years of smoking, and time since cessation) to assess possible residual confounding by smoking. To assess possible residual confounding by occupational exposure to lung carcinogens, we further Table 1  Characteristics of the study population

adjusted for the number of carcinogens substantially or not substantially exposed to. We also excluded information 5 years prior to date of diagnosis/interview in the calculation of average lifetime occupational PA to further minimize the potential for reverse causality bias.

Results This study included 727 self-responding incident lung cancer cases and 1,351 controls. We excluded 506 participants who used a proxy respondent ­(NMen = 351, ­NWomen = 155). In general, cases and controls who used a proxy respondent were more likely to currently smoke, to have fewer years of education and to have a lower income compared to self-responders (not shown). Among self-respondents, compared to controls, cases were more likely to currently smoke, have a lower income and education level, be of French ancestry, have a family history of lung cancer, and have lower intakes of fruit and vegetables (Table  1). No major differences were observed between men and women, except that male cases were more likely to be substantially exposed to occupational lung carcinogens than controls, while no difference was observed among women. In the

Selected characteristics

Mean age, years Smoking status, %  Never  Former  Current Median household income  Tertile 1  Tertile 2  Tertile 3 Highest education level: > high school, % French ancestry, % Mean BMI, kg/m2 Family history of lung cancer, % Mean vegetable intake, servings/week Mean fruit intake, servings/week Occupational lung carcinogens, %a Unexposed Not substantially exposed substantially exposed a

Cases (N = 727)

Controls (N = 1351)

Men

Women

Men

Women

(n = 442)

(n = 285)

(n = 805)

(n = 546)

63.6

61.0

64.9

61.2

2.9 33.0 64.1

6.7 21.0 72.3

17.8 53.5 28.7

49.3 28.6 22.1

37.8 33.0 29.2 13.5 76.5 25.4 22.9 16.4 13.4

43.5 31.9 24.6 13.9 76.1 24.4 27.0 16.0 12.0

29.8 35.0 35.2 24.5 64.6 25.9 11.7 19.9 18.7

27.5 35.0 37.5 32.0 72.3 25.3 20.0 21.6 18.5

27.4 41.9 30.7

79.7 19.6 0.7

45.1 35.4 19.5

82.8 15.9 1.3

 Occupational exposure to lung carcinogens was conceptualized as a three-level categorical variable. To be classified as exposed at a substantial level, a subject had to have had been exposed at confidence of probable or definite, concentration and frequency of medium or high and for duration >5 years. All other exposed subjects were then classified in the not substantial category

13

Cancer Causes Control

overall study population, the mean number of years worked was 32.0 (standard deviation = 12.4) and the mean number of jobs held was four jobs; on average, each job was held for 9 years (standard deviation = 9.1). To illustrate the variability of occupational MET values in the study population, over 85% of jobs were attributed a single MET value; 12% were attributed two different MET values, and 3% of jobs were assigned three to four different MET values. For jobs with several MET values assigned, the magnitude of the difference between the lowest and the highest value could be quite important: the difference was <1.0 METs for 54% of jobs, between 1.0 and <3.5 METs for 34% of jobs and for 12% of jobs, the difference was 3.5 METs or more. The mean levels of recreational PA among cases and controls in MET-h/week were 2.28 (SD = 2.74) and 2.96 (SD = 2.90), respectively, in men and 2.28 (SD = 2.24) and 3.47 (SD = 2.38), respectively, in women. The mean levels of occupational PA (MET-h/week) were much higher, at 154.44 (SD = 69.40) and 134.48 (SD = 63.58) for cases and controls, respectively, in men and 89.33 (SD = 31.75) and 83.57 (SD = 28.84) for cases and controls, respectively, in women. The difference observed between the mean levels of recreational versus occupational PA mainly reflects the greater proportion of time spent at work versus participating in recreational activities. Similar differences have been reported previously [8]. Table  2 presents the lung cancer risks associated with recreational and occupational PA. In both men and women, patterns of associations observed in the age- and smokingadjusted models were similar to those in the multivariable models. For both sexes, high recreational PA was associated with a reduced lung cancer risk (Table 2). In contrast, high occupational PA was associated with a significantly increased risk of lung cancer among males, with no association among women. When the different combinations of the tertiles of recreational and occupational PA were examined jointly, for both sexes, the general patterns observed in Table  2 were replicated in each stratum of the other variable, with slight variations reflecting mostly reduced statistical power (Table 3). In a sensitivity analysis that included all original case–control participants (i.e., proxy respondents also), the direction of associations for both recreational and occupational PA did not change though the ORs were slightly attenuated. Specifically, the OR (95% CI) for high versus low recreational PA was 0.52 (95% CI 033–0.88) for women and 0.77 (95% CI 0.56–1.04) for men, while for high versus low occupational PA was 0.99 95% CI (0.62–1.58) in women and 1.57 (95% CI 1.07–2.29) in men. In addition, sensitivity analyses with further adjustment for other smoking variables, with a more detailed adjustment for occupational lung carcinogen exposure, and

