High glycemic index and glycemic load are associated with moderately increased cancer risk

1 Introduction

Carbohydrate foods increase postprandial glycemia and insulin secretion at different rates, depending on the types of the carbohydrates, the amount and nature of fiber contained, the processing method, and the presence of other nutrients. These differences are captured by the glycemic index (GI), a ranking system for carbohydrate foods based on their ability to raise blood glucose concentrations 1. Glycemic load (GL), the product of a food's GI and total available carbohydrate content, was introduced to incorporate the effect of the total amount of carbohydrate consumed. Therefore, the overall GI reflects the quality of carbohydrates consumed, whereas the total dietary GL reflects both the quantity and quality of carbohydrates 2.

Factors influencing the metabolism of glucose may have a relevant role in the development of chronic diseases, including not only diabetes and cardiovascular diseases, but also cancer at selected sites 3-8. In particular, chronically elevated insulin concentrations are thought to represent the underlying mechanism for the GI/GL-related increased risks of several cancers. Insulin acts as a growth factor and increases the bioavailability and bioactivity of insulin-like growth factors (IGFs), such as IGF-1, by enhancing their synthesis and by decreasing their binding proteins. IGF-1 can promote tumor development by inhibiting apoptosis, stimulating cell proliferation and sex-steroid synthesis 6, 9, 10. IGF-1 has also angiogenic proprieties. Moreover, other conditions related to glucose metabolism, including insulin resistance, hyperglycemia, obesity, and diabetes may influence the risk of coronary heart disease and cancer 6, 7, 11-13.

In 2008, a meta-analysis reviewed results from 39 observational studies considering the relation between dietary GI/GL and the risk of cancer, and reported significantly increased risks for the highest versus the lowest categories of intake for colorectal (summary relative risk, RR = 1.26 for GI, and RR = 1.18 for GL) and endometrial cancers (RR = 1.36 for GI, and RR = 1.22 for GL). No significant association was observed for pancreatic and breast cancer 14. A number of studies were published subsequently on the issue, and their results have been considered in some reviews and meta-analyses focused on a single type of cancer 15-23. In 2012, a meta-analysis of cohort studies summarized the evidence on GI and GL and diabetes-related cancers (i.e. bladder, breast, colorectum, endometrium, liver, and pancreas), and suggested, overall, a modestly increased risk 24. Analyses on site-specific cancers revealed significant positive associations between GI and breast (RR for the highest versus the lowest category = 1.06; 95% confidence interval, CI: 1.02, 1.11) and colorectal cancer (RR = 1.08; 95% CI: 1.00–1.17), and between GL and endometrial cancer (RR = 1.21; 95% CI: 1.07–1.37).

Meta-analyses on GI, GL, and cancer risk have been recently made available on selected cancer sites, but most of them considered only cohort studies, and very recently published data have not been reviewed yet. Among them there are the National Enhanced Cancer Surveillance study 25, with more than 20 000 cases of a number of different cancers, the Australian Cancer Study on esophageal cancer 26, the Australian Ovarian Cancer Study 27, results on endometrial cancer from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) study 28 and from the Shanghai Endometrial Cancer study 29, results on breast and liver cancer from the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort, with about 200 liver cancer cases 30 and more than 11 000 breast cancer cases 31, results on colorectal cancer from the Italian segment of the EPIC cohort 32 and from a US case–control study with more than 1000 cases 33, and results on liver cancer from the Shanghai Men's and Women's Health study 34. Further, to our knowledge, no meta-analysis is currently available for esophageal or liver cancer.

To obtain an up-to-date quantification of the association between GI and GL and the risk of cancer from case–control and cohort studies and investigate between-study heterogeneity, we conducted a comprehensive systematic review and meta-analysis updated to January 2015, including all cancer sites for which adequate data were available.

2 Materials and methods

3 Results

3.2 GI, GL, and cancer risk

The study-specific and summary RRs of hormone-related cancers according to the highest versus the lowest category of GI and GL are shown in Fig. 1 (panels A and B, respectively). The summary RRs were all above unity but nonsignificant, ranging from 1.05 (breast) to 1.13 (endometrium) for GI, and from 1.04 (prostate) to 1.19 (ovary) for GL.

Figure 2 showed results for digestive-tract cancers. GI significantly increased colorectal cancer risk (summary RR = 1.16, 95% CI: 1.07–1.25). All other summary RRs were above unity, but not significant; they ranged from 1.10 (pancreas) to 1.46 (ESCC) for GI, and from 1.01 (pancreas) to 1.25 (ESCC) for GL.

4 Discussion

This comprehensive and updated meta-analysis of observational studies suggests that high-GI and high-GL diets are related to modest increased risks of cancer for most sites. The estimated summary RRs were all above unity and ranged from 1.05 (breast) to 1.46 (ESCC) for dietary GI, and from 1.01 (pancreas) to 1.25 (ESCC) for dietary GL. These associations, however, were not significant, with the exception of that between dietary GI and colorectal cancer.

