Main

Renal cancer is one of the top 10 cancer sites in the United States and Europe. Each year, about 65 000 new cases are diagnosed in the United States (Howlader et al, 2012) and about 60 000 new cases are diagnosed in the European Union (Boyle and Ferlay, 2005). Well-established unfavourable risk factors for renal cancer include smoking, obesity, hypertension, and type 2 diabetes mellitus (Scelo and Brennan, 2007). In contrast, physical activity may prevent the development of renal cancer, partly because it helps reduce obesity (Wing, 1999), blood pressure (Blair et al, 1984), and insulin resistance (Rosenthal et al, 1983). Physical activity may also independently decrease renal cancer risk by lowering lipid peroxidation levels (Vincent et al, 2002). However, few available epidemiologic studies have been able to report a clear inverse association between physical activity and renal cancer (Leitzmann, 2011). Moreover, no meta-analysis is available on the relation between physical activity and renal cancer. To address this research gap, we conducted a systematic literature search and meta-analysis to quantify the association between physical activity and renal cancer, taking into account the methodologic quality of the studies.

Materials and methods

Literature search

Our systematic review and meta-analysis adhered to the PRISMA guidelines (Moher et al, 2009). Both authors searched the literature using PubMed (see Appendix B for PubMed search options) and Web of Knowledge (see Appendix C for Web of Knowledge search options) to identify published non-ecologic epidemiologic studies quantifying the relationship between physical activity and renal cancer risk in individuals without a cancer history. That search was complemented by a scan of the reference lists of the identified studies and a scan of the reference list of a previous systematic review (Leitzmann, 2011). We considered all human research articles published in English through the end of September 2012 not classified as review, meta-analysis, editorial, comment, letter, practice guideline, or news. Our search strategy included the terms physical activity, exercise, cardiorespiratory fitness, cardiovascular fitness, resistance training, endurance training, aerobic, sport, athletes, players, lifestyle, kidney cancer, renal cancer, renal cell cancer, renal carcinoma, renal cell carcinoma, cancer, risk, incidence, and mortality. The search strategy excluded research on cancer survivors and research on specific types of cancer other than renal cancer. That search yielded 586 potential articles. Irrelevant articles were eliminated after screening titles and abstracts (n=477) or manuscripts (n=82). The 27 remaining studies (Goodman et al, 1986; Paffenbarger et al, 1987; Brownson et al, 1991; Lindblad et al, 1994; Mellemgaard et al, 1994, 1995; Bergstrom et al, 1999, 2001; Parker et al, 2002; Menezes et al, 2003; Mahabir et al, 2004; Nicodemus et al, 2004; van Dijk et al, 2004; Washio et al, 2005; Chiu et al, 2006; Pan et al, 2006; Setiawan et al, 2007; Tavani et al, 2007; Hu et al, 2008, 2009; Moore et al, 2008; Thompson et al, 2008; Yun et al, 2008; Spyridopoulos et al, 2009; Wilson et al, 2009; George et al, 2011; Parent et al, 2011) proved to be relevant.

To avoid duplicate information from overlapping studies, we removed eight of the 27 identified studies because their results were pooled (Lindblad et al, 1994; Mellemgaard et al, 1994) or updated (Parker et al, 2002; Menezes et al, 2003; Pan et al, 2006) in studies (Mellemgaard et al, 1995; Chiu et al, 2006; Hu et al, 2008) using the same database, because they reported results (Hu et al, 2009; Wilson et al, 2009) presented earlier (Mahabir et al, 2004; Hu et al, 2008), or because their investigations of total sitting time (George et al, 2011) were closely related to a previous study (Moore et al, 2008) on physical activity from the same cohort. The remaining 19 studies (Goodman et al, 1986; Paffenbarger et al, 1987; Brownson et al, 1991; Mellemgaard et al, 1995; Bergstrom et al, 1999, 2001; Mahabir et al, 2004; Nicodemus et al, 2004; van Dijk et al, 2004; Washio et al, 2005; Chiu et al, 2006; Setiawan et al, 2007; Tavani et al, 2007; Hu et al, 2008; Moore et al, 2008; Thompson et al, 2008; Yun et al, 2008; Spyridopoulos et al, 2009; Parent et al, 2011) were included in the meta-analysis.