with occupational PA calculated by excluding information from the 5 years preceding diagnosis/interview, the associations did not differ appreciably from those presented in Tables 2 and 3 (results not shown). The multivariable fractional polynomial analyses indicated no deviations from linearity (all p > 0.05). When PA was modeled continuously results were consistent with the categorical analyses as shown in Table 2. For recreational PA, the multivariable-adjusted ORs for lung cancer associated with a 1 standard deviation (SD) increase was 0.86 (95% CI 0.74, 0.99; SD = 2.86) in men and 0.73 (95% CI 0.60, 0.89; SD = 2.40) in women. For occupational PA, the ORs were 1.21 per 1 SD (95% CI 1.03, 1.41; SD = 66.37) in men and 1.09 per 1 SD (95% CI 0.88, 1.34; SD = 29.98) in women. To contextualize the meaning of 1 SD, the SD for recreational PA in men was 2.86 MET-h/week which is comparable to contrasting two persons who participated in activities with a MET value of 7.8 (e.g., boxing) versus 5.0 (e.g., rock climbing) for, on average, 1  h/week. For occupational PA in men, a SD of 66.37 MET-h/week is comparable to contrasting two persons who worked jobs with an average MET value of 3.5 (e.g., plumber) versus 2.0 (e.g., delivery person), for the same duration of 35 h/week. ORs for the associations of both recreational and occupational PA with lung cancer risk were not significantly modified by occupational lung carcinogen exposure, BMI, or smoking history in both sexes (p > 0.20 for all multiplicative interactions in Table  4). Associations for the main histologic types (Table  5) did not differ greatly from that observed for all lung cancers combined.

Discussion In one of the largest lung cancer case–control studies, the role of participation in recreational PA and occupational PA across the entire working life was examined. Increasing recreational PA was inversely associated with lung cancer risk. Conversely, in men, an increased lung cancer risk was observed with increasing occupational PA. These relationships did not differ when examining recreational and occupational PA jointly. Also, associations did not vary by occupational lung carcinogen exposure, BMI, and smoking status. In general, the relationships between recreational and occupational PA with histologic types of lung cancer did not greatly differ from what was seen for all types combined. The association between recreational PA and lung cancer risk has previously been examined in 28 studies and recently reviewed [6]. Overall, a consistent inverse association between recreational PA and lung cancer risk has been observed across studies with a pooled relative risk of 0.76 (95% CI 0.69–0.85); no sex difference was found

13

Table 2  Odds ratios (95% confidence intervals) of lung cancer associated with recreational and occupational physical activity in men and women

Cancer Causes Control

Men  Recreational ­PAc   Low   Intermediate   High   p-value for ­trendd  Occupational ­PAc   Low   Intermediate   High   p-value for ­trendd Women  Recreational ­PAc   Low   Intermediate   High   p-value for ­trendd Occupational ­PAc   Low   Intermediate   High   p-value for ­trendd

NCases

NControls

Adjusted for age and smoking OR (95% CI)

Multivariable-adjusteda,b OR (95% CI)

