Estimating cancers attributable to physical inactivity in Australia

Abstract

Objective:

It was previously estimated that 1,814 (1.6% of incident cancers) were attributable to physical inactivity in Australia in 2010, when only three sites were considered. We estimated the burden of cancer due to physical inactivity in Australia for 13 sites.

Design:

The population attributable fraction estimated site-specific cancer cases attributable to physical inactivity for 13 cancers. The potential impact fraction was used to estimate cancers that could have been prevented in 2015 if Australian adults had increased their physical activity by a modest amount in 2004–05.

Methods:

We used 2004–05 national physical activity prevalence data, 2015 national cancer incidence data, and contemporary relative-risk estimates for physical inactivity and cancer. We assumed a 10-year latency period.

Results:

An estimated 6,361 of the cancers observed in 2015 were attributable to physical inactivity, representing 4.8% of all cancers diagnosed. If Australian adults had increased their physical activity by one category in 2004–05, 2,564 cases (1.9% of all cancers) could have been prevented in 2015.

Conclusions:

More than three times as many cancers are attributable to physical inactivity than previously reported. Physical activity promotion should be a central component of cancer prevention programs in Australia.

INTRODUCTION

In 2012, Cancer Council Australia commissioned a series of reports (published in 2015) to estimate the burden of cancer attributable to modifiable risk factors.¹ The reports restricted analyses to risk factors deemed to cause cancer, as determined by the World Cancer Research Fund (WCRF) or the International Agency for Research on Cancer (IARC). The report on physical inactivity considered colon, endometrial and post-menopausal breast cancer.² Physical inactivity was reportedly responsible for 1.6% of incident cancers in Australia, ranking below tobacco (13.4%), ultra-violet radiation (6.2%), inadequate diet (6.1%), overweight/obesity (3.4%), infections (2.9%) and alcohol (2.8%).¹

Since that time, the U.S. Department of Health and Human Services convened a panel of international experts (Physical Activity Guidelines Advisory Committee) to assess the evidence relating to physical activity and cancer. The Committee agreed that there is strong causal evidence for seven cancer sites (breast, colon, bladder, endometrial, kidney, oesophageal adenocarcinoma and gastric cancer) and moderate/weak evidence for an additional eight sites (non-Hodgkin lymphoma, rectum, head and neck, myeloma, myeloid leukemia, liver, small intestine, gallbladder).^3,4 These conclusions were informed by large, contemporary studies including the landmark U.S. National Cancer Institute (NCI) Cohort Consortium’s analysis of harmonised data from 1.44 million adults, which reported that leisure-time physical activity was associated with lower risk of 13 types of cancer.⁵

This recent evolution and strengthening of the physical activity and cancer risk evidence base has not been translated into tangible policy reform or public health action in Australia. A clearer understanding of the magnitude of cancer burden related to physical inactivity is needed to prioritise investment in physical activity promotion. Thus, the aim of our study was to provide updated estimates of the site-specific burden of cancer due to physical inactivity in Australia in 2015.

METHODS

Cancer sites included in analyses

Hazard ratios (HRs) for the risk of 15 cancers identified by the Physical Activity Guidelines Advisory Committee, for four physical activity categories relative to no leisure-time physical activity, were reported by Matthews et al.⁶ For two sites (rectum and small intestine) the dose-response analysis showed a weak U-shaped trend, and the 95% confidence intervals (CIs) cross the null for each point estimate. These results raised doubts regarding the causal nature of physical activity on rectal and small intestine cancers. We therefore excluded these sites from our analyses and focussed on the remaining 13 cancers. Matthews et al. also provided (unpublished) sex-specific HRs; these were not available for women for oesophagus (adenocarcinoma) or gastric cardia as only one cohort had sufficient cases (so a meta-analysis could not be performed).

Physical activity prevalence

Data from the Confidentialised Unit Record Files, Australian Bureau of Statistics (ABS) National Health Survey for 2004–05 were used to estimate mean minutes of leisure-time physical activity per week for the Australian population. We applied the person weighting to benchmark our estimates to the total Australian population.