Quality score

The magnitude and heterogeneity of risk estimates may depend on the methodologic quality associated with the underlying study and with the risk estimate derivation. Similar to three previous systematic reviews (Monninkhof et al, 2007; Voskuil et al, 2007; Liu et al, 2011) on the association between physical activity and specific types of cancer, both authors employed a quality score proposed by Voskuil et al (2007) to assess the methodologic quality of the studies and the consistency of the available evidence. Please refer to Appendix A for a description of the items covered by the quality score.

Main statistical analysis

Because some studies presented risk estimates for men and women and some studies investigated more than one physical activity domain, the 19 identified studies reported a total of 37 risk estimates. If separate risk estimates were available for men and women, both risk estimates were included in the meta-analysis because they were based on independent samples. To prevent potential bias arising from the fact that the risk estimates for the various physical activity domains were based on the same study population, both authors allowed only one estimate per study and gender in the main analysis. Specifically, if more than one physical activity domain was studied, we selected the risk estimate with the highest quality score in the main analysis. Of the 37 risk estimates, 25 were included in the main analysis.

In the meta-analysis, we interpreted odds ratios and hazard ratios as relative risk estimates (RRi), computed the natural logarithms of those risk estimates log(RRi) with corresponding standard errors si=(log(upper 95% confidence interval (CI) bound of RR)−log(RR))/1.96, and employed a random-effects model to determine the weighted average of those log(RRi)s while allowing for heterogeneity of effects. In the random-effects model, the log(RRi)s were weighted by wi=1/(si2+t2) where si represented the standard error of log(RRi) and t2 represented the restricted maximum-likelihood estimate of the overall variance (Higgins and Thompson, 2002). In one case (Paffenbarger et al, 1987), we derived the standard error of the log(RRi) using the P-value accompanying the RR estimate. In five additional cases (Brownson et al, 1991; Mellemgaard et al, 1995; Bergstrom et al, 1999, 2001; Chiu et al, 2006), the reported RRs used the highest rather than the lowest activity level as the reference category, so we reversed those RRs for comparability. Heterogeneity of the risk estimates was assessed using the Q- and I2-statistics (Higgins and Thompson, 2002). Publication bias was tested using funnel plots (Egger et al, 1997), Egger’s regression test (Egger et al, 1997), and Begg’s rank correlation test (Begg and Mazumdar, 1994).

Statistical subanalyses

If a study presented separate risk estimates for recreational, occupational, and total physical activity, in a subanalysis all those risk estimates were included in the meta-analysis. Also, in a subanalysis we used all 37 risk estimates to investigate the impact of prespecified potentially influential methodologic factors on the summary risk estimate.

On the basis of pre-existing evidence, we hypothesised that the relations of physical activity to renal cell cancer would differ according to study design (cohort or case–control), physical activity domain (recreational, occupational, or total physical activity), and gender (men, women, or men and women combined). Thus, we conducted subanalyses within categories of those variables. We also performed exploratory analyses that were stratified by geographic region (North America, Europe, Asia), type of physical activity assessment (energy expenditure, physical fitness, moderate-to-vigorous physical activity duration, moderate-to-vigorous physical activity frequency, and qualitative assessments using categories, such as ‘sedentary’, ‘light’, ‘moderate’, or ‘high’ physical activity), timing in life of physical activity (recent physical activity, past physical activity, or consistent physical activity over time), number of adjustment factors (in quartiles), adjustments for smoking and obesity (adjusted for smoking and obesity, adjusted for smoking but not obesity, adjusted neither for smoking nor obesity; the option of adjusting for obesity but not smoking was not included because it did not occur), adjustment for hypertension (yes, no), adjustment for type 2 diabetes mellitus (yes, no), or methodologic quality score (in tertiles). To assess the influence of those factors, we applied random-effects meta-analysis regression comparing the model including the current factor of interest as a single explanatory variable with the null model not including any explanatory variables.

