Hostname: page-component-7c8c6479df-ws8qp Total loading time: 0 Render date: 2024-03-27T15:40:42.169Z Has data issue: false hasContentIssue false

Glycaemic index and glycaemic load in relation to risk of diabetes-related cancers: a meta-analysis

Published online by Cambridge University Press:  18 October 2012

Yuni Choi
Affiliation:
Women's Health Research Institute, Sookmyung Women's University, Seoul, Republic of Korea Department of Food and Nutrition, Sookmyung Women's University, 52 Hyochangwon-gil, Youngsan-gu, Seoul140-742, Republic of Korea
Edward Giovannucci
Affiliation:
Departments of Nutrition and Epidemiology, Harvard School of Public Health, Boston, MA, USA Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA, USA
Jung Eun Lee*
Affiliation:
Women's Health Research Institute, Sookmyung Women's University, Seoul, Republic of Korea Department of Food and Nutrition, Sookmyung Women's University, 52 Hyochangwon-gil, Youngsan-gu, Seoul140-742, Republic of Korea
*
*Corresponding author: J. E. Lee, fax +82 2 710 9479, email junglee@sm.ac.kr
Rights & Permissions [Opens in a new window]

Abstract

Diets high in glycaemic index (GI) or glycaemic load (GL) have been hypothesised to increase the risks of certain cancers by increasing blood glucose or insulin concentrations. We aimed to conduct a meta-analysis of prospective cohort studies to evaluate the association between GI or GL and diabetes-related cancers (DRC), including bladder, breast, colon–rectum, endometrium, liver and pancreas, which are associated with an increased risk for diabetes, and prostate cancer, which is associated with a reduced risk for diabetes. We searched Pubmed, EMBASE and MEDLINE databases up to September 2011 and reference lists of relevant articles. Relative risks (RR) and 95 % CI for the highest v. the lowest categories were extracted and pooled using a random-effects model. Thirty-six prospective cohort studies with a total of 60 811 DRC cases were included in the present meta-analysis. In a comparison of the highest and lowest categories, the pooled RR of DRC were 1·07 (95 % CI 1·04, 1·11; n 30) for GI and 1·02 (95 % CI 0·96, 1·08; n 33) for GL. In an analysis of site-specific cancer risks, we found significant associations for GI in relation to breast cancer (RR 1·06; 95 % CI 1·02, 1·11; n 11) and colorectal cancer (RR 1·08; 95 % CI 1·00, 1·17; n 9 studies). GL was significantly associated with the risk of endometrial cancer (RR 1·21; 95 % CI 1·07, 1·37; n 5). In conclusion, the findings of the present study suggest a modest-to-weak association between a diet that induces a high glucose response and DRC risks.

Type
Review Article
Copyright
Copyright © The Authors 2012

Diabetes and cancer are common chronic diseases that have contributed to many deaths worldwide. Recently, a consensus report of experts(Reference Giovannucci, Harlan and Archer1) representing the American Diabetes Association and the American Cancer Society reviewed the relationship between diabetes and cancers, and suggested that individuals with diabetes (primarily type 2) are more susceptible to developing cancers of the liver, pancreas, endometrium, colon/rectum, breast and bladder; however, they also have a lower risk of prostate cancer. This consensus report also discussed several possible biological mechanisms that may explain the direct link between diabetes and cancers, such as hyperinsulinaemia, hyperglycaemia and inflammation, but an explanation remains elusive.

The glycaemic index (GI) is, by definition, a unit of measurement used to rank carbohydrate-containing foods (scores ranging from 0 to 100) based on the postprandial blood glucose response compared with an equivalent amount of carbohydrate from a reference food (either glucose or white bread)(Reference Jenkins, Wolever and Taylor2). A related measure, the glycaemic load (GL), of a serving of a specific food is the product of the GI and the grams of carbohydrate content in a serving of a food, reflecting both the quality and quantity of dietary carbohydrates(Reference Foster-Powell, Holt and Brand-Miller3Reference Sheard, Clark and Brand-Miller5). Validation studies have shown that GL may be applicable to measuring degrees of overall postprandial plasma glucose and insulin response(Reference Brand-Miller, Thomas and Swan6, Reference Bao, Atkinson and Petocz7). Prospective cohort studies have shown that high-GI or -GL diets are associated with increased risks of adverse health outcomes, including CHD(Reference Beulens, de Bruijne and Stolk8), the metabolic syndrome(Reference Finley, Barlow and Halton9) and type 2 diabetes(Reference Salmeron, Ascherio and Rimm10, Reference Salmeron, Manson and Stampfer11) compared with low GI or GL. Additionally, a number of epidemiological studies have been conducted for the associations between GI or GL and the risk of common cancer sites, although results have been mixed (generally positive or null, not showing a clear association). Given the evidence for the potential link between GI or GL and cancer risks, presumably through effects of a diet stimulating postprandial glucose or insulin response, it is important to evaluate the hypothesis that GI and GL can be potential predictive factors for cancer risks, particularly those related to high levels of blood glucose or insulin.

We therefore assessed the associations between GI or GL and diabetes-related cancers (DRC) in a meta-analysis of observational prospective cohort studies. We did not include case–control studies because recall and selection bias is often encountered in case–control studies of diet and cancer risk.

Materials and methods

Literature search

A single author (Y. C.) conducted a literature search of the published studies using Pubmed, EMBASE and MEDLINE databases up to September 2011, and another author (J. E. L.) checked the extracted studies. The keyword ‘glycemic index’ OR ‘glycemic load’ was combined with the following search terms in each turn: (1) ‘liver cancer’ OR ‘liver neoplasm’ OR ‘liver carcinoma’ OR ‘hepatocellular carcinoma’; (2) ‘pancreas’ OR ‘pancreatic cancer’ OR ‘pancreatic neoplasm’ OR ‘pancreatic carcinoma’; (3) ‘endometrium’ OR ‘endometrial cancer’ OR ‘endometrial neoplasm’ OR ‘corpus uteri’ OR ‘endometrial carcinoma’; (4) ‘colorectal cancer” OR ‘colon cancer’ OR ‘rectal cancer’ OR ‘colorectal neoplasm’ OR ‘colorectal carcinoma’; (5) ‘breast cancer’ OR ‘breast carcinoma’ OR ‘breast neoplasm’; (6) ‘bladder cancer’ OR ‘bladder neoplasm’ OR ‘bladder carcinoma’; (7) ‘prostate cancer’ OR ‘prostate neoplasm’ OR ‘prostate carcinoma’. We also reviewed the reference lists of the retrieved articles to identify additional studies.