The National Health Survey collected self-reported time spent walking or doing moderate or vigorous exercise for sport, recreation and fitness in the previous two weeks, which we converted to weekly estimates for our analysis. Consistent with the truncation rules applied for the International Physical Activity Questionnaire-short form, a maximum of nine hours per day of total leisure-time physical activity was applied (up to three hours each for walking, moderate- and vigorous-intensity physical activity).⁷ We applied a weighting of three metabolic equivalent of tasks (METs) for walking and moderate activity and six METs for vigorous activity, then added these together to derive total leisure physical activity MET-hours/week. For example, for someone reporting 6 hours of walking and 4 hours of fast swimming over a 2-week period, the total leisure physical activity would be 21 MET-hours/week ([6 hours/2] × 3 + [4 hours/2] × 6).

Physical activity prevalence was estimated by sex for seven age groups, 18–24, 25–34, 35–44, 45–54, 55–64, 65–74 and ≥ 75 years, using the physical activity categories corresponding to the estimated relative risks (≥30, 15-<30, 7.5-<15, >0-<7.5 and 0 MET-hour/week; Supplementary Table 1). Physical activity prevalence was also estimated for the 18–34 and 18–44 year age groups to correspond with the aggregated cancer counts (see ‘Incident cancers’ below).

Our unexposed reference category was ≥30 MET-hours/week of physical activity. This high cut-off point was chosen to fully capture the potential protective effect of physical activity on the risk of certain cancers. All categories below ≥30 MET-hours/week will hereafter be referred to as ‘suboptimal activity’.

Incident cancers

We used cancer incidence data for Australian adults in 2015 in seven age groups, 25–34, 35–44, 45–54, 55–64, 65–74, 75–84 and ≥85 years, obtained via an ad hoc data request from the Australian Institute of Health and Welfare (AIHW) (Supplementary Table 2). We assumed a latency period of 10 years, a time frame applied by Olsen et al.² and Friedenreich et al.⁸ This assumes that the effects of physical inactivity from 2004–05 will subsequently affect cancer risk in 2015. For cancers with <16 incident cases in a given age group (bladder, oesophagus (adenocarcinoma), gastric cardia, myeloma, liver and gallbladder), data were combined with those for subsequent 10-year age groups until this threshold was met. This threshold was used to ensure data confidentiality and stability, aligning with advice from the Surveillance, Epidemiology, and End Results Program at the U.S. National Cancer Institute.

Relative risk estimates

We transformed the HRs (Supplementary Table 3) so they were relative to the highest physical activity category (≥30 MET-hours/week) using the formula:

$$H{R}_i^{\prime }=\frac{H{R}_j}{H{R}_5}$$

where ${H{R}^{\prime }}_i$ is the transformed HR relative to the highest physical activity category for the i^th physical activity category ($i=1$ to 5; $i=1$ denotes the new reference category, the most active and $i=5$, the inactive category) and where $j=6-i$; thus for the original untransformed HRs, $j=1$ to 5; where $j=1$ denotes the inactive group and $j=5$ the most active.

Whilst it was possible to transform the point estimates, we were unable do this for the 95% CIs presented by Matthews et al.⁶ This is because of the way that variance (which is used to calculate CIs) propagates through formulas, e.g. if x and y are independent then $\mathrm{v}\mathrm{a}\mathrm{r}(\mathrm{x}+\mathrm{y})=\mathrm{v}\mathrm{a}\mathrm{r}\left(\mathrm{x}\right)+\mathrm{v}\mathrm{a}\mathrm{r}\left(\mathrm{y}\right)$. However, if the coefficient estimates are correlated (as is the case for the HRs for the different physical activity categories) there are non-zero off-diagonal elements in the variance-covariance matrix for the physical activity categories. It is impossible to correctly calculate the lower and upper CIs for the PAF or PIFs correctly without knowing the off-diagonal components of the variance-covariance matrix.⁹

Statistical analysis

The population attributable fraction (PAF) was calculated to estimate the proportional change in cancer incidence if suboptimal physical activity was eliminated;¹⁰ that is, if all Australian adults engaged in ≥30 MET-hours/week. This was done using Levin’s PAF formula for >2 exposure categories:¹¹

$$PAF=\frac{\sum _{i=1}^k\left(p_i\right)\left(H{R}_i^{\prime }-1\right)}{1+\sum _{i=1}^k\left(p_i\right)\left(H{R}_i^{\prime }-1\right)}$$

Where $p_i$ is the proportion of Australian population in i^th activity category, $k$ is the number of physical activity categories, $p_1$ is the proportion of population in the highest physical activity (unexposed, reference) category, $p_5$ is the proportion in the lowest physical activity category (most exposed), and $H{R}_i^{\prime }$ is the HR for cancer for the i^th activity category relative to the most active category (i = 1, unexposed, ≥30 MET-hours/week). PAFs were calculated separately by age group and sex.