Inference is therefore difficult due to nonsignificant results. Still, some effects might be of some relevance at the population level because of the high incidence of several of these neoplasms, in particular breast and colorectal cancers 65. If real, the associations observed can be related to the unfavorable impact of GI/GL on blood glucose levels, insulin, and IGFs, which may promote tumor growth 66.

Significant associations of high GL intake were found with either endometrial or pancreatic cancer risks only among overweight individuals, suggesting that BMI represents an effect modificator of the associations of GL with these two forms of cancer. Overweight strongly influences insulin resistance and hyperinsulinemia 67. It is thus likely that overweight individuals have a greater insulin response to their diet for their physiological condition compared to lean individuals. Similar effects have been reported for GL, which appears to increase the risk of coronary heart disease only in overweight women 5.

High heterogeneity across studies was detected in most of the meta-analyses by cancer site. Therefore, even if we used random effects models to take heterogeneity into account, our summary estimates should be interpreted with caution. We tried to overcome the problem of heterogeneity by pooling results from more homogeneous subsets of studies (i.e. subgroup analyses). In analyses by study design, we found generally higher summary RR estimates in case–control than cohort studies. Still, a number of cohorts supported a detrimental role of high-GI/GL diets in cancer risk. In particular, positive associations for dietary GI were found in men of the National Institute of Health-AARP (NIH-AARP) cohort for stomach and liver cancer 40, in the ORDET study for breast cancer 68, and in the Shanghai Women's Health study (SWHS) for liver cancer 34. For GL, positive associations were observed in the Women's Health Study (WHS) for colorectal cancer 69, in the Hormones and Diet in the Etiology of Breast Tumors Study (ORDET) for breast cancer 68, in the EPIC cohort for breast cancer 31, in the National Breast Screening Study (NBSS) for endometrial 58 and ovarian cancers 70, and in the Nurses’ Health Study (NHS) for endometrial cancer 47. A possible explanation of the higher RR estimates in case–control than in cohort studies is that the former are more prone to problems of recall and/or selection bias. Still, most of the case–control studies included in our meta-analyses had validated FFQs administered by trained interviewers, and this offers additional control over the quality of the measurement compared to self-administered FFQ used in cohorts. Further, case–control studies could include subjects with a diet characterized by higher GI/GL than cohort studies 14. This observation derives from the fact that analyses in case–control studies were based on greater upper quantiles of GI/GL, in the absence of marked differences in the lowest quantiles. If associations of small magnitude exist, these are difficult to be detected in cohort studies on populations with more homogenous diets and a narrower range of total dietary GI/GL. Further, the small differences between case–control and cohort studies may reflect possible differences in dietary habits between European and North American populations. Indeed, 12 out of 20 cohorts were conducted in the United States, and only six in Europe; in contrast, 17 out of 32 case–control studies were conducted among Europeans and only six among North Americans. European populations tended to consume high amount of bread, refined carbohydrates, and pasta 71-73, whereas the diet of North American populations is rich in high-fat foods and therefore tends to have low GL values.

Adjustment for potential confounding factors was different across studies, and this may have resulted in residual confounding. However, most studies allowed for BMI/physical activity, energy intake, tobacco, and some indicators of social class, as showed in Supporting Information Table 1. Thus, residual confounding should be limited.

Furthermore, studies used different methods of segregations according to GI/GL (e.g. tertiles, quartiles, quintiles), and there was heterogeneity of thresholds of various levels, including the highest one used in the current meta-analysis. This may reflect different dietary habits among populations, as well as the different choice of the reference food (glucose or white bread) for GI/GL calculation. Indeed, when white bread was the reference food, the GI value against glucose was calculated simply by multiplying by 0.7 2. Since we considered quantiles of the distribution and not absolute values, the choice of white bread or glucose as reference did not materially influence the results. When conducting stratified analyses by reference food, we observed no material differences in the summary RR estimates for white bread and glucose (data not shown).

Some stratified analyses considered a limited number of studies, and their results must therefore be interpreted with caution. Further, some positive findings in stratified analyses may be attributed to the role of chance following multiple testing.

Criticisms have arisen concerning the use of GI in epidemiological studies. Indeed, GI is an indicator of quality of carbohydrates but not of the cumulative GL of the diet, and does not vary in response to the amount of food consumed 74.

The main advantages of the present meta-analysis were the inclusion of a large number of studies and cases and the consequent statistical power to detect moderate associations, and the use of a systematic meta-analytic approach to summarize our results and investigate reasons for between-study heterogeneity. Finally, results from Begg's and Egger's tests indicate that publication bias is unlikely to have appreciably influenced our results.

In conclusion, our results, derived from the most up-to-date evidence, are suggestive of modest increased risks for most cancer sites for high-GI and high-GL diets.

The authors have declared no conflict of interest.