Study selection

Studies were included if they met the following criteria: (1) a cohort with GI or GL as an exposure and cases of DRC as an outcome were described; (2) estimates of relative risk (RR) and corresponding 95 % CI were provided; and (3) it was published in English. If multiple publications from one study were found, the most recent study was included in the present meta-analysis.

Data extraction

The following data were retrieved from the publications: the first author's last name, the year of publication, the sex of the participants, the study name, the country where the study was performed, the duration of follow-up, the age at baseline, the number of cases, the sample size, the dietary assessment, the comparison level (the highest intake category v. the lowest) of the GI and GL and confounding factors included in the multivariable-adjusted model. For each study, we used the most fully adjusted RR in the multivariate model. The two authors (Y. C. and J. E. L.) independently assessed the study quality using the Newcastle–Ottawa Scale, which ranged from 1 to 9 stars (poor to excellent, respectively)(Reference Wells, Shea and O'Connell12), and disagreements were resolved through consensus.

Statistical analysis

As a primary analysis, we pooled the RR estimates by comparing the highest intake category with the lowest from each study for site-specific cancer risks and overall DRC risk according to the GI and GL. The pooled RR estimates with corresponding 95 % CI were derived using a random-effects analysis, which considers both within- and between-study variance components(Reference DerSimonian and Laird13). For studies that only provided separate estimates by sex(Reference George, Mayne and Leitzmann14Reference Patel, McCullough and Pavluck18) and the one study that reported breast cancer risk by menopausal status(Reference Holmes, Liu and Hankinson19), we included the results separately. Studies describing relationships between the GI or GL and prostate cancer, which is considered to be inversely associated with diabetes(Reference Giovannucci, Harlan and Archer1), were not included in the main analysis to examine the hypothesis that the GI or GL was positively associated with some cancers that are associated with an increased risk for diabetes. The statistical heterogeneity between the studies was tested with Q statistics and I 2 statistics(Reference Higgins and Thompson20). We also evaluated the non-linearity of the association using restricted cubic splines(Reference Durrleman and Simon21Reference Orsini, Li and Wolk23) for studies that provided the number of participants or person-years and two or more categories of GI or GL intake. We conducted sensitivity analyses by omitting each study, one at a time, to evaluate whether the pooled estimates were influenced substantially by any single study. For studies(Reference Nielsen, Olsen and Christensen24) that provided RR as continuous variables only, we recalculated them into estimates per ten increments in GI and per 100 increments in GL (treated as top and bottom estimates) and then pooled them with categorical variables in the additional analysis. We also performed subgroup analyses (highest intake v. lowest intake) and random-effects meta-regression analyses to explore potential sources of heterogeneity between the studies by selected study characteristics, including the cancer site, the geographic region (North America, Europe and Asia), sex (males, females and both sexes), obesity status ( < 25 v. ≥ 25 kg/m2), study quality and the exclusion of diabetic individuals. Potential publication bias was assessed with Egger's regression asymmetry test(Reference Egger, Davey Smith and Schneider25). P < 0·05 was considered statistically significant. All analyses were two-sided and performed using STATA software (version 11; StataCorp) and SAS statistical software, version 9.2 (SAS Institute).

Results

Study characteristics

Fig. 1 shows the detailed literature search steps. The preliminary literature search resulted in the retrieval of 235 articles. Of these, thirty-nine articles and three additional relevant articles were considered to be of interest for the full-text review. After the full-text review, six articles were excluded for the reasons described in the flowchart and thirty-six were included in the final meta-analysis. The characteristics of the included articles are shown in Table 1. Overall, the meta-analysis for GI was based on thirty-three prospective cohort studies (n 44 RR estimates) and the meta-analysis for GL was based on thirty-six prospective cohort studies (n 48 RR estimates). The total number of cases of DRC was 60 811 (bladder, n 1481; breast, n 26 551; colon–rectum, n 16 793; endometrium, n 3200; liver, n 310; pancreas, n 3272; and prostate, n 9204), with a mean follow-up duration ranging from 5 to 21 years. To assess the habitual diet, all studies used either self- or interviewer-administered FFQ that included sixty-one(Reference Michaud, Fuchs and Liu17, Reference Holmes, Liu and Hankinson19, Reference Michaud, Liu and Giovannucci26) to 208 food items(Reference Lajous, Boutron-Ruault and Fabre27). For dietary GI intake, twenty-seven studies used a single dietary assessment and four studies used a cumulative average dietary assessment. For dietary GL intake, thirty studies used a single dietary assessment and four studies used a cumulative average dietary assessment. All studies were given a score of 7 or 8 stars, representing the high quality of studies of the studies, twenty-six were conducted in North America, seven in Europe and two in Asia.

Fig. 1 Flowchart of the study selection process in the meta-analysis. GI, glycaemic index; GL, glycaemic load.

Table 1 Characteristics of the studies included in the meta-analysis

M, males; F, females; NIH-AARP, National Institutes of Health–American Association of Retired Persons; FFQ(s), self-reported FFQ; FH, family history; PA, physical activity; TEI, total energy intakes; HRT, hormone replacement therapy; WHI, Women's Health Initiative; AAM, age at menarche; SWHS, Shanghai Women's Health Study; FFQ(I), interviewed FFQ; SMC, Swedish Mammography Cohort; MGEN, Mutuelle Générale de l'Education Nationale; ORDET, Ormoni e Dieta nella Eziologia dei Tumori (Hormones and Diet in the Etiology of Breast Cancer); NBSS, National Breast Screening Study; DDCHS, Danish Diet, Cancer, and Health Study; NHS, Nurses' Health Study; WHS, Women's Health Study; CPSII, Cancer Prevention Study II; MEC, Multiethnic Cohort study; Q, quartiles or quintiles; ref, reference; NLCS, Netherlands Cohort Study; BCDDP, Breast Cancer Detection Demonstration Project; NSAID, non-steroidal anti-inflammatory drug; IWHS, Iowa Women's Health Study; HPFS, Health Professional Follow-Up Study; EPIC, European Prospective Investigation into Cancer and Nutrition; EU, European Union; NA, not available; C, males and females; PLCO, Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial; PSA, prostate-specific antigen.

* Value expressed as mean, median value or range corresponding to the category (g/d).

Value expressed as mean or median; baseline and end of follow-up years given for the studies by George et al. (Reference George, Mayne and Leitzmann14) and Nimptsch et al. (Reference Nimptsch, Kenfield and Jensen54).

Value expressed as mean.

Glycaemic index and glycaemic load intake associated with the risk of overall diabetes-related cancers and each cancer site

Figs. 2 and 3 show the associations between DRC risk and either GI or GL, respectively, when comparing the highest with the lowest category intake.