189 150 103

266 274 265

1.00 (Ref) 0.94 (0.69, 1.28) 0.70 (0.51, 0.97) 0.09

1.00 (Ref) 0.88 (0.64, 1.21) 0.66 (0.47, 0.92) 0.05

75 162 205

265 275 265

1.00 (Ref) 1.65 (1.16, 2.34) 2.05 (1.45, 2.91) <0.01

1.00 (Ref) 1.60 (1.09, 2.35) 1.96 (1.27, 3.01) <0.01

159 72 54

180 185 181

1.00 (Ref) 0.58 (0.38, 0.89) 0.46 (0.29, 0.71) <0.01

1.00 (Ref) 0.63 (0.40, 0.99) 0.55 (0.34, 0.88) 0.02

86 72 127

181 185 180

1.00 (Ref) 0.65 (0.41, 1.02) 0.97 (0.63, 1.49) 0.10

1.00 (Ref) 0.70 (0.43, 1.13) 0.94 (0.58, 1.53) 0.28

In women for occupational PA: adjusted for age (continuous), ethnicity (French, English and Other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous), and vegetable intake (continuous)

a

 In men: adjusted for age (continuous), ethnicity (French, English and Other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous) and occupational carcinogen exposure (substantial, not substantial and never)

b

 In women for recreational PA: adjusted for age (continuous), ethnicity (French, English and Other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous), and fruit intake (continuous)

c

 Categorization of PA in average MET-h per week per year according to sex-specific tertiles among controls. For recreational PA in males: low (<0.61), intermediate (0.61–<4.05), and high (>4.05). For occupational PA in men: low (<99.51), intermediate (99.51–<145.63), and high (>145.63). For recreational PA in women: low (<2.36), intermediate (2.36–<4.42), and high (>4.42). For occupational PA in women: low (<67.78), intermediate (67.78–<92.13), and high (>92.13)

d

 Trend tests were performed by assigning the median of each category as scores and computed by adding the continuous variable to the logistic model predicting lung cancer risk

[6]. Among the six studies that examined the relationship by histologic type [9, 14, 18, 31–33], some studies reported statistically significant inverse associations for adenocarcinoma [18, 31], squamous cell carcinoma [18, 32], and small cell carcinoma [14, 31, 32]. Occupational PA in relation to lung cancer risk has been examined in six cohorts [9–14] and three case–control studies [15–17], among which only one study, conducted by our group, examined lifetime PA and reported a nonsignificant inverse association [17]. Statistically significant associations were observed in three studies, where, consistent with our findings among men, increased risks of lung

13

cancer were associated with standing versus sitting jobs in two cohort studies [9, 11] or increasing occupational PA levels in a case–control study [15]. Only three studies [9, 11, 14] examined the occupational PA–lung cancer relationship by sex, two of which reported evidence of an increased risk associated with standing jobs among men only [9] and in both men and women [11], while no association was reported in the third study [14]. In the only study that evaluated associations by histologic type, there was no evidence of differences by type [12]. In our analysis, higher levels of recreational PA were associated with a decreased risk of lung cancer, similar

Cancer Causes Control Table 3  Odds ratios (95% confidence intervals) of lung cancer associated with combined occupational and recreational physical activity in men and women NCASES/NCONTROLS Multivariable-adjusteda, b OR (95% CI) Men Recreational PA l­owc Recreational PA ­intermediatec Recreational PA h­ ighc Women Recreational PA l­owc Recreational PA ­intermediatec Recreational PA H ­ ighc

Occupational ­PAc Low

Intermediate High

28/69 1.00 (Ref) 27/97 0.77 (0.39, 1.53) 20/99 0.70 (0.34, 1.43)

66/89 1.69 (0.91, 3.11) 55/94 1.25 (0.67, 2.34) 41/92 1.04 (0.55, 1.99)

95/108 1.74 (0.94, 3.23) 68/83 1.97 (1.03, 3.77) 42/74 1.15 (0.58, 2.28)

46/47 1.00 (Ref) 23/68 0.54 (0.25, 1.19) 17/66 0.53 (0.23, 1.24)

37/61 0.54 (0.26, 1.13) 20/67 0.42 (0.19, 0.93) 15/57 0.50 (0.21, 1.19)

76/72 0.87 (0.43, 1.74) 29/50 0.68 (0.30, 1.52) 22/58 0.37 (0.16, 0.82)

a

 In men: adjusted for age (continuous), ethnicity (French, English and Other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous) and occupational carcinogen exposure (substantial, not substantial and never)

b  In women: adjusted for age (continuous), ethnicity (French, English and Other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous) and fruit intake (continuous), and vegetable intake (continuous) c

