Glycemic Load, Index, and Diabetes Risk

Introduction

Type 2 diabetes has been one of the foremost causes of cardiovascular disease. It has about 10 percent global prevalence (van Bakel et al., 2009). The diet of an individual has been under consideration for the development of type 2 diabetes. In this context, specifically the foods’ capacity to contain carbohydrates must have to be augmenting the blood glucose (Mann, 2007). It is suggestive that diets having high GL (glycemic load) or GI (glycemic index) may have the predisposition of higher postprandial insulin and blood glucose concentrations that in turn have been increasing glucose intolerance and eventual risk of type 2 diabetes (Venn and Green, 2007). There have been several studies indicating a relationship between GL, GI and type 2 diabetes. However, there have been several other big studies that have findings that there is no evidence supporting the hypothesis (Jenkins et al., 1988). As per the dietary guidelines of American Diabetes Association related to the prevention of diabetes that currently states that the evidence is saying that diabetes risk is reduced with diets low in GL is insufficiently consistent. Inconsistencies of considerable amount are also found in results with regards to the role of the total intake of carbohydrate. There has been conclusion related to the systematic reviews with regards to the evidence of positive relation between both dietary GL and GI and type 2 diabetes risk (Moghaddam et al., 2006), although with significant amount of unexplored heterogeneity. By comparing the most extreme categories only on the basis of varied definitions in each study reviewed, additional heterogeneity is introduced and information discarded in the categories of middle exposure that leads to uncertainty with regards to the association’s strength. Combining various definitions of the lowest and the highest categories of exposure indicates that their estimates of summary cannot be in relation to a specific exposure level that limits the applicability of results in terms of public health. Moreover, the assessment was not done with the review that any dose response relationship’s nature has been a key criterion to judge the chances of any casual association. The scrutiny that surrounds the carbohydrate foods quality in the last decade have began approaching the fats. Big epidemiological studies have established linkages between the chronic diseases with the glycemic response with the association of GL or GI diets with several positive health impacts. These are inclusive of: (a) in diabetic subjects, improvement of glycemic control having association with diabetes’ reduced risk; (b) lipid profiles that are more favorable having association with cardiovascular disease’s lower risk; and (c) inflammation’s reduced markers having association with overweight, metabolic syndrome’s lower risk and other chronic diseases. There have been some studies that are associated between GL/GI and the risk of breast, colon, and other cancers. In spite of studies indicating associations, several studies are there that fail in linking GL/GI of the diet with disease and health risk.

There have been two main possibilities accounting for the lacking of consistencies amongst studies. The first one has to deal with the potential confounding because of the quality of diet. The second one has been in relation to the variability having association with the method to determine GL and GI in the studies. Low GL/GI diets, in terms of confounding, can have the linkages to the reduction of the disease risk as they are consistimg of whole grains, legumes, nuts, vegetables, and fruits. These nutrients of high fiber, phytochemicalrich, and intense foods have capacity in lowering the risk of the disease irrespective of their GL/GI (Buscemi, et al., 2013). The diets and foods, on the other hand, may have low GL/GI in case they are contained with low GI sweeteners like high in fat and meat or fructose. These diets that are widely variant have the likelihood of having different effects. In the same way, the foods having high GL/GI can have a positive association with highly processed snacks’ daily consumption and negative association with nutrient-rich foods’ consumption. Therefore, a high GL/GI dietary may be reflective of food patterns having less likelihood of delivering disease protecting nutrients. Therefore, the GL/GI of certain diets can be about whether it is present or what is it with CHO rather than CHO’s glycemic effect. The GI’s determination of a food, even in the conditions of closely controlled laboratory by experienced analysts, where large standard deviations are created even for a single individual. This makes the measure’s accuracy and precision been questioned by many (Elmazar et al., 2014). In a similar way, dietary GL or GI is determined from the frequency of food also having several potential error source. This has been subject to compounding by the changes in food amount or food combination been changed in the measurement of glycemic response. The contention of the proponent that, in spite the variability, the use of GL/GI concept has importance to make the selection of diet and that GL/GI should be under control for both prevention and treatment of chronic disease. The contention of the detractors has been that the variability of the measure makes the usage of GL/GI valuable in the setting of a laboratory only.

Therefore, the carbohydrate foods are placed in the front and center of the nutritional controversy. There are not viewed as merely as a source of energy or as a stable energy anymore, instead they find themselves to be the addition in the list of components that must be subject to evaluation when deciding the quantity and the type to have inclusion in the diet.