Fig. 2 The pooled relative risks (RR) and 95 % CI of the glycaemic index in association with diabetes-related cancer and each cancer site. The pooled RR estimates were obtained using a random-effects model. On the x axis, the centre of each square indicates the RR of the study with its corresponding 95 % CI (the horizontal line). The size of the indicates the relative sample sizes in each study. The ◆ indicates the pooled RR estimates for each cancer site and the ◇ at the bottom indicates the pooled RR estimate for total cancers. F, females; M, males; pre-, premenopausal status; post-, postmenopausal status; C, both sexes.

Fig. 3 The pooled relative risks (RR) and 95 % CI of the glycemic load in association with diabetes-related cancer and each cancer site. The pooled RR estimates were obtained using a random-effects model. On the x axis, the centre of each square indicates the RR of the study with its corresponding 95 % CI (the horizontal line). The size of the indicates the relative sample sizes in each study. The ◆ indicate the pooled RR estimates for each cancer site, and the ◇ at the bottom indicates the pooled RR estimate for total cancers. F, females; M, males; pre-, premenopausal status; post-, postmenopausal status; C, both sexes.

Overall diabetes-related cancer risk

We combined thirty prospective studies of GI and thirty-three studies of GL that examined associations with potential diabetes-induced cancers, including bladder cancer(Reference George, Mayne and Leitzmann14), breast cancer(Reference George, Mayne and Leitzmann14, Reference Holmes, Liu and Hankinson19, Reference Lajous, Boutron-Ruault and Fabre27Reference Cho, Spiegelman and Hunter35), colorectal cancer(Reference George, Mayne and Leitzmann14Reference Michaud, Fuchs and Liu17, Reference Li, Yang and Shu36Reference Terry, Jain and Miller42), endometrial cancer(Reference George, Mayne and Leitzmann14, Reference Cust, Slimani and Kaaks43Reference Folsom, Demissie and Harnack46), liver cancer(Reference George, Mayne and Leitzmann14) and pancreatic cancer(Reference George, Mayne and Leitzmann14, Reference Patel, McCullough and Pavluck18, Reference Michaud, Liu and Giovannucci26, Reference Simon, Shikany and Neuhouser47Reference Johnson, Anderson and Harnack52). We pooled three risk estimates of prostate cancer(Reference George, Mayne and Leitzmann14, Reference Shikany, Flood and Kitahara53, Reference Nimptsch, Kenfield and Jensen54)separately, which could possibly be lower among diabetic individuals(Reference Giovannucci, Harlan and Archer1). When comparing the highest intake category with the lowest category, the pooled multivariable-adjusted RR of the overall DRC risk were 1·07 (95 % CI 1·04, 1·11) for GI, with no evidence of heterogeneity (P = 0·36, I 2 = 6·1 %) across studies, and 1·02 (95 % CI 0·96, 1·08) for GL, with modest heterogeneity (P < 0·001, I 2 = 45·4 %). Egger's regression test showed no evidence of a publication bias for GI (P = 0·99) or GL (P = 0·54). When we added one more study(Reference Nielsen, Olsen and Christensen24) that provided RR for continuous GI or GL, RR comparing the highest with the lowest category were 1·07 (95 % CI 1·03, 1·10) for GI and 1·02 (95 % CI 0·97, 1·08) for GL. When we evaluated whether there were non-linear relationships between the GI or GL and overall DRC risks, we found modestly suggestive evidence of non-linearity for GI (P = 0·06) or GL (P = 0·21) intakes.

The subgroup analyses, meta-regression and sensitivity analysis were performed on the associations of the overall DRC risk in relation to the GI and GL (highest v. lowest intake; Table 2). In the meta-regression analyses, we could not find any evidence of between-study heterogeneity in the overall risk estimates by cancer site (P for difference: P = 0·23 for GI, P = 0·06 for GL), geographic location (P for difference: P = 0·99 for GI, P = 0·22 for GL), sex (P for difference: P = 0·13 for GI, P = 0·64 for GL) or obesity (P for difference: P = 0·59 for GI, P = 0·37 for GL). When we examined whether associations differed by contrast in levels in the comparison categories, we found similar associations, with a pooled RR of 1·08 (95 % CI, 1·02, 1·13) for a difference of ≥ 12 in the GI and 1·06 (1·00–1·14) for a difference of < 12 in the GI. With respect to the GL, the pooled RR were 0·98 (0·91–1·06) for a difference of ≥ 65 and 1·07 (0·98–1·16) for a difference of < 65. Additionally, the associations did not vary by study quality (P for difference: P ≤0·6 for GI or GL) or the exclusion of diabetic individuals (P for difference: P ≤ 0·6 for GI or GL). Although there were no statistically significant differences, a significant positive association between GI and overall DRC risk was more apparent in the thirty-three combined estimates conducted in North America (RR 1·07; 95 % CI 1·03, 1·12) compared to other regions. A significant positive association between GI and overall DRC risk was observed in the twenty-seven combined estimates for women (RR 1·06; 95 % CI 1·02, 1·10; n 34 RR estimates). In the sensitivity analyses, where one study was omitted at a time, no particular study unduly influenced the pooled RR estimates for overall cancer sites or the P for heterogeneity.

Table 2 Subgroup analysis and meta-regression for the effects of characteristics on diabetes-related cancer risk* (Relative risks (RR) and 95 % confidence intervals)

* Prostate cancer was excluded from the analysis.

The number of RR estimates.

All pooled RR estimates for the comparison of the highest v. lowest categories were calculated from random- effects model.

§ P value for test of heterogeneity.

The analysis included studies that assessed the associations by BMI; data were available for cancers of breast, colon–rectum, pancreas and endometrium.

Bladder cancer risk

One large prospective study of older US adults examined the associations between GI or GL and bladder cancer risk. The RR (95 % CI) for men and women combined were 1·14 (0·82, 1·58) for GI and 0·97 (0·73, 1·31) for GL. Notably, a significant positive association between GI and bladder cancer for the comparisons of the highest with the lowest category of intake was found among men but not women. The author speculated, however, that there may be a residual confounding effect of smoking in this association. We were not able to test for a publication bias due to limited number of studies.

