This is the first systematic review of the evidence to date on GI and GL intake and endometrial cancer risk in relation to BMI and provides evidence of a positive association between high GL diets and endometrial cancer risk but little evidence of an association for dietary GI intake. The finding that a high GL diet increases the risk of endometrial cancer compared with a low GL diet demonstrates the advantage of a meta-analysis because only one of the individual studies reported statistically significant results, although all studies showed associations in a positive direction. It should be noted that some individual studies had a relatively narrow range of GI and GL intakes, which could explain the lack of statistically significant results observed in these studies. In the cohort studies, comparing the highest versus the lowest category of GL intake corresponded to an approximate difference of 50 GL units (Folsom et al, 2003
; Silvera et al, 2005
; Cust et al, 2007a
; Larsson et al, 2007
), whereas in the case–control studies this observed difference was approximately 100 units (Augustin et al, 2003a
). This could partly explain the stronger association observed in the meta-analysis of endometrial cancer risk and GL when the case–control study was included.
We did not observe an association between dietary GI and endometrial cancer risk. This would suggest that endometrial cancer risk is related to the actual blood glucose, and hence insulin, demand induced by the consumption of normal portion sizes of carbohydrates rather than the standard 50
g used to calculate GI values. Despite this, GI is still an important contributor to the GL value of a food, and it would be preferable to advise individuals to consume a diet composed of low GI/moderate total carbohydrate content as opposed to high GI/low total carbohydrate content to achieve a low dietary GL intake. There is a paucity of research examining the association between dietary factors and endometrial or ovarian cancer risk. Investigations of total carbohydrate or cereals have provided weak or null associations (Jain et al, 2000
; Trichopoulou et al, 2000
). Identifying a strong association for any modifiable dietary factor and endometrial or ovarian cancer is therefore extremely valuable. In addition, there is little evidence to suggest that a low GI diet may have any accompanying adverse effects, and thus advising women to consume a low GL diet would seem a reasonable approach (Colombani, 2004
Previous reports have shown a linear dose–response relationship between BMI and endometrial cancer risk (Jain et al, 2000
; Furberg and Thune, 2003
; Schouten et al, 2004
; Friedenreich et al, 2007
). Our findings that a high GL diet increases the risk of endometrial cancer as BMI increases, suggests that BMI may be an effect modifier of the association between GL and endometrial cancer, and that high GL diets may exaggerate endometrial cancer risk in women who are more likely to be insulin resistant. Additionally, a recent meta-analysis of 16 studies demonstrated that diabetics have over twice the risk of developing endometrial cancer compared with that of non-diabetics (Friberg et al, 2007
). Similar elevated risks of ovarian cancer incidence and mortality were observed in a large UK cohort of diabetics compared with the general population (Swerdlow et al, 2005
). Consuming low GI carbohydrates has been associated with improved glycaemic control in diabetic patients compared with high GI diets in randomised controlled trials (Brand-Miller et al, 2003
). Other subgroup analyses in the papers included in our review have suggested that the association between GL and endometrial cancer may be modified by diabetes, menopausal status, HRT use or physical activity (Augustin et al, 2003a
; Folsom et al, 2003
; Silvera et al, 2005
; Cust et al, 2007a
; Larsson et al, 2007
). Unfortunately, we were unable to perform meta-analyses on the basis of HRT use, diabetes or physical activity as too few studies reported results by these stratifications, and therefore no conclusions can be drawn with respect to these variables.
Although the mechanisms are recognised to differ by cancer site, the IGF system is often reported as the proposed mediator between GI, and therefore potentially GL and cancer risk (Du et al, 2006
). A validation study of the effect of GI on the insulin response did report an overall 70% reduced insulin response after consumption of a low GI food compared with a high GI food (Brand-Miller et al, 2005
). However, alterations to IGF-1 and IGFBP-3 levels were minimal following the low GI food compared with the high GI food. Notably, this study was conducted in lean young subjects, so the application of these findings to obese people is currently unknown. Obese subjects are known to have elevated circulating IGF-1 levels as a result of overnutrition (Augustin et al, 2002
), and it is therefore plausible that a high GL diet in people with a higher BMI has a more profound effect on IGF-1 levels. Unfortunately, studies that have investigated the association between the IGF system and endometrial and ovarian cancer risk have produced inconsistent findings (Lukanova et al, 2004
; Peeters et al, 2007
; Tworoger et al, 2007
). Despite this, C-peptide, a marker of pancreatic insulin production, has repeatedly been shown to be directly related to endometrial cancer risk in well-designed studies (Lukanova et al, 2004
; Cust et al, 2007b
), suggesting a key role for hyperinsulinaemia. The majority of endometrial cancers are oestrogen-related (WCRF/AICR, 2007
), therefore hyperinsulinaemia induced by a habitually high GL diet may explain the increased endometrial, and possibly ovarian, cancer risk by the ‘unopposed oestrogen' hypothesis.
