Weight Change and Risk of Colorectal Cancer
A total of 1,494 articles were identified via Scopus and Web of Science searches. Of these, 326 duplicate articles were excluded, and a further 1,116 articles were excluded on the basis of their title and abstract, leaving 52 articles for further evaluation. After obtaining the full articles, we excluded a further 39 papers, leaving 13 articles appropriate for the meta-analysis. The reasons for excluding studies are outlined in Figure 1; the majority of the excluded papers did not present results for change in weight or waist circumference (36%), or the incidence of colorectal cancer was not the outcome of interest (59%).
(Enlarge Image)
Figure 1.
Selection of studies for inclusion in a meta-analysis of weight change and risk of colorectal cancer, 1974–2014.
Table 1 shows a summary of the characteristics of the studies eligible for inclusion in the meta-analysis. Seven studies were conducted in Europe, 5 in the United States, and 1 in Australia. All studies examined change in weight (n = 10) or body mass index (n = 3) as the exposure of interest; none examined change in waist or hips circumference, and all presented results separately for men and women. Eight studies assessed change retrospectively by using self-reported weights in early adulthood and, of these, 3 studies measured weight at baseline. Five studies assessed change prospectively; of these, 3 studies measured weight at both waves of data collection. Thygesen et al. used self-reported weight assessments, recorded prospectively. Steins Bisschop et al. measured weight at baseline and self-reported weight 5 years after baseline. Three studies presented results for colon and rectal cancers combined, 6 studies presented results separately for colon and rectal cancers, and the remaining 4 studies presented results for colon cancer only.
Web Table 1 http://aje.oxfordjournals.org/content/181/11/832/suppl/DC1 available at http://aje.oxfordjournals.org/ shows the hazard ratios and corresponding 95% confidence intervals for each study for weight/body mass index change and the risk of colorectal cancer for the categorical and dose-response associations.
Thirteen prospective studies were included in this meta-analysis and the multivariable-adjusted hazard ratios comparing the largest weight gain category with the reference category for each study, and all studies combined are presented in Figure 2. There was evidence of an increased risk of colorectal cancer for the largest weight gain group, such that the risk of colorectal cancer was 16% higher for those in the highest weight gain category compared with those in the reference group (hazard ratio (HR) = 1.16, 95% CI: 1.08, 1.24). Studies that assessed change from midlife to older age (Figure 2B) had a lower hazard ratio compared with studies examining change from early adulthood to midlife (Figure 2A) (ratio of HRs = 0.83, 95% CI: 0.71, 0.97; P = 0.02). Studies that included physical activity in the analysis had higher hazard ratios than studies that did not (ratio of HRs = 1.21, 95% CI: 1.03, 1.42; P = 0.02) (Table 2).
(Enlarge Image)
Figure 2.
Adjusted hazard ratio (HR) for the risk of colorectal cancer comparing the largest weight gain group with the reference group for change measured from early adulthood to midlife (A) and change measured from midlife to older age (B) for males (squares) and females (circles), 1997–2014. Overall estimate reflects a combined hazard ratio from parts A and B. Dashed line, overall estimate; bars, 95% confidence interval (CI).
There was little heterogeneity across the studies (I = 10.2%, P = 0.30 from the χ test for heterogeneity) (Table 2). There did not appear to be any small-study effects; the corresponding funnel plot appeared symmetrical (Figure 3), and there was weak evidence of small-study effects from Egger's regression asymmetry test (P = 0.62).
(Enlarge Image)
Figure 3.
Funnel plot of the studies included in the meta-analysis comparing the largest weight gain group with the reference group for the risk of colorectal cancer, 1997–2014. The x-axis is on the log scale. Continuous line, no effect; dashed line, upper and lower 95% confidence limits.
Figure 4 shows the results for the meta-analysis comparing weight loss with the reference group. Of the 13 studies, 10 studies were included in this meta-analysis; 3 studies did not present results for a weight-loss group compared with a weight-stable or a reference group. There was no association between weight loss compared with a reference or stable group and the risk of colon or rectal cancer (HR = 0.96, 95% CI: 0.89, 1.05), regardless of whether change was assessed from early adulthood to midlife (Figure 4A) or from midlife to an older age (Figure 4B) (ratio of HRs = 1.05, 95% CI: 0.88, 1.26; P = 0.55).
(Enlarge Image)
Figure 4.
Adjusted hazard ratio (HR) for the risk of colorectal cancer comparing the largest weight loss group with the reference group for change measured from early adulthood to midlife (A) and change measured from midlife to older age (B) for males (squares) and females (circles), 1997–2014. Overall estimate reflects a combined hazard ratio from parts A and B. Dashed line, overall estimate; bars, 95% confidence interval (CI).
There was little heterogeneity across the studies (I = 0.0%; P = 0.71 from the χ test for heterogeneity) (Table 3). The funnel plot appeared symmetrical (Figure 5), and Egger's regression asymmetry test did not suggest the presence of small-study effects (P = 0.30).
(Enlarge Image)
Figure 5.