Breast cancer risk

Eleven studies evaluated GI and GL intake in relation to breast cancer risk. The majority of the prospective studies were conducted in North America and Europe; only one study was conducted in China. The meta-analysis suggested that the highest GI intake was associated with a 6 % relative increase in breast cancer risk compared with the lowest intake (95 % CI 1·02, 1·11); however, no significant associations were found between GL and breast cancer risk (RR 1·04; 95 % CI 0·96, 1·12). Statistical heterogeneity was not observed for GI (P = 0·64, I 2 = 0 %), but it was observed for GL (P = 0·03, I 2 = 49·2 %). Although heterogeneity was observed in the association between GL and breast cancer risk, this heterogeneity disappeared when a study by Sieri et al. (Reference Sieri, Pala and Brighenti31) was omitted, resulting in a pooled estimate of 1·03 (95 % CI 0·97, 1·08; P = 0·52, I 2 = 0 %). We found no indications of a publication bias for either GI (P = 0·75) or GL (P = 0·41) using Egger's test. In an additional analysis, where we added one more study(Reference Nielsen, Olsen and Christensen24) that provided RR for continuous GI or GL, RR for breast cancer comparing the highest with the lowest category were 1·06 (95 % CI 1·01, 1·10) for GI and 1·04 (95 % CI 0·97, 1·11) for GL. In a further subgroup analysis of breast cancer by menopausal status(Reference Holmes, Liu and Hankinson19, Reference Lajous, Boutron-Ruault and Fabre27, Reference Wen, Shu and Li29, Reference Sieri, Pala and Brighenti31Reference Higginbotham, Zhang and Lee33), the pooled RR were 1·05 (95 % CI 0·83, 1·33) for GI and 1·28 (95 % CI 0·94, 1·75) for GL among pre-menopausal woman, and 1·07 (95 % CI 0·88, 1·28) for GI and 1·11 (95 % CI 0·91, 1·36) for GL among post-menopausal woman.

Colorectal cancer risk

Nine studies for GI and eleven studies for GL were included in the meta-analysis of colorectal cancer risk. The meta-analysis of nine prospective studies showed a borderline positive association between GI and colorectal cancer risk when comparing the highest with the lowest category of intake (RR 1·08; 95 % CI 1·00, 1·17); there was no evidence of heterogeneity (P = 0·16, I 2 = 29·4 %) or a publication bias (P = 0·44). No significant association was observed between GL intake and colorectal cancer risk (RR 0·99; 95 % CI 0·90, 1·09) with modest heterogeneity (P = 0·04, I 2 = 42·2 %) and evidence of a publication bias (P = 0·03). When a study by Higginbotham et al. (Reference Higginbotham, Zhang and Lee41) was omitted from the analysis of GL intake and colon–rectal cancer, however, no heterogeneity was found among the remaining ten studies (P = 0·28, I 2 = 15·9 %), and the pooled RR for the highest v. lowest category of GL intake was 0·97 (95 % CI 0·90, 1·05).

Endometrial cancer risk

Five prospective studies examined the association of endometrial cancer with either GI or GL intake. There was no significant association with endometrial cancer for the highest v. the lowest GI intake (RR 1·00; 95 % CI 0·87, 1·14). In contrast, the highest category of GL intake was significantly associated with a 21 % greater risk of developing endometrial cancer compared with the lowest category of intake (95 % CI 1·07, 1·37). No statistical heterogeneity among studies was observed for either GI (P = 0·28, I 2 = 21·6 %) or GL (P = 0·91, I 2 = 0 %); we also found no evidence of a publication bias using Egger's test (P = 0·08 for GI and P = 0·23 for GL).

Liver cancer risk

Only one longitudinal study (National Institutes of Health – American Association of Retired Persons) evaluated the association of GI or GL intake with the risk of liver cancer. A combined analysis of men and women together showed no evidence of an association between GI intake and liver cancer risk. In contrast, the comparison of the highest v. lowest category of GL intake showed a significant reduction in the risk of liver cancer (RR 0·37; 95 % CI 0·16, 0·84). We could not test for a publication bias due to the limited number of studies.

Pancreatic cancer risk

With regard to pancreatic cancer, eight studies for GI and nine studies for GL were conducted in North America. There was no association between the GI and pancreatic cancer risk when comparing the highest with the lowest intake (RR 1·05; 95 % CI 0·93, 1·19), with little evidence of heterogeneity (P = 0·87, I 2 = 0 %). Similarly, no association was found for the highest GL intake compared with lowest intake (RR 0·95; 95 % CI 0·79, 1·13; P for heterogeneity P = 0·06, I 2 = 43·5 %).

Prostate cancer risk

We pooled three RR estimates of prostate cancer, the risk of which has been suggested to be lower among diabetic individuals, and found no significant associations. The pooled RR were 0·97 (95 % CI 0·91, 1·04) for GI and 0·90 (95 % CI 0·74, 1·11) for GL when comparing the highest with lowest category of intake. There was no evidence of a publication bias, as determined by Egger's regression test for GI (P = 0·99) or GL (P = 0·54).

Discussion

To our knowledge, the present study is the first systematic literature review and meta-analysis of the association between the risks of DRC and the GI or GL. The present results from the meta-analysis of prospective cohort studies suggest that high GI was modestly associated with overall DRC risks, whereas high GL was not related to overall DRC risks. In the cancer-specific analysis, we found that high GI was weakly, but significantly, associated with an increased risk of breast or colorectal cancer. We also found that high GL was significantly associated with an increased risk of developing endometrial cancer.

We could draw several inferences from other studies that may be likely explanations for the modest-to-weak associations between GI or GL and DRC risks observed in the present study. Of several aetiological hypotheses on cancer, insulin resistance, hyperinsulinemia and an increased level of insulin-like growth factor-I may be most probable as currently understood, as they have been implicated as key mediators in the underlying mechanism relating dietary and associated lifestyle factors to carcinogenesis(Reference Giovannucci, Harlan and Archer1, Reference Renehan, Zwahlen and Minder55). Elevated circulating insulin levels could promote carcinogenesis, either directly by stimulating the production of insulin receptors or indirectly by suppressing insulin-like growth factor-binding proteins 1 and 3, which may increase the bioavailability of insulin-like growth factor-I for its receptors(Reference Cohen, Peehl and Lamson56). Growing evidence from epidemiological studies has supported the mechanisms described earlier. Elevated levels of insulin (or C-peptide as a surrogate) and insulin-like growth factor-I have also been associated with an increased risk of several DRC cancers(Reference Ma, Giovannucci and Pollak57Reference Gunter, Hoover and Yu60). The increasing evidence for an association between hyperinsulinaemia and cancer risk has led to interest in examining factors that increase insulin in relation to various cancers(Reference Giovannucci and Michaud61). Epidemiological studies have shown an elevated risk of cancers with factors that increase insulin levels and a reduced risk of cancers with factors related to decreased insulin levels: increased cancer risks among individuals who had obesity (or visceral obesity)(Reference Moghaddam, Woodward and Huxley62) or consumed a C-peptide dietary pattern(Reference Fung, Hu and Schulze63) or western dietary patterns enriched in fat and red meat(Reference Schulz, Hoffmann and Weikert64, Reference Giovannucci, Rimm and Stampfer65), and reduced cancer risks among those with high physical activity(Reference Harriss, Atkinson and Batterham66, Reference Moore, Gierach and Schatzkin67) or who were coffee drinkers(Reference Yu, Bao and Zou68, Reference Wu, Willett and Hankinson69).