For each of the 13 cancer sites, and for men and women separately, the number of cancers attributable to physical inactivity was estimated by multiplying the PAF for each age group by the observed number of incident cancers in the corresponding age group. To estimate the total number of cancers attributable to physical inactivity for each cancer site, the observed and attributable cancers were summed across each age group and sex; the overall PAF for each cancer type was derived from these totals ((attributable/observed) × 100). Similarly, the observed and attributable cancers across all cancer sites were estimated by summing the relevant sex-age-specific counts. For women, HRs for oesophageal (adenocarcinoma) or gastric cardia were not available, as there were too few cases reported within the NCI Cohort Consortium to enable sex-specific estimates. Therefore, we used the overall HRs (for men and women combined) from Matthews et al.⁶ for these sites to calculate the totals.

While the PAF represents the maximum benefit achievable at the population level, the potential impact fraction (PIF) estimates the number of potentially preventable cancers in 2015 under counterfactual scenarios of changing physical activity prevalence. The PIF is the proportional change in cancer incidence if the exposure to the risk factor were reduced, but not necessarily eliminated, as is assumed in the PAF.¹⁰ We applied the ‘proportional shift’ method, to estimate the number of preventable cancers under the hypothetical scenario in which everyone moves up one ordinal physical activity category (i.e. the counterfactual prevalence [P*]; Supplementary Table 2):

$$PIF=\frac{\sum _{i=1}^k{P}_{i}H{R^{\prime }}_i-\sum _{i=1}^k{P}_i^{\mathrm{*}}H{R^{\prime }}_i}{\sum _{1=1}^k{P}_{i}H{R^{\prime }}_i}$$

Where $P_i$ is the observed prevalence in the i^th physical activity category, $k$ is the number of physical activity categories, $P_i$ is the observed prevalence in the highest physical activity category (reference),$P_5$ is the observed prevalence in the inactive category,$P_i^{\mathrm{*}}$ is the counterfactual prevalence of physical activity, and $H{R}_i^{\prime }$ is the HR for cancer for the i^th physical activity category relative to the most active category. After the intervention the counterfactual cancer incidence rates are $I^{\mathrm{*}}=(1-PIF)I$ where $I$ is the incident rate before the intervention, and the cancers preventable are given by $I-I^{\mathrm{*}}$. PIFs, counterfactual incidence and preventable cancers were estimated separately by cancer type, sex and 10-year age group; for bladder, oesophagus (adenocarcinoma), gastric cardia, myeloma, liver and gallbladder cancer, the younger age groups were combined to 18–34 years or 18–44 years due to there being too few incident cancer cases in the youngest age groups.

To estimate the overall PIF for each cancer type under the hypothetical physical activity prevalence, the observed and counterfactual incidence and preventable cases were summed across each age group and sex and the PIF for each cancer type was derived from these totals $((1-(I^{\mathrm{*}}/I))\mathrm{*}100).$

Sensitivity Analyses

Our primary analyses assume that physical activity reduces cancer risk at least partly through reductions in body mass index (BMI), i.e. BMI is a mediator. Because BMI could also confound the relationship between physical activity and cancer risk, we repeated all analyses using HRs from Matthews et al.⁶ that had adjusted for BMI. The transformed HRs relative to the highest physical activity category (≥30 MET-hours/week) are presented in Supplementary Table 5.

RESULTS

Population Attributable Fractions

The site-specific PAFs and estimated number of cancers attributable to suboptimal physical activity (<30 MET-hours/week) in 2015 are shown in Table 1. We estimated that 4,904 (13.0%) of the 37,743 observed incident cancer cases from the seven cancer sites for which evidence is deemed strong were attributable to suboptimal activity. For the six moderate/weak evidence cancer sites there were 15,878 observed incident cancers in 2015, of which we estimated that 1,457 (9.2%) were attributable to suboptimal activity. Overall, we estimated that 6,361 (11.9%) of the 53,621 cancers diagnosed at the 13 selected cancer sites in 2015 were attributable to suboptimal activity. If we consider the total 133,022 cancer cases combined from 2015, the physical inactivity attributable cancers equate to 4.8% of the burden in Australia.