Methods

There has been conduction of systematic literature search that covered all prospective research with the provision of evidence on all aspects of cardiometabolic health and dietary carbohydrates that includes obesity, glycemic response, insulin resistance, and cardiovascular disease. There has been searching of the online databases for the published prospective studies in English language in the period of 2009 to 2018. The searches include the ISI Web of Science, CAB Abstracts, Embase, MEDLINE in-process, MEDLINE, Cochrane Library, and BIOSIS. The search was then updated with the use of primary sources (Embase and MEDLINE that includes MEDLINE in-process). The updated search will only include the cohort studies that investigate the GL, GI, and type 2 diabetes and carbohydrate intake. The eligibility was limited to cohort studies only including the case cohort studies and nested case control studies within a cohort. The criteria of inclusion were studies on the basis of adult population since 2009 and published in English. These studies have assessment of GL, GI, and intake of dietary carbohydrate with over two exposure categories.

The extraction of the information has been the following ones from the identified publications: numbers of non-cases and cases, length of follow up, type of study (case cohort, nested case control, full cohort), participants’ mean age, range of age, sex of the participants, name of the study, the study’s geographical region, year of publication, authors, dietary assessment method, outcome assessment method, dietary exposure level (either as midpoint, median, mean, or range of each category or the increment unit for continuous estimates), RRs’ estimates with CIs, the standards usable in deriving GL or GI, and the controlled characteristics for either stratification, matching or modeling. The extraction of data has been carried out by V.J.B., C.W., C.N., C.L.C., C.E.L.E., D.E.T., and D.C.G. and the accuracy of it has been checked by D.C.G. and D.E.T. The full text copies were extracted of the articles that are potentially relevant. These are read by the two members of the team of review independently. The settlement of any disagreement would be a third reviewer. A detailed guideline and a structured flowchart have been used in determining the inclusion eligibility.

Results

The eligibility includes only the cohort studies that comprise case cohort studies and case control studies nested within a cohort. The criteria of inclusion have been studies on the basis of an adult population since 1990 published in English language with the assessing of the GL, GI or the intake of total dietary carbohydrate with over two exposure categories. These have some control at least to confront either by adjusting in a matching or model, and as an outcome with type 2 diabetes and some RR (relative risk) estimate with the measurement of uncertainty such as CIs of 95 percent. The studies that are included are only with generally healthy participants i.e. if there is recruitment of only cohort participants especially due to disease’s personal history or ill health. The cases’ mean dietary exposure in comparison with non-cases does not have eligibility if not they are adjusted mean. There is no eligibility of the dietary patterns if the intake is not quantified by them. There is also no eligibility of gestational diabetes outcomes.

Synthesis and analysis of data

In enabling the pooling of study results of individual with the use of various exposure categorizations, there was derivation of linear dose-response trend for each study with the use of the method of Greenland and Longnecker (Greenland and Longnecker, 1992). The estimation of this method has been with dose-response slopes that are study specific and the related CIs on the basis of presented results for each category of GL, GI or the intake of total dietary carbohydrate prior to being combined into pooled estimate.

For the derivation of the dose-response trend, median or mean exposure had been used for each category in case it has been used and presented the midpoint when there was the presentation of the exposure ranges instead. At the time when there was unbinding of the highest and lowest categories, it has been assumed that the category width to be identical as the category adjacent when the midpoint is estimated. Greenland and Longnecker’s method has the requirement too with regards to the distribution of person years and cases, or non-cases and cases, with the estimates of uncertainty (for example, CI) and RRs for the three categories at least of the quantified GL, GI, or intake carbohydrate. Where, the person-years or the total number of cases in the publication was presented, but not in the case of distribution, it was estimated that this has been on the basis of quantiles’ definitions. The exposure level estimation on the basis of midpoint, mean, or median, has been assigned to the RR corresponding to it for each study. For the presentation of the studies with respect to the exposure per given energy intake unit, this was rescaled with the use of intake of estimated energy if this was presented for each category. The studies that have reported already a linear dose–response trend with a precision measure such as standard error or a CI have used this directly. Where there is only separate presentation of the results, for women and men, these have been in combination first with the use of fixed effects meta-analysis prior to the combination of other studies. This has been ensuring that there is no underestimation of heterogeneity between the studies. For each study, all dose-response trends estimated have been pooled with the use of random effects model in taking into account the anticipation of heterogeneity between studies (Zheng et al., 2012). The presentation of linear dose–response trend involves the choosing of an increment size around a single standard deviation’s equivalent in U.S. or European population in easing across exposure comparison.