The lack of association or modest association for GI and GL in the present meta-analysis may suggest that a mechanism linking insulin to cancer development could be more plausible than the effect of blood glucose on cancer development(Reference Giovannucci, Harlan and Archer1). Experimental study(Reference Heuson and Legros70) has shown that rats that were hyperglycaemic and insulin deficient, a condition similar to human type 1 diabetes, had reduced tumour cell proliferation, as assessed by the size, number and aggressiveness of the tumour. The differences in cancer development between type 1 and 2 diabetes partly support this hypothesis. Hyperglycaemia occurs in both type 1 and 2 diabetes, but insulin resistance and endogenous hyperinsulinaemia are only observed in type 2 diabetes(71). Cancers frequently observed in association with type 2 diabetes are bladder, breast, colorectal, endometrial, liver and pancreatic cancers(Reference Giovannucci, Harlan and Archer1), and those associated with type 1 diabetes are stomach and squamous cell skin carcinomas and leukaemia(Reference Shu, Ji and Li72). The finding that no association was identified with colorectal cancer among those who have been diagnosed with type 2 diabetes for more than 15 years(Reference Hu, Manson and Liu73), possibly under a condition of insulin deprivation (the Starling Curve of the pancreas)(Reference DeFronzo74), could also support the hypothesis that hyperinsulinaemia may be a more important contributor to tumour development than hyperglycaemia. The dietary insulin index is another approach that has been recently developed to directly quantify the postprandial plasma insulin secretion compared with a reference food(Reference Holt, Miller and Petocz75), and it has been found to be more precise in assessing the insulin response than the GL or carbohydrate amount(Reference Bao, de Jong and Atkinson76). The evidence that postprandial insulin concentrations do not change proportionally with the blood glucose response(Reference Holt, Miller and Petocz75) and that GI or GL, measures of the carbohydrates in blood glucose levels, may not ideally predict insulin secretion through the consumption of no or low carbohydrate-containing food, may suggest that the dietary insulin index is a more acceptable measure for assessing insulin secretion and cancer risk. Only a few studies, however, have been conducted regarding insulin index/load or C-peptide, an indicator of insulin production(Reference Bao, de Jong and Atkinson76), in relation to cancer risks, which warrants further observational studies to investigate indicators reflecting insulin secretion and its effect on cancer risks.

Other conceptual and practical considerations may contribute to the weak association observed in the present study. From a conceptual perspective, GI and GL may be relatively moderate contributors to overall insulin exposure, which is influenced by genetic factors, adiposity level, physical activity and non-carbohydrate components in foods that influence insulin secretion, and dietary factors, such as coffee, that influence insulin resistance but not insulin secretion directly. From a practical perspective, the attenuation of associations could also be explained by a potential measurement error or between-study variation in the estimated amount of carbohydrates, as measured through the different types of FFQ, given the likely influence of heterogeneity by diets with a high GL. Although we could not clearly identify the sources of heterogeneity in the relationship between diets with a high GL and overall DRC risks, the significant heterogeneity observed in the association of diets with a high GL with risks of breast cancer and colorectal cancer disappeared after omitting individual studies from the breast cancer studies(Reference Sieri, Pala and Brighenti31) and from the colorectal studies(Reference Higginbotham, Zhang and Lee41) that had extreme values compared to the other studies. A better understanding requires further prospective cohort studies for each site of DRC.

A previous meta-analysis of both case–control and prospective cohort studies(Reference Gnagnarella, Gandini and La Vecchia77) observed increased risks of colorectal cancer, breast cancer and endometrial cancer with the highest v. the lowest levels of GI and/or GL diets. The present meta-analysis was restricted to observational prospective cohort studies because such a design minimises the possible effects of selection and recall bias compared to case–control studies, showing a weaker association for GI and GL in relation to colorectal, breast or endometrial cancers than a meta-analysis of both case–control and prospective cohort studies.

There are possible limitations to the present study. First, measurement error with regard to random variation in the estimated GI values might have occurred in some studies included in the present analysis because the GI values of some foods are presently based on the results provided in only one or two GI calculation studies, which frequently had small sample sizes(Reference Foster-Powell, Holt and Brand-Miller3). Second, because most of the studies assessed diets using a single FFQ, which may have contained a measurement error, the possibility of misclassification of GI or GL cannot be precluded(Reference Freedman, Potischman and Kipnis78). Third, the limited number of studies for certain cancer sites (e.g. liver and bladder cancer) did not allow us to draw conclusive summaries for those sites. Lastly, the majority of studies included in the present meta-analysis were conducted in Western countries, thus it is uncertain whether the present findings for different geographic locations or populations are generalised, especially in Asian populations whose typical diets on average consist of a greater proportion of carbohydrates. Further studies should provide information on the potential differences based upon geographic location or ethnic difference. However, the present study also has several major strengths, including the inclusion of many prospective studies with long durations of follow-up and a large number of cases of DRC. The present results were also unlikely to be attributed to publication bias with regard to GI or GL and DRC risk based on Egger's regression test.

In conclusion, the findings of the present meta-analysis suggest a modest or weak association between a diet inducing high glucose response and the risks of overall cancers, particularly those positively related to diabetes. GI or GL may not be strong predictors of DRC risks, and presumably other factors associated with insulin response per se may contribute relatively more to DRC risks. The present findings warrant further studies to explore a diet that stimulates the postprandial insulin response in relation to cancer risk.

Acknowledgements

The present study was supported by the National Research Foundation of Korea (NRF) grant, which was funded by the Korean government (no. 2012-0003287), and by the SRC Research Center for Women's Diseases of Sookmyung Women's University (2011). The authors’ responsibilities were as follows: Y. C. contributed to the concept and design of the study, data collection, statistical analysis and writing of the manuscript; E. G. provided invaluable advice and consultations during draft development and critically revised the manuscript for important intellectual content; and J. E. L. contributed to the concept and design of the study, provided guidance during study selection, data analysis, draft development and final submission. All authors were responsible for the critical interpretation of the findings, and all contributed to the final manuscript and gave final approval of the version to be published. None of the authors had a conflict of interest.