 Categorization of PA in average MET-h per week per year according to sex-specific tertiles among controls. For recreational PA in males: low (<0.61), intermediate (0.61–<4.05) and high (>4.05). For occupational PA in men: low (<99.51), intermediate (99.51–<145.63), and high (>145.63). For recreational PA in women: low (<2.36), intermediate (2.36–<4.42), and high (>4.42). For occupational PA in women: low (<67.78), intermediate (67.78–<92.13), and high (>92.13)

to the previous reports. We hypothesized a similar inverse association with lifetime occupational PA; however, in men, we observed a positive association with lung cancer risk, consistent with a few previous studies [9, 11, 15]. The observed positive association could represent a chance finding. Interestingly, research on cardiovascular disease has analogously shown that recreational PA decreases risk, while occupational PA increases risk [34]. This pattern of contrasting associations could be attributed to differences in the nature of recreational versus occupational PA. For instance, recreational PA usually takes place over shorter periods and on non-consecutive days, while occupational PA occurs repetitively on a daily basis for several hours which may limit recovery

time [35]. Although PA in general has been associated with lower levels of inflammation, lack of recovery from PA may aggravate levels of inflammation [36, 37] which could contribute to cancer development [38, 39]. PA may modify the patterns of respiratory airflow, for instance, by encouraging mouth breathing [40]. In terms of occupational PA, physically demanding jobs are generally conducted in surroundings that may contain hazardous exposures and chemicals [9]. Thus, modified breathing patterns could increase the deposition and impaction of carcinogenic particles. In our study, we measured and controlled for exposure to occupational lung carcinogens with arguably the most valid method used in any analysis of occupational PA and lung cancer. Nonetheless, there may be other unrecognized lung carcinogens present in the workplace, thereby limiting our ability to fully control for occupational carcinogen exposure. We controlled for established lung cancer risk factors and identified additional confounders using DAGs and the change-in-estimate approach. A major strength of this study was our adjustment of smoking based on detailed smoking history. We used various measures, including smoking status, duration of smoking, time since cessation of smoking, and smoking intensity, incorporated into one parsimonious measure. In sensitivity analysis, we further adjusted for additional measures of smoking. Nevertheless, some degree of residual confounding is still plausible. Another strength of our study is our measure of lifetime occupational PA, which was determined by experts in industrial hygiene and exercise physiology blinded to case/ control status, who used participant lifetime occupational histories that included specific task information for each job. We believe our method greatly improves upon previous studies that simply used job titles without task information, or relied on self-reported PA levels, both of which would contribute to information bias. Furthermore, many previous studies based their occupational PA measure on one time point, usually current job [9, 15, 16] or job held in the past year [11, 14] at the time of study participation. Indeed, our study is among the first to have examined lifetime occupational PA which is more likely to better capture the period of etiologic relevance [17]. To further reduce the potential for information bias, we restricted the study population to self-respondents, who provided more complete information on their job-related tasks, and who presumably reported their recreational activity 20 years prior with less error than a proxy respondent. However, although having reduced the potential for information bias, it is possible that our findings reflect associations for less severe lung cancer, if proxy respondents were mostly used by more advanced cases. We had no information on lung cancer stage at diagnosis.

13

Table 4  Odds ratios (95% confidence intervals) of lung cancer risk associated with occupational physical activity in men and women stratified by body mass index, exposure to occupational lung carcinogens, and smoking history

Cancer Causes Control NCases

Men  Occupational lung carcinogen exposure   Never 121   Ever 321  Body mass index   Under and normal weight 356   Overweight and obese 86  Smoking ­historye   Never smoker 13   Light smoker 118   Heavy smoker 311 Women  Occupational lung carcinogen exposure   Never 227   Ever 58  Body mass index   Under and normal weight 264   Overweight and obese 21  Smoking ­historye   Never smoker 19   Light smoker 80   Heavy smoker 186

NControls

Recreational PA Per 1 S ­ Da OR (95% CI)b, c

pd

Occupational PA Per 1 S ­ Da OR (95% CI)b, c

pd

326 479

0.92 (0.73, 1.17) 0.84 (0.70, 0.99)

0.51

1.20 (0.80, 1.81) 1.23 (1.04, 1.45)