Funnel plot of the studies included in the meta-analysis comparing the largest weight loss group with the reference group for the risk of colorectal cancer, 1997–2014. The x-axis is on the log scale. Continuous line, no effect; dashed line, upper and lower 95% confidence limits.
Nine studies were included in the dose-response analysis. Rapp et al. and Samanic et al. presented results for change in body mass index, but because they did not present the mean height, we were unable to convert the hazard ratios for change in body mass index to hazard ratios for change in weight. Colditz and Coakley did not provide estimates for each category of weight change and, thus, a dose-response association could not be estimated. Han et al. presented results for percent change, which could not be converted to an absolute change.
The pooled hazard ratio was 1.03 (95% CI: 1.02, 1.05) per 5-kg weight gain (Figure 6), with no evidence of departure from linearity (P = 0.49). The pooled hazard ratios for weight gain from early adulthood to midlife (Figure 6A) and from midlife to older age (Figure 6B) were similar (ratio of HRs = 1.03, 95% CI: 0.95, 1.12; P = 0.51). There was some heterogeneity across the estimates (I = 41.6%; P = 0.02 from the χ test for heterogeneity).
(Enlarge Image)
Figure 6.
Adjusted hazard ratio (HR) for 5-kg weight gain and the risk of colorectal cancer for change measured from early adulthood to midlife (A) and change measured from midlife to older age (B) for males (squares) and females (circles), 1997–2014. Overall estimate reflects a combined hazard ratio from parts A and B. Dashed line, overall estimate; bars, 95% confidence interval (CI).
Table 4 shows the results of the meta-regression analyses for the prespecified covariates. Slightly stronger associations were found for men than for women (ratio of HRs = 1.03, 95% CI: 1.00, 1.07; P = 0.07). The pooled hazard ratios did not vary by cancer site or the method used to ascertain weight.
Visual inspection of the funnel plot did not indicate the presence of small-study effects, and Egger's regression asymmetry test did not suggest any small-study effects (P = 0.89) (Figure 7).
(Enlarge Image)
Figure 7.
Funnel plot of the studies included in the meta-analysis comparing a dose-response relationship between weight gain and the risk of colorectal cancer, 1997–2014. The x-axis is on the log scale. Continuous line, no effect; dashed line, upper and lower 95% confidence limits.
We conducted 3 sensitivity analyses where we excluded the estimates from the studies by Steins Bisschop et al., Renehan et al., and Oxentenko et al.; these did not change the results (data not shown).
Results
Study Selection
A total of 1,494 articles were identified via Scopus and Web of Science searches. Of these, 326 duplicate articles were excluded, and a further 1,116 articles were excluded on the basis of their title and abstract, leaving 52 articles for further evaluation. After obtaining the full articles, we excluded a further 39 papers, leaving 13 articles appropriate for the meta-analysis. The reasons for excluding studies are outlined in Figure 1; the majority of the excluded papers did not present results for change in weight or waist circumference (36%), or the incidence of colorectal cancer was not the outcome of interest (59%).
(Enlarge Image)
Figure 1.
Selection of studies for inclusion in a meta-analysis of weight change and risk of colorectal cancer, 1974–2014.
Study Characteristics
Table 1 shows a summary of the characteristics of the studies eligible for inclusion in the meta-analysis. Seven studies were conducted in Europe, 5 in the United States, and 1 in Australia. All studies examined change in weight (n = 10) or body mass index (n = 3) as the exposure of interest; none examined change in waist or hips circumference, and all presented results separately for men and women. Eight studies assessed change retrospectively by using self-reported weights in early adulthood and, of these, 3 studies measured weight at baseline. Five studies assessed change prospectively; of these, 3 studies measured weight at both waves of data collection. Thygesen et al. used self-reported weight assessments, recorded prospectively. Steins Bisschop et al. measured weight at baseline and self-reported weight 5 years after baseline. Three studies presented results for colon and rectal cancers combined, 6 studies presented results separately for colon and rectal cancers, and the remaining 4 studies presented results for colon cancer only.
Web Table 1 http://aje.oxfordjournals.org/content/181/11/832/suppl/DC1 available at http://aje.oxfordjournals.org/ shows the hazard ratios and corresponding 95% confidence intervals for each study for weight/body mass index change and the risk of colorectal cancer for the categorical and dose-response associations.
Largest Category of Weight Gain Compared With the Reference Category and the Risk of Colorectal Cancer
Thirteen prospective studies were included in this meta-analysis and the multivariable-adjusted hazard ratios comparing the largest weight gain category with the reference category for each study, and all studies combined are presented in Figure 2. There was evidence of an increased risk of colorectal cancer for the largest weight gain group, such that the risk of colorectal cancer was 16% higher for those in the highest weight gain category compared with those in the reference group (hazard ratio (HR) = 1.16, 95% CI: 1.08, 1.24). Studies that assessed change from midlife to older age (Figure 2B) had a lower hazard ratio compared with studies examining change from early adulthood to midlife (Figure 2A) (ratio of HRs = 0.83, 95% CI: 0.71, 0.97; P = 0.02). Studies that included physical activity in the analysis had higher hazard ratios than studies that did not (ratio of HRs = 1.21, 95% CI: 1.03, 1.42; P = 0.02) (Table 2).