References

1Giovannucci, E, Harlan, DM, Archer, MC, et al. (2010) Diabetes and cancer: a consensus report. Diabetes Care 33, 16741685.CrossRefGoogle Scholar
2Jenkins, DJ, Wolever, TM, Taylor, RH, et al. (1981) Glycemic index of foods: a physiological basis for carbohydrate exchange. Am J Clin Nutr 34, 362366.Google Scholar
3Foster-Powell, K, Holt, SH & Brand-Miller, JC (2002) International table of glycemic index and glycemic load values: 2002. Am J Clin Nutr 76, 556.CrossRefGoogle ScholarPubMed
4Jenkins, DJ, Kendall, CW, Augustin, LS, et al. (2002) Glycemic index: overview of implications in health and disease. Am J Clin Nutr 76, 266S273S.CrossRefGoogle ScholarPubMed
5Sheard, NF, Clark, NG, Brand-Miller, JC, et al. (2004) Dietary carbohydrate (amount and type) in the prevention and management of diabetes: a statement by the American Diabetes Association. Diabetes Care 27, 22662271.CrossRefGoogle ScholarPubMed
6Brand-Miller, JC, Thomas, M, Swan, V, et al. (2003) Physiological validation of the concept of glycemic load in lean young adults. J Nutr 133, 27282732.Google Scholar
7Bao, J, Atkinson, F, Petocz, P, et al. (2011) Prediction of postprandial glycemia and insulinemia in lean, young, healthy adults: glycemic load compared with carbohydrate content alone. Am J Clin Nutr 93, 984996.Google Scholar
8Beulens, JW, de Bruijne, LM, Stolk, RP, et al. (2007) High dietary glycemic load and glycemic index increase risk of cardiovascular disease among middle-aged women: a population-based follow-up study. J Am Coll Cardiol 50, 1421.CrossRefGoogle ScholarPubMed
9Finley, CE, Barlow, CE, Halton, TL, et al. (2010) Glycemic index, glycemic load, and prevalence of the metabolic syndrome in the Cooper Center Longitudinal Study. J Am Diet Assoc 110, 18201829.CrossRefGoogle ScholarPubMed
10Salmeron, J, Ascherio, A, Rimm, EB, et al. (1997) Dietary fiber, glycemic load, and risk of NIDDM in men. Diabetes Care 20, 545550.Google Scholar
11Salmeron, J, Manson, JE, Stampfer, MJ, et al. (1997) Dietary fiber, glycemic load, and risk of non-insulin-dependent diabetes mellitus in women. JAMA 277, 472477.Google Scholar
12Wells, G, Shea, B & O'Connell, D, et al. (2012) The Newcastle–Ottawa Scle (NOS) for assessing the quality of nonrandomised sutdies in meta-analysis.http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp (accesses 6 June 2012).Google Scholar
13DerSimonian, R & Laird, N (1986) Meta-analysis in clinical trials. Control Clin Trials 7, 177188.Google Scholar
14George, SM, Mayne, ST, Leitzmann, MF, et al. (2009) Dietary glycemic index, glycemic load, and risk of cancer: a prospective cohort study. Am J Epidemiol 169, 462472.CrossRefGoogle ScholarPubMed
15Howarth, NC, Murphy, SP, Wilkens, LR, et al. (2008) The association of glycemic load and carbohydrate intake with colorectal cancer risk in the Multiethnic Cohort Study. Am J Clin Nutr 88, 10741082.Google Scholar
16Weijenberg, MP, Mullie, PF, Brants, HA, et al. (2008) Dietary glycemic load, glycemic index and colorectal cancer risk: results from the Netherlands Cohort Study. Int J Cancer 122, 620629.CrossRefGoogle ScholarPubMed
17Michaud, DS, Fuchs, CS, Liu, S, et al. (2005) Dietary glycemic load, carbohydrate, sugar, and colorectal cancer risk in men and women. Cancer Epidemiol Biomarkers Prev 14, 138147.Google Scholar
18Patel, AV, McCullough, ML, Pavluck, AL, et al. (2007) Glycemic load, glycemic index, and carbohydrate intake in relation to pancreatic cancer risk in a large US cohort. Cancer Causes Control 18, 287294.CrossRefGoogle Scholar
19Holmes, MD, Liu, S, Hankinson, SE, et al. (2004) Dietary carbohydrates, fiber, and breast cancer risk. Am J Epidemiol 159, 732739.Google Scholar
20Higgins, JP & Thompson, SG (2002) Quantifying heterogeneity in a meta-analysis. Stat Med 21, 15391558.Google Scholar
21Durrleman, S & Simon, R (1989) Flexible regression models with cubic splines. Stat Med 8, 551561.CrossRefGoogle ScholarPubMed
22Greenland, S & Longnecker, MP (1992) Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 135, 13011309.Google Scholar
23Orsini, N, Li, R, Wolk, A, et al. (2012) Meta-analysis for linear and nonlinear dose–response relations: examples, an evaluation of approximations, and software. Am J Epidemiol 175, 6673.Google Scholar
24Nielsen, TG, Olsen, A, Christensen, J, et al. (2005) Dietary carbohydrate intake is not associated with the breast cancer incidence rate ratio in postmenopausal Danish women. J Nutr 135, 124128.CrossRefGoogle Scholar
25Egger, M, Davey Smith, G, Schneider, M, et al. (1997) Bias in meta-analysis detected by a simple, graphical test. BMJ 315, 629634.Google Scholar
26Michaud, DS, Liu, S, Giovannucci, E, et al. (2002) Dietary sugar, glycemic load, and pancreatic cancer risk in a prospective study. J Natl Cancer Inst 94, 12931300.Google Scholar
27Lajous, M, Boutron-Ruault, MC, Fabre, A, et al. (2008) Carbohydrate intake, glycemic index, glycemic load, and risk of postmenopausal breast cancer in a prospective study of French women. Am J Clin Nutr 87, 13841391.Google Scholar
28Shikany, JM, Redden, DT, Neuhouser, ML, et al. (2011) Dietary glycemic load, glycemic index, and carbohydrate and risk of breast cancer in the Women's Health Initiative. Nutr Cancer 63, 899907.Google Scholar
29Wen, W, Shu, XO, Li, H, et al. (2009) Dietary carbohydrates, fiber, and breast cancer risk in Chinese women. Am J Clin Nutr 89, 283289.Google Scholar
30Larsson, SC, Bergkvist, L & Wolk, A (2009) Glycemic load, glycemic index and breast cancer risk in a prospective cohort of Swedish women. Int J Cancer 125, 153157.Google Scholar
31Sieri, S, Pala, V, Brighenti, F, et al. (2007) Dietary glycemic index, glycemic load, and the risk of breast cancer in an Italian prospective cohort study. Am J Clin Nutr 86, 11601166.Google Scholar
32Silvera, SA, Jain, M, Howe, GR, et al. (2005) Dietary carbohydrates and breast cancer risk: a prospective study of the roles of overall glycemic index and glycemic load. Int J Cancer 114, 653658.Google Scholar
33Higginbotham, S, Zhang, ZF, Lee, IM, et al. (2004) Dietary glycemic load and breast cancer risk in the Women's Health Study. Cancer Epidemiol Biomarkers Prev 13, 6570.Google Scholar
34Jonas, CR, McCullough, ML, Teras, LR, et al. (2003) Dietary glycemic index, glycemic load, and risk of incident breast cancer in postmenopausal women. Cancer Epidemiol Biomarkers Prev 12, 573577.Google ScholarPubMed
35Cho, E, Spiegelman, D, Hunter, DJ, et al. (2003) Premenopausal dietary carbohydrate, glycemic index, glycemic load, and fiber in relation to risk of breast cancer. Cancer Epidemiol Biomarkers Prev 12, 11531158.Google Scholar
36Li, HL, Yang, G, Shu, XO, et al. (2011) Dietary glycemic load and risk of colorectal cancer in Chinese women. Am J Clin Nutr 93, 101107.Google Scholar
37Kabat, GC, Shikany, JM, Beresford, SA, et al. (2008) Dietary carbohydrate, glycemic index, and glycemic load in relation to colorectal cancer risk in the Women's Health Initiative. Cancer Causes Control 19, 12911298.Google Scholar
38Strayer, L, Jacobs, DR Jr, Schairer, C, et al. (2007) Dietary carbohydrate, glycemic index, and glycemic load and the risk of colorectal cancer in the BCDDP cohort. Cancer Causes Control 18, 853863.Google Scholar
39Larsson, SC, Giovannucci, E & Wolk, A (2007) Dietary carbohydrate, glycemic index, and glycemic load in relation to risk of colorectal cancer in women. Am J Epidemiol 165, 256261.Google Scholar
40McCarl, M, Harnack, L, Limburg, PJ, et al. (2006) Incidence of colorectal cancer in relation to glycemic index and load in a cohort of women. Cancer Epidemiol Biomarkers Prev 15, 892896.Google Scholar
41Higginbotham, S, Zhang, ZF, Lee, IM, et al. (2004) Dietary glycemic load and risk of colorectal cancer in the Women's Health Study. J Natl Cancer Inst 96, 229233.CrossRefGoogle ScholarPubMed