0.92

664 141

0.88 (0.75, 1.03) 0.77 (0.55, 1.08)

0.47

1.18 (0.98, 1.41) 1.25 (0.95, 1.65)

0.20

145 427 233

0.76 (0.39, 1.46) 0.84 (0.67, 1.06) 0.88 (0.73, 1.06)

0.90

0.70 (0.34, 1.45) 1.20 (0.94, 1.54) 1.25 (1.03, 1.53)

0.31

452 94

0.71 (0.57, 0.88) 0.80 (0.52, 1.22)

0.62

1.09 (0.85, 1.40) 1.00 (0.68, 1.48)

0.71

504 42

0.71 (0.58, 0.88) 0.86 (0.51, 1.46)

0.52

1.09 (0.88, 1.35) 1.03 (0.43, 2.46)

0.73

269 192 85

0.62 (0.36, 1.07) 0.73 (0.53, 1.01) 0.76 (0.58, 1.00)

0.82

1.23 (0.85, 1.78) 1.25 (0.86, 1.83) 0.94 (0.70, 1.25)

0.34

In women for occupational PA: adjusted for age (continuous), ethnicity (French, English and other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous), and vegetable intake (continuous)

a

 In men: SD of recreational PA was 2.86 and SD of occupational PA was 66.37. In women: SD of recreational PA was 2.40 and SD of occupational PA was 29.98

b

 In men: adjusted for age (continuous), ethnicity (French, English and Other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous) and occupational carcinogen exposure (substantial, not substantial and never)

c

 In women for recreational PA: adjusted for age (continuous), ethnicity (French, English and Other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous), and fruit intake (continuous)

d e

 p-value for interaction term

 The categories of “Lighter” versus “Heavier” smokers were based on dichotomizing at the median CSI value among all smokers in the study

We assessed recreational PA for a period of 20 years prior to the date of diagnosis/interview, as this was believed to correspond to a relevant biological time window for cancer initiation. Although measuring PA for only a single moment might lead to exposure misclassification particularly for individuals whose behaviors have changed throughout time, it can be speculated from studies that assessed recreational PA at several lifetime periods that a high proportion of individuals follow a rather constant PA pattern [41, 42]. On the other hand, despite a relatively high participation rate even among controls, it is possible that the inverse association between recreational PA and lung cancer risk may reflect selection bias resulting from

13

the possibility that controls that participated had healthier lifestyles, such as participating in recreational PA, than the underlying population. Furthermore, we restricted the study population to those who self-responded to the interview to reduce the potential for exposure misclassification. As cases were more likely to use a proxy respondent than controls, the restriction to self-respondents may further contribute to selection bias if those who self-responded differed from those who used a proxy respondent on the basis of occupational PA. However, sensitivity analysis including proxy responders revealed similar associations to those found in analyses based on restricting to self-respondents only.

Cancer Causes Control Table 5  Odds ratios (95% confidence intervals) of histological types of lung cancer associated with recreational and occupational physical activity Multivariable-Adjusted OR (95% CI)a,b NCases/NControls Men  Recreational PA 152/805   Per 1 S ­ Dc  Occupational PA 152/805   Per 1 S ­ Dc Women  Recreational PA 151/546   Per 1 S ­ Dc  Occupational PA 151/546   Per 1 S ­ Dc

Adenocarcinoma

NCases/NControls

Squamous cell carcinoma

NCases/NControls

Small cell carcinoma

0.90 (0.74, 1.09)

173/805

0.78 (0.64, 0.95)

65/805

0.89 (0.66, 1.19)

1.25 (1.01, 1.54)

173/805

1.10 (0.90, 1.36)

65/805

1.31 (0.98, 1.74)

0.81 (0.65, 1.01)

51/564

0.59 (0.41, 0.86)

39/564

0.70 (0.47, 1.05)

1.07 (0.83, 1.36)

51/564

1.26 (0.89, 1.79)

39/564

1.37 (0.94, 1.99)

In women for occupational PA: adjusted for age (continuous), ethnicity (French, English and Other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous), and vegetable intake (continuous) a