Table 1.

Population Attributable Fraction (PAF) and estimated number of cancers attributable to physical inactivity in Australia in 2015

Men				Women			Persons4
	Obs.	PAF	Attrib.	Obs.	PAF	Attrib.	Obs.	PAF	Attrib.
Strongest evidence of association 1
Breast	-	-	-	16,971	11.1	1,885	16,971	11.1	1,885
Colon	5,353	27.8	1,489	5,062	6.4	323	10,415	17.4	1,812
Bladder	2,054	3.2	65	672	14.4	97	2,726	5.9	162
Endometrial	-	-	-	2,690	13.6	367	2,690	13.6	367
Kidney	2,282	14.1	322	1,182	17.3	205	3,464	15.2	527
Oesophagus (adenocarcinoma)2,3	628	5.1	32	-	-	-	742	9	67
Gastric cardia2,3	601	8.3	50	-	-	-	735	11.4	84
Subtotal (7 types, strong association)	10,918	17.9	1,958	-	-	-	37,743	13	4,904
Head and neck	3,402	15.8	538	1,308	−5.2	−68	4,710	10	470
Myeloma3	1,078	1.9	21	806	−4.2	−34	1,884	−0.7	−13
Myeloid leukemia	718	11.3	81	529	11.5	61	1,247	11.4	142
Liver3	1,529	22.1	338	562	−0.5	−3	2,091	16	335
Non-Hodgkin lymphoma	2,831	−2.1	−60	2,153	12.4	267	4,984	4.2	207
Gallbladder 3	459	65.4	300	503	3.2	16	962	32.8	316
Subtotal (6 types, weak/moderate association)	10,017	12.2	1,218	5,861	4.1	239	15,878	9.2	1,457
Total (13 types)	20,935	15.2	3,176	-	-	-	53,621	11.9	6,361

¹According to the USA Physical Activity Guidelines, 2018

²Due to few cases of oesophagus (adenocarcinoma) and gastric cardia in women, HRs were not available, thus PAF and attributable cancers have not been estimated.

³Due to low numbers of incident cases for bladder, oesophagus (adenocarcinoma), gastric cardia, myeloma, liver and gallbladder in young age groups, age groups with <16 incidence counts were combined with the subsequent 10-year age group until this threshold had been met.

⁴Attributable cases were derived from the sex-specific values, and the PAFs were estimated from these derived values except for oesophagus (adenocarcinoma) and gastric cardia, where the Person’s PAF and attributable cancers were obtained using the overall HRs (i.e. men and women combined), as these were not available separately for women.

The PAF and estimated number of cancers attributable to suboptimal levels of physical activity by sex and age group are presented in Supplementary Table 6. Cancers with the highest PAF for men and women combined were gallbladder at 32.8%, followed by colon at 17.4% and liver at 16.0%. In some instances, negative PAF values were derived; this occurred when the HR <1.00 for most physical activity categories, relative to the most active group (see Supplementary Table 6). Site-specific PAFs differed by sex. There were substantially more cases of colon, head and neck, liver and gallbladder cancers attributable to suboptimal physical activity for men than women, whereas the attributable cases of bladder and non-Hodgkin lymphoma were higher for women.

Our sensitivity analysis utilising the HRs adjusted for BMI demonstrated a decrease in the number of cancers attributable to suboptimal physical activity. For the seven sites with strong evidence of a causal effect, we observed 3,458 (9.2%) attributable cancers. This equates to 1,446 (29.5%) fewer than the 4,904 estimated using the HRs that were not adjusted for BMI. For the six sites with moderate/weak evidence, we observed 1,079 (6.8%) attributable cancers, 378 (25.9%) fewer than in our primary analyses (Supplementary Table 7).