All statistical analyses were performed in R (R Development Core Team, 2011) using the R-package ‘metafor’ (Viechtbauer, 2010). Risk estimates are reported with 95% CIs. Statistical significance is based on the 5% significance level.

Results

Description of underlying study characteristics

Table 1 presents the 19 studies on physical activity and renal cancer risk included in the meta-analysis. Because six studies stratified results by gender and nine studies investigated more than one physical activity domain, the 19 studies reported a total of 37 risk estimates.

Table 1 Characteristics of the 19 studies on physical activity and renal cancer risk included in the meta-analysis
Full size table

When grouping studies by potentially effect modifying factors (Table 2), we noted that there was an equal number of risk estimates from case–control and prospective cohort studies, with the vast majority of studies originating in the United States or Europe. Half of the risk estimates were based on recreational physical activity, one-third of the risk estimates were based on occupational activity, and four risk estimates were based on total physical activity. Half of the physical activity assessments were of a qualitative type and the remaining half were of a quantitative type. Nearly two-thirds of the risk estimates were adjusted for smoking and obesity, one-third of the risk estimates were adjusted for hypertension, and one-sixth of the risk estimates were adjusted for history of type 2 diabetes mellitus.

Table 2 Summary risk estimates and I2 measures of heterogeneity from random-effects models stratified by selected study characteristics
Full size table

Meta-analysis

The random-effects model summarising the 25 risk estimates with the highest quality scores from each of the 19 studies (Figure 1) revealed a statistically significant 12% reduction in renal cancer risk when comparing a high with a low level of physical activity (RR=0.88; 95% CI=0.79–0.97; I2=33%). The magnitude of that summary risk estimate did not materially change when grouping those 25 risk estimates by study design (Figure 1), physical activity domain (Figure 2), or gender (Figure 3). That meta-analysis combined a total of 2 327 322 subjects and 10 756 cases. No publication bias was indicated by the funnel plot (Figure 4), Egger’s regression test (P=0.89), or Begg’s rank correlation test (P=0.53).

Figure 1
figure 1

Forest plot corresponding to the main random-effects meta-analysis including 25 risk estimates quantifying the relationship between high physical activity and renal cancer risk. Relative risks (RRs) compare high vs low levels of physical activity and are grouped by study design. The size of the box representing each risk estimate is proportional to the weight that the risk estimate contributed to the summary risk estimate.

Full size image
Figure 2
figure 2

Forest plot corresponding to the main random-effects meta-analysis including 25 risk estimates quantifying the relationship between high physical activity and renal cancer risk. Relative risks (RRs) compare high vs low levels of physical activity and are grouped by physical activity domain. The size of the box representing each risk estimate is proportional to the weight that the risk estimate contributed to the summary risk estimate.

Full size image
Figure 3
figure 3

Forest plot corresponding to the main random-effects meta-analysis including 25 risk estimates quantifying the relationship between high physical activity and renal cancer risk. Relative risks (RRs) compare high vs low levels of physical activity and are grouped by gender. The size of the box representing each risk estimate is proportional to the weight that the risk estimate contributed to the summary risk estimate.

Full size image
Figure 4
figure 4

Funnel plot corresponding to the main random-effects meta-analysis including 25 risk estimates quantifying the relationship between high physical activity and renal cancer risk.

Full size image

Renal cancer end point

Because physical activity may differentially impact the incidence vs mortality of kidney cancer, we repeated the main analysis after excluding the two risk estimates from the papers on kidney cancer mortality (Washio et al, 2005; Thompson et al, 2008). The summary risk estimate remained unchanged (RR=0.88; 95% CI=0.80–0.98).