In examining the possible nonlinear associations, the restricted cubic splines are calculated for each study with the exposure of over three categories with the use of fixed knots at 10, 50, and 90 percent. This has been with the reported intake’s total distribution and then the combination of usage of multivariate meta-analysis. There have been 4 studies presenting results over a continuous exposure for a linear trend (Franz et al., 2010). Therefore, there should not be inclusion of these in the nonlinear dose–response analyses.

Findings

Twenty four publications have been identified from 21 cohort studies reporting GL, GI, and total carbohydrate intake along with type 2 diabetes incidence (Appendix 1) The Meta analysis could not be used in one publication as intake was not quantified by it (Villegas et al., 2007). This one is not usable as the results presented by it have been for the lowest and highest categories. The other reasons for not using it has been due to the form of results that have been presented (Sonestedt et al., 2012). The 18 cohorts remaining have the provision of sufficient information to include the dose– response meta-analyses (Appendix 2). The bias assessment’s risk has the provision in the Appendix 3. From United States, there have been nine studies; from Europe, there are four studies and the remainder from China, Japan, and Australia. The results presented by one cohort study in 3 publications (Halton et al., 2008). Thus, data has been used in the publications that are most recent (Mekary et al., 2011). A study further reported the load and GI from the intake of total carbohydrate in a separate paper. For the inclusion of one study, the estimated standard errors have been used with the P value reported along with the estimates (Schulz et al., 2006). For the inclusion of another, there is estimation of the category means on the basis of the assumption of normal distribution with derivation of approximate standard deviation and mean from the publication (Hopping et al., 2010). The studies that are excluded to report the unadjusted estimates had had the resulting losses of two studies that present results for the intake of carbohydrate which would be included otherwise (Montonen et al., 2007).

GI

The extraction of data has been 15 publications that investigated the relationship between the type 2 diabetes and the GI. The estimated mean of the category of the intakes have the range from around 45 to 90 GI units with individual studies spanning between 6-36 units. The estimate that is pooled of RR from the Meta analysis of linear dose-response has been 1.08 (1.02–1.15 CI, 95%) per 5 GI units (P = 0.01). There has been considerable heterogeneity among the cohort studies (95% CI 80–92%; I2 = 87%; P < 0.001; df = 14). Studies that adjust for type 2 diabetes family history seemingly have much higher estimates compared to the ones that do not adjust (P < 0.001). The stronger relationship between the diabetes and GI has the restriction to the studies that have had the adjustment for this. This leads to the improvement in heterogeneity within each subgroup (Appendix 4) estimates have consistency across other subgroups that are predefined. The funnel plot has approximate symmetry not having much evidence of the effects of small study such as publication bias. The dose–response meta-analysis of nonlinear nature has shown consistent increased risk having association with increase in GI. There has not been much evidence in the plot with a threshold effect.

GL

The extraction of data has been 16 publications that investigated the relationship between the type 2 diabetes and the GL (5,6,8) (Fig 1 B). The estimated mean of the category of the intakes have the range from around 55 to 245 GL units with individual studies spanning between 48-190 units. The estimate that is pooled of RR from the Meta analysis of linear dose-response has been 1.03 (1.00–1.05 CI, 95%) per 20 GL units (P = 0.02). There has been moderate heterogeneity among the cohort studies (95% CI 19–74%; I2 = 54%; P = 0.005; df = 15). Much like GI, Studies that adjust for type 2 diabetes family history seemingly have much higher estimates compared to the ones that do not adjust for this covariate (P = 0.03). The stronger relationship between the diabetes and GL has been apparent to the studies that have had the adjustment for this. The family stratifying it leads to the improvement in heterogeneity within each subgroup (Appendix 4). Longer follow up has association with stronger association between type 2 diabetes and GL (P = 0.03). The estimates have consistency across other subgroups that are predefined. The funnel plot has approximate symmetry not having much evidence of the effects of small study such as publication bias.

The dose–response meta-analysis of nonlinear nature has shown consistent increased risk having association with increase in GL. There has not been much evidence in the plot with a threshold effect.