(Enlarge Image)
Figure 2.
Adjusted hazard ratio (HR) for the risk of colorectal cancer comparing the largest weight gain group with the reference group for change measured from early adulthood to midlife (A) and change measured from midlife to older age (B) for males (squares) and females (circles), 1997–2014. Overall estimate reflects a combined hazard ratio from parts A and B. Dashed line, overall estimate; bars, 95% confidence interval (CI).
There was little heterogeneity across the studies (I = 10.2%, P = 0.30 from the χ test for heterogeneity) (Table 2). There did not appear to be any small-study effects; the corresponding funnel plot appeared symmetrical (Figure 3), and there was weak evidence of small-study effects from Egger's regression asymmetry test (P = 0.62).
(Enlarge Image)
Figure 3.
Funnel plot of the studies included in the meta-analysis comparing the largest weight gain group with the reference group for the risk of colorectal cancer, 1997–2014. The x-axis is on the log scale. Continuous line, no effect; dashed line, upper and lower 95% confidence limits.
Weight Loss Compared With the Reference Category and the Risk of Colorectal Cancer
Figure 4 shows the results for the meta-analysis comparing weight loss with the reference group. Of the 13 studies, 10 studies were included in this meta-analysis; 3 studies did not present results for a weight-loss group compared with a weight-stable or a reference group. There was no association between weight loss compared with a reference or stable group and the risk of colon or rectal cancer (HR = 0.96, 95% CI: 0.89, 1.05), regardless of whether change was assessed from early adulthood to midlife (Figure 4A) or from midlife to an older age (Figure 4B) (ratio of HRs = 1.05, 95% CI: 0.88, 1.26; P = 0.55).
(Enlarge Image)
Figure 4.
Adjusted hazard ratio (HR) for the risk of colorectal cancer comparing the largest weight loss group with the reference group for change measured from early adulthood to midlife (A) and change measured from midlife to older age (B) for males (squares) and females (circles), 1997–2014. Overall estimate reflects a combined hazard ratio from parts A and B. Dashed line, overall estimate; bars, 95% confidence interval (CI).
There was little heterogeneity across the studies (I = 0.0%; P = 0.71 from the χ test for heterogeneity) (Table 3). The funnel plot appeared symmetrical (Figure 5), and Egger's regression asymmetry test did not suggest the presence of small-study effects (P = 0.30).
(Enlarge Image)
Figure 5.
Funnel plot of the studies included in the meta-analysis comparing the largest weight loss group with the reference group for the risk of colorectal cancer, 1997–2014. The x-axis is on the log scale. Continuous line, no effect; dashed line, upper and lower 95% confidence limits.
Dose-response Relationship for Weight Gain and Risk of Colorectal Cancer
Nine studies were included in the dose-response analysis. Rapp et al. and Samanic et al. presented results for change in body mass index, but because they did not present the mean height, we were unable to convert the hazard ratios for change in body mass index to hazard ratios for change in weight. Colditz and Coakley did not provide estimates for each category of weight change and, thus, a dose-response association could not be estimated. Han et al. presented results for percent change, which could not be converted to an absolute change.
The pooled hazard ratio was 1.03 (95% CI: 1.02, 1.05) per 5-kg weight gain (Figure 6), with no evidence of departure from linearity (P = 0.49). The pooled hazard ratios for weight gain from early adulthood to midlife (Figure 6A) and from midlife to older age (Figure 6B) were similar (ratio of HRs = 1.03, 95% CI: 0.95, 1.12; P = 0.51). There was some heterogeneity across the estimates (I = 41.6%; P = 0.02 from the χ test for heterogeneity).
(Enlarge Image)
Figure 6.
Adjusted hazard ratio (HR) for 5-kg weight gain and the risk of colorectal cancer for change measured from early adulthood to midlife (A) and change measured from midlife to older age (B) for males (squares) and females (circles), 1997–2014. Overall estimate reflects a combined hazard ratio from parts A and B. Dashed line, overall estimate; bars, 95% confidence interval (CI).
Table 4 shows the results of the meta-regression analyses for the prespecified covariates. Slightly stronger associations were found for men than for women (ratio of HRs = 1.03, 95% CI: 1.00, 1.07; P = 0.07). The pooled hazard ratios did not vary by cancer site or the method used to ascertain weight.
Visual inspection of the funnel plot did not indicate the presence of small-study effects, and Egger's regression asymmetry test did not suggest any small-study effects (P = 0.89) (Figure 7).
(Enlarge Image)
Figure 7.
Funnel plot of the studies included in the meta-analysis comparing a dose-response relationship between weight gain and the risk of colorectal cancer, 1997–2014. The x-axis is on the log scale. Continuous line, no effect; dashed line, upper and lower 95% confidence limits.
Sensitivity Analyses
We conducted 3 sensitivity analyses where we excluded the estimates from the studies by Steins Bisschop et al., Renehan et al., and Oxentenko et al.; these did not change the results (data not shown).
Source...