Potentially influential methodologic factors

In subanalyses investigating potentially influential methodologic factors, all 37 risk estimates were used. The random-effects summary risk estimate (RR=0.90; 95% CI=0.82–0.98; I2=26%) of those 37 risk estimates did not substantially differ from that of the main analysis. We found that the methodologic quality score significantly influenced the magnitude of the summary risk estimate (P=0.02; Table 2) but not the underlying overall variation t2. The best evidence synthesis of studies that fell into the high tertile of the quality score yielded a meta-analysis estimate for the relation of physical activity to renal cancer of 0.78 (95% CI=0.66–0.92; t2=0.02). In contrast, the meta-analysis RRs for studies falling into the intermediate (RR=1.00; 95% CI=0.89–1.13; t2=0) and lower (RR=0.93; 95% CI=0.80–1.07; t2=0.02) tertiles of the quality score were statistically nonsignificant.

When stratifying by the type of physical activity assessment, the summary risk estimates based on frequency of moderate-to-vigorous physical activity (RR=0.72; 95% CI=0.53–0.97) or duration of moderate-to-vigorous physical activity (RR=0.85; 95% CI=0.69–1.04) appeared to be stronger than those based on energy expenditure (RR=0.97; 95% CI=0.84–1.12) or qualitative physical activity assessments (RR=0.98; 95% CI=0.85–1.14). However, that variation was not statistically significant (P=0.24). Similarly, the magnitude of the inverse association between physical activity and renal cancer appeared to be stronger with a larger number of adjustment factors, although that difference was not statistically significant (P=0.28). The meta-analysis RR for studies in the top tertile of the number of adjustment factors was 0.83 (95% CI=0.71–0.97), whereas the RR for studies in the bottom tertile of the number of adjustment factors was 0.96 (95% CI=0.85–1.08). There was no difference in risk estimates between study designs (RR for case–control studies=0.91; 95% CI=0.79–1.04; RR for cohort studies=0.89; 95% CI=0.80–0.99; P-value for interaction=0.93). Similarly, none of the following remaining study characteristics affected the summary risk estimates: physical activity domain (P=0.84), timing in life of physical activity (P=0.18), gender (P=0.41), geographic region (P=0.63), joint adjustment for smoking and obesity (P=0.31), hypertension adjustment (P=0.30), and diabetes adjustment (P=0.18).

We further examined study characteristics according to the quality score (Table 3). Studies that fell into the top tertile of the quality score tended to employ quantitative physical activity assessments, to investigate recreational activity, to examine recent physical activity, to use a cohort design, and to adjust for smoking, obesity, hypertension, and diabetes. In contrast, studies in the bottom tertile of the quality score tended to employ qualitative physical activity assessments, to investigate occupational activity, to examine past physical activity, to use a case–control design, and to not adjust for smoking, obesity, hypertension, or diabetes.

Table 3 Distribution of methodologic characteristics (absolute frequencies) of all 37 risk estimates by tertile of quality scorea
Full size table

After adjusting the main random-effects model for study quality (in tertiles), the previously observed heterogeneity (P-heterogeneity=0.03) of risk estimates was no longer evident (P-heterogeneity=0.12).

Discussion

Main results

This comprehensive meta-analysis revealed a statistically significant 12% reduction in renal cancer risk associated with a high vs low level of physical activity.

Potentially influential factors

The summary RR estimate was not affected by individual potentially influential factors, such as type of physical activity assessment, physical activity domain, timing in life of physical activity, gender, study design, study region, number of adjustment factors, and adjustments for smoking, obesity, hypertension, or diabetes. However, the quality score representing a combination of specific factors affected the summary risk estimate. Summary risk estimates based on studies that fell into the top quality score tertile were statistically significantly inverse, whereas summary risk estimates based on studies that fell into the intermediate or bottom quality score tertiles were not. After adjusting for study quality, the previously observed heterogeneity in the random-effects model was no longer statistically significant.