 In men: adjusted for age (continuous), ethnicity (French, English and Other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous) and occupational carcinogen exposure (substantial, not substantial and never)

b

 In women for recreational PA: adjusted for age (continuous), ethnicity (French, English and other), number of years of schooling (categorical: <7, 7–<12, 12+), income (tertiles among controls), CSI (continuous), and fruit intake (continuous) c

 In men: SD of recreational PA was 2.86 and SD of occupational PA was 66.37. In women: SD of recreational PA was 2.40 and SD of occupational PA was 29.98

In summary, our results support previous research suggesting a lower risk of lung cancer with increasing recreational PA. We observed not only a positive association with occupational PA for men, which may be due to chance or to residual or uncontrolled confounding, but may also reflect the different nature of occupational versus recreational PA which can be explored in further research, particularly among never-smokers. Although smoking remains the most important modifiable risk factor for lung cancer, the difficulties in smoking cessation warrant the identification of other modifiable factors, such as PA, with a goal towards prevention. Acknowledgments  This work was supported by the Canadian Cancer Society Research Institute (Grant #19912). Dr. Ho received a postdoctoral fellowship from the Canadian Institutes of Health Research and Lung Cancer Canada to conduct this work and is currently supported by the Cancer Research Society, Fonds de recherche du Québec–Santé (FRQS) and Ministère de l’Économie, de la Science et de l’Innovation du Québec (MESI). Dr. Parent received Career Investigator Awards from the FRQS. Dr. Abrahamowicz is a James McGill Professor. Dr. Siemiatycki holds the Guzzo Chair in Environment and Cancer. Dr. Koushik was supported by a New Investigator award of the Canadian Institutes of Health Research. Author contributions  All authors contributed to this work. Compliance with ethical standards  Conflict of interest  All authors declare that they have no conflict of interests.

Research involving human rights  All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Informed consent  Informed consent was obtained from all individual participants included in the study.

References 1. Parkin DM, Sasco AJ (1993) Lung cancer: worldwide variation in occurrence and proportion attributable to tobacco use. Lung Cancer 9:1–16 2. Abraham J, Gulley JL, Allegra CJ (2005) Bethesda handbook of clinical oncology. Lippincott Williams & Wilkins, Philadelphia 3. Friedenreich CM, MR Orenstein, (2002) Physical activity and cancer prevention: etiologic evidence and biological mechanisms. J Nutr 132(11 Suppl):3456S-3464S 4. Lee IM (2003) Physical activity and cancer prevention–data from epidemiologic studies. Med Sci Sports Exerc 35(11):1823–1827 5. Kruk J, Aboul-Enein HY (2006) Physical activity in the prevention of cancer. Asian Pac J Cancer Prev 7(1):11–21 6. Brenner DR et al (2016) Leisure-time physical activity and lung cancer risk: A systematic review and meta-analysis. Lung Cancer 95:17–27 7. Quadrilatero J, Hoffman-Goetz L (2003) Physical activity and colon cancer. A systematic review of potential mechanisms. J Sports Med Phys Fit 43(2):121–138 8. Csizmadi I et  al (2011) Hours spent and energy expended in physical activity domains: results from the tomorrow project cohort in Alberta, Canada. Int J Behav Nutr Phys Act 8:110