Oestrogen is a known mitogen, and overweight and obesity could increase endometrial and ovarian cancer risk due to ovarian hyperandrogenism (surplus ovarian androgen production), which is promoted by hyperinsulinaemia, resulting in a subsequent oestrogen excess derived from the aromatisation of androgens in adipose tissue (Kaaks et al, 2002
). In postmenopausal women, the cessation of progesterone synthesis results in oestrogen concentrations being insufficiently counterbalanced, while in hyperinsulinaemic premenopausal women, absent ovulation and ensuing progesterone deficiency may also enhance the mitogenic potential of oestrogen (Kaaks and Lukanova, 2001
; Kaaks et al, 2002
). Others have recognised that insulin levels per se
cannot explain the disparity in the observed risk between obesity and premenopausal and postmenopausal breast cancer risk, whereas oestrogen concentrations are well correlated (Key, 2001
The precise mechanisms through which a high GL diet increases endometrial, and potentially ovarian, cancer risk needs to be clarified in further research that utilises independent biomarkers such as C-peptide or components of the IGF system, in addition to dietary exposure. The inclusion of biomarkers could help to overcome the limitations associated with dietary GI and GL evaluation. For example, GI and GL values apply only to single foods, so when composite meals are consumed the ability to predict insulinaemic responses from GI values has been questioned (Flint et al, 2004
). Blood glucose and insulin concentrations can also be influenced by other dietary components such as protein and fats (Jenkins et al, 1984
). Furthermore, the studies in our systematic review used a mixture of glucose and white bread reference values for GI and GL. The reproducibility of GI and GL values and their application to different population groups are often highlighted as important methodological issues (Feskens and Du, 2006
), and to our knowledge, the validity of combining GI and GL results from different populations is currently unknown.
Other limitations to individual study designs reported in this review include the possibility of recall bias in case–control studies, whereby cases recall their diet differently compared with healthy controls (Augustin et al, 2003a
). Self-administered FFQs were used in all of the prospective cohorts, none of which were specifically designed for assessing GI or GL intake. Food Frequency Questionnaires are known to incorporate some dietary measurement error, especially among overweight or obese individuals but are the most convenient assessment tool available for large-scale studies (Black et al, 1993
). The FFQ employed in one of the case–control studies was also relatively short, including only 37 items (Augustin et al, 2003a
). Only two cohorts incorporated repeat dietary measures (Cust et al, 2007a
; Larsson et al, 2007
), one of which used a single 24-h recall in a stratified random sample of participants in addition to an FFQ (Cust et al, 2007a
), whereas the other obtained dietary information using an FFQ at two different time points (Larsson et al, 2007
). Despite potential dietary measurement error, it is unlikely that the women in these studies were aware of any potential link between dietary GI or GL and cancer at the time of participation, which ranged between the 1980s and 1990s.
Our meta-analysis of GI and GL intake and endometrial cancer risk does have limitations. The overall results are based on only five studies, and only four studies are included in the analysis stratified by BMI. However, all studies incorporated a large number of cases, particularly the EPIC study, which is a multicentre investigation of the association between diet and cancer risk in 10 European countries. Potential confounders such as parity, hormonal use, age at menarche and age at menopause were not universally adjusted for, which is not ideal when combining results; however, age, energy intake and BMI were adjusted for in all studies. We only compared low GI (GL) intake with high GI (GL) intake in reported categorisations, which differed from study to study. It is also difficult to determine with certainty any effect of publication bias or heterogeneity in such a small sample of studies, as indicated by the wide confidence intervals shown in the I2 test for heterogeneity. In addition, the possibility of residual confounding cannot be ruled out.
Future well-designed studies or consortium-based analyses with a large number of cases, and consequently, more power should be conducted, particularly for the examination of interactions in high-risk insulin-resistant population groups, that is overweight, obese and sedentary women, to confirm our results. Further research is required on the effects of a high GI or GL diet in ovarian cancer, as our systematic review included only two studies, both of which showed results similar to those seen for endometrial cancer risk.
In conclusion, consuming a high GL diet is associated with an increased risk of endometrial cancer and risk is further increased in obese women. Dietary GI does not appear to be related to endometrial cancer risk. Further research is required on GI, GL and ovarian cancer risk.