Potential Impact Fraction

We estimated that 1,544 (4.1%) of the 37,743 incident cancers at the seven sites with strong evidence could have been prevented if everyone had shifted up one physical activity category a decade earlier, in 2004–05 (Table 2). For the six moderate/weak-evidence cancer sites, 1,020 (6.4% of 15,878) incident cancers could have been prevented under the same counterfactual scenario. In total, 2,564 (4.8%) of 53,621 incident cancers at 13 cancer sites could have been prevented by increasing physical activity a decade earlier. The cancers with the highest PIFs were gallbladder (25%), liver (12%) and gastric cardia (8%).

Table 2:

Potential Impact Fraction (PIF) and estimated number of preventable cancers in Australia in 2015 assuming an increase in physical activity prevalence levels (shifting up one category) in 2004–05

			Men		Women				Persons5
	Observed	PIF	CF Incidence2	Prevented	Observed	PIF	CF Incidence2	Prevented	Observed	PIF	CF Incidence2	Prevented
Strongest evidence of association 1
Breast	-	-	-	-	16,971	3.7	16,349	622	16,971	3.7	16,349	622
Colon	5,353	6.4	5,011	342	5,062	1.6	4,980	82	10,415	4.1	9,991	424
Bladder	2,054	2.8	1,997	57	672	0.6	668	4	2,726	2.2	2,665	61
Endometrial	-	-	-	-	2,690	4.1	2,581	109	2,690	4.1	2,581	109
Kidney	2,282	4.4	2,182	100	1,182	10.4	1,059	123	3,464	6.4	3,241	223
Oesophagus (adenocarcinoma)3,4	628	6.7	586	42	-	-	-	-	742	6.2	696	46
Gastric cardia3,4	601	7.5	556	45	-	-	-	-	735	8	676	59
Subtotal (7 types, strong association)	10,918	5.4	10,332	586	-	-	-	-	37,743	4.1	36,199	1,544
Weak/moderate evidence of association 1
						-
Head and neck	3,402	5.8	3,205	197	1,308	−0.7	1,317	−9	4,710	4	4,522	188
Myeloma4	1,078	6	1,013	65	806	7.9	742	64	1,884	6.8	1,755	129
Myeloid leukemia	718	3.1	696	22	529	10.2	475	54	1,247	6.1	1,171	76
Liver4	1,529	14	1,315	214	562	5.2	533	29	2,091	11.6	1,848	243
Non-Hodgkin lymphoma	2,831	−1.6	2,876	−45	2,153	9	1,960	193	4,984	3	4,836	148
Gallbladder4	459	38.6	282	177	503	11.7	444	59	962	24.5	726	236
Subtotal (6 types, weak/moderate association)	10,017	6.3	9,387	630	5,861	6.7	5,471	390	15,878	6.4	14,858	1,020
Total (13 types)	20,935	5.8	19,719	1,216	-	-	-	-	53,621	4.8	51,057	2,564

¹According to the USA Physical Activity Guidelines, 2018

²CF Incidence = Counterfactual Incidence = (1 − PIF) × Observed

³Due to few cases for oesophagus (adenocarcinoma) and gastric cardia in women, HRs were not available, thus PIF, counterfactual incidence and preventable cancers have not been estimated

⁴Due to low numbers of incident cases for bladder, oesophagus (adenocarcinoma), gastric cardia, myeloma, liver and gallbladder in young age groups, age groups with <16 incidence counts were combined with the subsequent 10-year age group until this threshold had been met.

⁵Preventable cases and CF incidence were derived from the sex-specific values, and the PIFs were estimated from these derived values. For oesophagus (adenocarcinoma) and gastric cardia, the Person’s CF incidence, PIF, preventable cancers were obtained using the overall HRs (i.e. men and women combined), as these were not available separately for women.

The PIF related to suboptimal levels of physical activity by sex and age group are presented in Supplementary Table 8. We observed substantial differences in PIF by site for men and women, but limited variability across age groups.

Applying HRs adjusted for BMI decreased cancer cases potentially prevented by moving up one physical activity a decade earlier. For the seven cancer sites with strong evidence of causation, the number decreased from 1,544 to 962 (37.7% fewer cases potentially prevented). For the six sites with moderate/weak evidence, the estimate decreased from 1,020 to 869 (14.8% fewer cases potentially prevented; Supplementary Table 9).