42Terry, PD, Jain, M, Miller, AB, et al. (2003) Glycemic load, carbohydrate intake, and risk of colorectal cancer in women: a prospective cohort study. J Natl Cancer Inst 95, 914916.Google Scholar
43Cust, AE, Slimani, N, Kaaks, R, et al. (2007) Dietary carbohydrates, glycemic index, glycemic load, and endometrial cancer risk within the European Prospective Investigation into Cancer and Nutrition cohort. Am J Epidemiol 166, 912923.Google Scholar
44Larsson, SC, Friberg, E & Wolk, A (2007) Carbohydrate intake, glycemic index and glycemic load in relation to risk of endometrial cancer: a prospective study of Swedish women. Int J Cancer 120, 11031107.Google Scholar
45Silvera, SA, Rohan, TE, Jain, M, et al. (2005) Glycaemic index, glycaemic load and risk of endometrial cancer: a prospective cohort study. Public Health Nutr 8, 912919.Google Scholar
46Folsom, AR, Demissie, Z, Harnack, L, et al. (2003) Glycemic index, glycemic load, and incidence of endometrial cancer: the Iowa women's health study. Nutr Cancer 46, 119124.Google Scholar
47Simon, MS, Shikany, JM, Neuhouser, ML, et al. (2010) Glycemic index, glycemic load, and the risk of pancreatic cancer among postmenopausal women in the women's health initiative observational study and clinical trial. Cancer Causes Control 21, 21292136.Google Scholar
48Meinhold, CL, Dodd, KW, Jiao, L, et al. (2010) Available carbohydrates, glycemic load, and pancreatic cancer: is there a link? Am J Epidemiol 171, 11741182.CrossRefGoogle ScholarPubMed
49Heinen, MM, Verhage, BA, Lumey, L, et al. (2008) Glycemic load, glycemic index, and pancreatic cancer risk in the Netherlands Cohort Study. Am J Clin Nutr 87, 970977.CrossRefGoogle ScholarPubMed
50Nothlings, U, Murphy, SP, Wilkens, LR, et al. (2007) Dietary glycemic load, added sugars, and carbohydrates as risk factors for pancreatic cancer: the Multiethnic Cohort Study. Am J Clin Nutr 86, 14951501.CrossRefGoogle ScholarPubMed
51Silvera, SA, Rohan, TE, Jain, M, et al. (2005) Glycemic index, glycemic load, and pancreatic cancer risk (Canada). Cancer Causes Control 16, 431436.Google Scholar
52Johnson, KJ, Anderson, KE, Harnack, L, et al. (2005) No association between dietary glycemic index or load and pancreatic cancer incidence in postmenopausal women. Cancer Epidemiol Biomarkers Prev 14, 15741575.CrossRefGoogle ScholarPubMed
53Shikany, JM, Flood, AP, Kitahara, CM, et al. (2011) Dietary carbohydrate, glycemic index, glycemic load, and risk of prostate cancer in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) cohort. Cancer Causes Control 22, 9951002.Google Scholar
54Nimptsch, K, Kenfield, S, Jensen, MK, et al. (2011) Dietary glycemic index, glycemic load, insulin index, fiber and whole-grain intake in relation to risk of prostate cancer. Cancer Causes Control 22, 5161.CrossRefGoogle ScholarPubMed
55Renehan, AG, Zwahlen, M, Minder, C, et al. (2004) Insulin-like growth factor (IGF)-I, IGF binding protein-3, and cancer risk: systematic review and meta-regression analysis. Lancet 363, 13461353.Google Scholar
56Cohen, P, Peehl, DM, Lamson, G, et al. (1991) Insulin-like growth factors (IGFs), IGF receptors, and IGF-binding proteins in primary cultures of prostate epithelial cells. J Clin Endocrinol Metab 73, 401407.Google Scholar
57Ma, J, Giovannucci, E, Pollak, M, et al. (2004) A prospective study of plasma C-peptide and colorectal cancer risk in men. J Natl Cancer Inst 96, 546553.Google Scholar
58Michaud, DS, Wolpin, B, Giovannucci, E, et al. (2007) Prediagnostic plasma C-peptide and pancreatic cancer risk in men and women. Cancer Epidemiol Biomarkers Prev 16, 21012109.Google Scholar
59Endogenous Hormones and Breast Cancer Collaborative Group, Key, TJ, Appleby, PN, et al. (2010) Insulin-like growth factor 1 (IGF1), IGF binding protein 3 (IGFBP3), and breast cancer risk: pooled individual data analysis of 17 prospective studies. Lancet Oncol 11, 530542.Google Scholar
60Gunter, MJ, Hoover, DR, Yu, H, et al. (2008) A prospective evaluation of insulin and insulin-like growth factor-I as risk factors for endometrial cancer. Cancer Epidemiol Biomarkers Prev 17, 921929.Google Scholar
61Giovannucci, E & Michaud, D (2007) The role of obesity and related metabolic disturbances in cancers of the colon, prostate, and pancreas. Gastroenterology 132, 22082225.Google Scholar
62Moghaddam, AA, Woodward, M & Huxley, R (2007) Obesity and risk of colorectal cancer: a meta-analysis of 31 studies with 70,000 events. Cancer Epidemiol Biomarkers Prev 16, 25332547.Google Scholar
63Fung, TT, Hu, FB, Schulze, M, et al. (2012) A dietary pattern that is associated with C-peptide and risk of colorectal cancer in women. Cancer Causes Control 23, 959965.CrossRefGoogle ScholarPubMed
64Schulz, M, Hoffmann, K, Weikert, C, et al. (2008) Identification of a dietary pattern characterized by high-fat food choices associated with increased risk of breast cancer: the European Prospective Investigation into Cancer and Nutrition (EPIC)-Potsdam Study. Br J Nutr 100, 942946.Google Scholar
65Giovannucci, E, Rimm, EB, Stampfer, MJ, et al. (1994) Intake of fat, meat, and fiber in relation to risk of colon cancer in men. Cancer Res 54, 23902397.Google Scholar
66Harriss, DJ, Atkinson, G, Batterham, A, et al. (2009) Lifestyle factors and colorectal cancer risk (2): a systematic review and meta-analysis of associations with leisure-time physical activity. Colorectal Dis 11, 689701.Google Scholar
67Moore, SC, Gierach, GL, Schatzkin, A, et al. (2010) Physical activity, sedentary behaviours, and the prevention of endometrial cancer. Br J Cancer 103, 933938.Google Scholar
68Yu, X, Bao, Z, Zou, J, et al. (2011) Coffee consumption and risk of cancers: a meta-analysis of cohort studies. BMC Cancer 11, 96.Google Scholar
69Wu, T, Willett, WC, Hankinson, SE, et al. (2005) Caffeinated coffee, decaffeinated coffee, and caffeine in relation to plasma C-peptide levels, a marker of insulin secretion, in U.S. women. Diabetes Care 28, 13901396.Google Scholar
70Heuson, JC & Legros, N (1972) Influence of insulin deprivation on growth of the 7,12-dimethylbenz(a)anthracene-induced mammary carcinoma in rats subjected to alloxan diabetes and food restriction. Cancer Res 32, 226232.Google Scholar
71Expert Committee on the Diagnosis and Classification of Diabetes Mellitus (2003) Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Diabetes Care S5S20.Google Scholar
72Shu, X, Ji, J, Li, X, et al. (2010) Cancer risk among patients hospitalized for type 1 diabetes mellitus: a population-based cohort study in Sweden. Diabet Med 27, 791797.Google Scholar
73Hu, FB, Manson, JE, Liu, S, et al. (1999) Prospective study of adult onset diabetes mellitus (type 2) and risk of colorectal cancer in women. J Natl Cancer Inst 91, 542547.Google Scholar
74DeFronzo, RA (1988) Lilly lecture 1987. The triumvirate: beta-cell, muscle, liver. A collusion responsible for NIDDM. Diabetes 37, 667687.Google Scholar
75Holt, SH, Miller, JC & Petocz, P (1997) An insulin index of foods: the insulin demand generated by 1000-kJ portions of common foods. Am J Clin Nutr 66, 12641276.Google Scholar
76Bao, J, de Jong, V, Atkinson, F, et al. (2009) Food insulin index: physiologic basis for predicting insulin demand evoked by composite meals. Am J Clin Nutr 90, 986992.Google Scholar
77Gnagnarella, P, Gandini, S, La Vecchia, C, et al. (2008) Glycemic index, glycemic load, and cancer risk: a meta-analysis. Am J Clin Nutr 87, 17931801.CrossRefGoogle ScholarPubMed
78Freedman, LS, Potischman, N, Kipnis, V, et al. (2006) A comparison of two dietary instruments for evaluating the fat–breast cancer relationship. Int J Epidemiol 35, 10111021.Google Scholar
Figure 0