13

9. Steindorf K et al (2006) Physical activity and lung cancer risk in the European prospective investigation into cancer and nutrition cohort. Int J Cancer 119(10):2389–2397 10. Rundle A et al (2010) Physical activity and lung cancer among non-smokers: a pilot molecular epidemiological study within EPIC. Biomarkers 15(1):20–30 11. Bak H et al (2005) Physical activity and risk for lung cancer in a Danish cohort. Int J Cancer 116(3):439–445 12. Colbert LH et al (2002) Physical activity and lung cancer risk in male smokers. Int J Cancer 98(5):770–773 13. Severson RK et  al (1989) A prospective analysis of physical activity and cancer. Am J Epidemiol 130(3):522–529 14. Thune I, Lund E (1997) The influence of physical activity on lung-cancer risk: a prospective study of 81,516 men and women. Int J Cancer 70(1):57–62 15. Brownson RC et al (1991) Physical activity on the job and cancer in Missouri. Am J Public Health 81(5):639–642 16. Dosemeci M et al (1993) Occupational physical activity, socioeconomic status, and risks of 15 cancer sites in Turkey. Cancer Causes Control 4(4):313–321 17. Parent ME et  al (2011) Occupational and recreational physical activity during adult life and the risk of cancer among men. Cancer Epidemiol 35(2):151–159 18. Leitzmann MF et al (2009) Prospective study of physical activity and lung cancer by histologic type in current, former, and never smokers. Am J Epidemiol. doi:10.1093/aje/kwn371 19. Steenland K, Nowlin S, Palu S (1995) Cancer incidence in the National Health and Nutrition Survey I. Follow-up data: diabetes, cholesterol, pulse and physical activity. Cancer Epidemiol Biomark Prev 4(8):807–811 20. Yun YH et  al (2008) Dietary preference, physical activity, and cancer risk in men: national health insurance corporation study. BMC Cancer 8:366 21. Parent ME et  al (2007) Exposure to diesel and gasoline engine emissions and the risk of lung cancer. Am J Epidemiol 165(1):53–62 22. Harrison RM, Smith DJ, Kibble AJ (2004) What is respon sible for the carcinogenicity of PM2.5? Occup Environ Med 61(10):799–805 23. Ainsworth BE et  al (2000) Compendium of physical activities: an update of activity codes and MET intensities. Med Sci Sports Exerc 32(9 Suppl):S498–S504 24. Anshell M, Freedson P, Haywood K (1991) Dictionary of the sports and exercise sciences, ed. H.K. Books, Champaign 25. Leffondre K et  al (2006) Modelling smoking history using a comprehensive smoking index: application to lung cancer. Stat Med 25(24):4132–4146 26. Siemiatycki J et  al (2004) Listing occupational carcinogens. Environ Health Perspect 112(15):1447–1459

13

Cancer Causes Control 27. Evans D et al (2012) Combining directed acyclic graphs and the change-in-estimate procedure as a novel approach to adjustmentvariable selection in epidemiology. BMC Med Res Methodol 12:156 28. Royston P, Sauerbrie W (2008) Multivariable model—building: a pragmatic approach to regression analysis based on fractional polynomials for modelling continuous variables. John Wiley & Sons, West Sussex 29. Royston P, Altman DG (1994) Regression using fractional polynomials of continuous covariates: parsimonious parametric modelling. J Royal Stat Soc 43(3):429–467 30. WHO (1995) Physical status: the use and interpretation of anthropometry. Report of a WHO Expert Committee, WHO Technical Report Series 854. World Health Organization, Geneva 31. Kubik A et  al (2008) A case–control study of lifestyle and lung cancer associations by histological types. Neoplasma 55(3):192–199 32. Mao Y et al (2003) Physical activity and the risk of lung cancer in Canada. Am J Epidemiol 158(6):564–575 33. Sinner P et  al (2006) The association of physical activity with lung cancer incidence in a cohort of older women: the Iowa Women’s Health Study. Cancer Epidemiol Biomark Prev 15(12):2359–2363 34. Kukkonen-Harjula K (2007) Physical activity and cardiovascular health—work and leisure differ. Scandinavian J Work Environ Health 33(6):401–404 35. Karlqvist L, Leijon O, Harenstam A (2003) Physical demands in working life and individual physical capacity. Eur J Appl Physiol 89(6):536–547 36. Peake JM et  al (2016) Recovery of the immune system after exercise. J Appl Physiol. doi:10.1152/japplphysiol.00622.2016 37. Pedersen BK, Rohde T, Ostrowski K (1998) Recovery of the immune system after exercise. Acta Physiol Scand 162(3):325–332 38. Grivennikov SI, Greten FR, Karin M (2010) Immunity, inflammation, and cancer. Cell 140(6):883–899 39. Schottenfeld DJF Fraumeni (2006) Cancer epidemiology and prevention 3rd ed. Oxford University Press, New York 40. Shephard RJ, Shek PN (1998) Associations between physi cal activity and susceptibility to cancer: possible mechanisms. Sports Med 26(5):293–315 41. Steindorf K et al (2005) Case–control study of lifetime occupational and recreational physical activity and risks of colon and rectal cancer. Eur J Cancer Prev 14(4):363–371 42. Friedenreich CM, Courneya KS, Bryant HE (2001) Influence of physical activity in different age and life periods on the risk of breast cancer. Epidemiology 12(6):604–612