The influence of individual factors was examined in previous meta-analyses of physical activity and cancers of the colorectum (Harriss et al, 2009; Boyle et al, 2012), pancreas (Bao and Michaud, 2008), and prostate (Liu et al, 2011). In agreement with our observations, no statistically significant heterogeneity across gender (Bao and Michaud, 2008; Boyle et al, 2012), study design (Liu et al, 2011), geographic region (Bao and Michaud, 2008; Harriss et al, 2009), physical activity domain (Boyle et al, 2012), or obesity adjustment (Bao and Michaud, 2008; Harriss et al, 2009) was reported.

The influence of a quality score combining several factors was previously studied with respect to the associations between physical activity and cancers of the breast (Monninkhof et al, 2007), endometrium (Voskuil et al, 2007), prostate (Liu et al, 2011), colon (Boyle et al, 2012), and pancreas (O’Rorke et al, 2010). In agreement with our findings, the meta-analysis on physical activity and breast cancer (Monninkhof et al, 2007) detected a more pronounced risk reduction with increased quality score, while the remaining analyses (Voskuil et al, 2007; O’Rorke et al, 2010; Liu et al, 2011; Boyle et al, 2012) did not detect any statistically significant association between quality score and summary risk estimates. Two reviews (O’Rorke et al, 2010; Boyle et al, 2012), however, described decreased variation in risk estimates with increasing quality score. No such observation was made in this study.

Potential biological mechanisms

A high level of physical activity has been shown to reduce adiposity (Wing, 1999), hypertension (Blair et al, 1984), insulin resistance (Rosenthal et al, 1983), circulating levels of insulin-like growth factor 1 (Eliakim et al, 1996, 1998), and lipid peroxidation (Vincent et al, 2002) – factors positively associated with the development of renal carcinoma (Kellerer et al, 1995; Chow et al, 2000; Gago-Dominguez et al, 2002; van Dijk et al, 2004; Vatten et al, 2007; Yuen et al, 2009). Further potential cancer preventing mechanisms include the beneficial effects of physical activity on chronic inflammation and immune function (McTiernan, 2008). It is hypothesised, however, that the effects of physical activity on chronic inflammation are mediated, in part, through avoidance of adiposity. The exact mechanisms linking physical activity to immune function related to tumour suppression have not yet been established, but it is thought that physical activity improves the number or the function of natural killer cells.

Strengths and limitations

This is the first meta-analysis of physical activity and renal cancer. It bears the strengths and limitations inherent in any meta-analysis combining results from studies with heterogeneous study designs (Greenland and O’Rourke, 2008). Particular strengths of the current meta-analysis are that it is based on an extensive systematic literature review, that it rigorously excluded duplicate information induced by overlapping studies, and that it combined information from 19 studies, including a total of 2 327 322 subjects and 10 756 cases. A further strength is that it is among the few meta-analyses of physical activity and a specific type of cancer to assess the heterogeneity of summary estimates by potentially influential factors underlying the RR estimates. The employed quality score addressed potential selection, misclassification, and confounding biases, and accounted for heterogeneity of the results. An inverse association between physical activity and renal cancer risk was observed in analyses including all risk estimates and in analyses including only risk estimates from high-quality studies. In addition, no publication bias was detected.

One limitation of this meta-analysis is the large variation in the underlying studies regarding their definitions of exposure to physical activity – ranging from ‘physically very active’ to ‘5 h of vigorous physical activity per week or more’. Similarly, the definitions of physical activity referent groups ranged from ‘not physically active’ to ‘<5 h of vigorous physical activity per week’. Such variation did not allow us to conduct stratified analyses according to comparable groups of exposed and non-exposed individuals. Thus, we were not able to identify the specific type, intensity, frequency, and duration of physical activity required to lower renal cancer risk.

Conclusion

In conclusion, our comprehensive meta-analysis provides strong support for an inverse relation of physical activity to the risk of renal cancer. On the basis of high-quality studies, physical activity may decrease the risk of renal cancer by 22%. Future research is required to discern which specific types, intensities, frequencies, and durations of physical activity are needed for renal cancer risk reduction. High-quality studies that employ standardised physical activity assessments and uniform definitions of physical activity are warranted.