DISCUSSION

We estimated the magnitude of cancer burden related to physical inactivity in Australia in 2015. We considered 13 cancer sites and found that 6,361 (11.9%) of these cancer cases could be attributed to suboptimal physical activity. This PAF represents the ‘best case’ scenario for cancer prevention, where every adult participates in the highest category of physical activity each week (≥30 MET-hours/week). The PIF provides a more realistic and potentially achievable public health intervention effect. If Australian adults had modestly increased their physical activity (moved up one category) in 2004–05, 2,564 cancers at 13 cancer sites could have been prevented in 2015.

A strength of this study was the incorporation of the updated evidence-base demonstrating the broad range of cancer sites that may be prevented by physical activity.^3,4 We incorporated recently published, robust risk estimates for physical inactivity and cancer risk from a harmonised dataset of ~750,000 adults.⁶ We utilised nationally representative data on physical activity prevalence to generate estimates.

Nonetheless, the physical activity prevalence estimates were derived from self-report, and limited to leisure-time physical activity. A previous PAF project conducted sensitivity analyses, incorporating a bias-adjusted estimate of physical activity from accelerometer data, which resulted in their PAF increasing from 8% to 11%.¹² This suggests that the burden of cancer attributable to physical inactivity in Australia may be higher than estimated by the present study. It would be helpful if future National Health Survey waves integrated an accelerometer sub-study to enable adjustment for measurement error in prevalence estimates. An accelerometer sub-study would also allow generation of attenuation factors from regression calibration, so that studies of cancer risk could address measurement error when estimating HRs and 95% CIs.^13,14 Both of these measures would help estimate more accurate PAF estimates.

Our findings assume a 10-year latency period, measuring the time between exposure assessment in 2004–05 and cancer diagnosis in 2015. This is in line with other similar studies investigating the PAF.^2,12,15 However, this assumption may be more accurate for certain cancer sites. Our analysis did not follow people over time, nor take into consideration potential protective benefits of physical activity across the life course.

Our physical activity PAF estimate of 4.8% of all cancers is higher than the 1.6% from the Cancer Council Australia series. The PAF estimates for other modifiable risk factors generated for the Cancer Council Australia series are unlikely to have changed substantially. This suggests that the proportion of cancers attributable to physical inactivity is like that for overweight/obesity.¹ It is reasonable to argue that the full cancer prevention benefit of optimal levels of physical activity will not be realised at a population level, and therefore the PIF presents a more achievable intervention aim. We estimate that 2,564 (4.8%) cases from the 13 cancer sites in 2015 could have been prevented if physical activity had been increased by a modest amount a decade earlier. Even this number exceeds the number of cancers attributed to physical inactivity by Olsen et al.,² highlighting that physical activity has been underutilised as a cancer prevention strategy in Australia.

This objective of this research was to provide updated evidence, reflecting the current expert consensus on the role of physical activity in cancer prevention,^3,4 to inform cancer control policy and strategy in Australia. With over 6,300 Australian cancer cases attributable to suboptimal levels of physical activity, cancer control agencies should prioritise investment into strategies to increase participation in physical activity across all ages, such as the International Society for Physical Activity and Health’s ‘Eight Investments that Work for Physical Activity’. This systems-based approach was designed to address the growing burden of non-communicable diseases globally, with a target of a 15% reduction in physical inactivity by 2030.¹⁶

Conclusion

More than three times as many cancers are attributable to physical inactivity than previously reported. Our report provides a contemporary understanding of the cancer burden due to physical inactivity, building a case for Australian cancer control agencies to prioritise and invest in physical activity promotion.

Practical implications.

1. Physical activity is not widely promoted by Australian cancer control agencies. Currently only colon, postmenopausal breast and endometrial cancer are cited as cancers attributable to low levels of physical activity.

2. We estimate that more than three times as many cancers are attributable to physical inactivity as currently thought.

3. Australia has no physical activity plan and no coordinated physical activity strategy; the findings of our research are central to physical activity advocacy going forward.

Abbreviations

WCRF

World Cancer Research Fund

IARC

International Agency for Research on Cancer

NCI

National Cancer Institute

HR

Hazard ratio

CI

Confidence interval

ABS

Australian Bureau of Statistics

MET

Metabolic equivalent of task

AIHW

Australian Institute of Health and Welfare

PAF

Population attributable fraction

PIF

Potential impact fraction

BMI

Body mass index