Fig. 1 Flowchart of the study selection process in the meta-analysis. GI, glycaemic index; GL, glycaemic load.

Figure 1

Table 1 Characteristics of the studies included in the meta-analysis

Figure 2

Fig. 2 The pooled relative risks (RR) and 95 % CI of the glycaemic index in association with diabetes-related cancer and each cancer site. The pooled RR estimates were obtained using a random-effects model. On the x axis, the centre of each square indicates the RR of the study with its corresponding 95 % CI (the horizontal line). The size of the indicates the relative sample sizes in each study. The ◆ indicates the pooled RR estimates for each cancer site and the ◇ at the bottom indicates the pooled RR estimate for total cancers. F, females; M, males; pre-, premenopausal status; post-, postmenopausal status; C, both sexes.

Figure 3

Fig. 3 The pooled relative risks (RR) and 95 % CI of the glycemic load in association with diabetes-related cancer and each cancer site. The pooled RR estimates were obtained using a random-effects model. On the x axis, the centre of each square indicates the RR of the study with its corresponding 95 % CI (the horizontal line). The size of the indicates the relative sample sizes in each study. The ◆ indicate the pooled RR estimates for each cancer site, and the ◇ at the bottom indicates the pooled RR estimate for total cancers. F, females; M, males; pre-, premenopausal status; post-, postmenopausal status; C, both sexes.

Figure 4

Table 2 Subgroup analysis and meta-regression for the effects of characteristics on diabetes-related cancer risk* (Relative risks (RR) and 95 % confidence intervals)