Total carbohydrate

The extraction of data has been 8 studies that investigated the relationship between the type 2 diabetes and the carbohydrate (5,6,8) (Fig 14). The estimated mean of the category of the intakes have the range from around 130 to 340 g with individual studies spanning between 72-210 g. The estimate that is pooled of RR from the Meta analysis of linear dose-response has been 0.97 (0.90–1.06 CI, 95%) per 50 g per day of total intake of carbohydrate (P = 0.5). There has been considerable heterogeneity among the cohort studies (95% CI 50–88%; I2 = 75%; P < 0.001; df = 7). There are estimates which have high consistency across the subgroups that are predefined, although the studies have a tendency with longer follow up in having larger estimates (Appendix 4). The funnel plot has approximate symmetry not having much evidence of the effects of small study such as publication bias. The dose–response meta-analysis of nonlinear nature has shown a relatively flat curve over intakes that are typical and having a broad range.

Data Interpretation, Discussion, and Conclusion

It has been quantified that there is a clear positive relationship between GL and GI with type 2 diabetes’ increasing incidence. There is stronger association for GI compared to GL with around a single GI intake standard deviation having association with over two times the increased risk having association with GL. Comparing the dietary GI data; the GL’s evidence base has more inconsistency in terms of the association’s direction.

In spite of the usage of linear dose–response trends in combining the studies that uses the categorizations of various exposures, for all exposures the heterogeneity is still high. This heterogeneity’s exploration with the investigations of the estimates in various predefined subgroups is suggestive of the family history’s adjustment of diabetes that has been the key potentially, with studies failing in adjusting for it to have much lower estimates between the type 2 diabetes and GL, and GI. These findings have the consistency with the ones of the previous two systematic reviews (Dong et al., 2011). This study has quantified the association’s strength and has explored in the results related to some of the heterogeneity. This paper has also removed some portions of this heterogeneity with the combined dose-response trends. This study has also investigated the possibility of nonlinear associations. The results have been included from the large prospective studies and nine publications that since the review are most recent that have been published. These are inclusive of roughly 20,000 more type 2 diabetes cases from more than 250,000 more participants which strengthens further the evidence on which the conclusion of this study is based. The observational studies’ meta-analysis has susceptibility to the identical biases that are contained in the studies and are prone to. The pooled estimate still can be containing the element of bias so that the reviewed studies are also biased. All reviewed studies, in particular, have the usage of some kinds of self-reported dietary exposure and thus have the susceptibility to the potential and big measurement error. Additionally, there are several adjustments for dietary covariates that are self reported and therefore are devoid of having full adjustment for true intake. This can lead to bias in either direction with regards to the association. Moreover, it could not be conclusively proved that any association has casualness based alone on observational studies. There is even greater number of uncorrected confounding in all or some of the studies. However, the found estimates for the GL and GI are strong and the dose–response trends are clear, and there is lack of evidence with regards to the effects of small study such as bias in publication.

Given the databases’ limited nature of the foods’ GI values, to assign to the diet of an individual with GI as captured by the FFQ (food frequency questionnaire) has the potential to be problematic. The GI values for each item of the food, typically, have been taken in a questionnaire from the foods’ GI values’ 2002 international table (Evert et al., 2013). Broad foods’ groupings within each item of FFQ at times necessitate the average GI’s allocation for that item. This has led to some concerns of expressions related to the use of FFQ-derived GL and GI values in exploring the association of the disease (Franz et al., 2014). The foods’ dietary GI has been varying considerably and depends on the extent of cooking and processing duration and method, extent of the ripeness of starch gelatinization, and the duration of the storage (Venn and Green, 2007). There are more issues concerning whether the consumption of the food impact one another in altering the whole meal’s GI (Cozma et al., 2012). Therefore, this exposure has the potential of being prone to the error bias of the measurement. The GL’s estimation has the requirement of the carbohydrate amount’s additional estimate in the diet that provides the dilution’s greater scope of results through the error bias measurement. Although the estimation of the GL and GI’s absolute values, in all probability, do not form the actual values’ accurate estimates in a number of studies, these can still be used in order to combine different studies on dose-response trends and on the same scale and nonlinear trends for estimation. However, to interpret these, the focus must be on the relative ranking in the same magnitude as on the estimated GL and GI. It has been reported that there is a plethora of exposures across publications. However, the individual studies that have reported the intakes usually have been varying by small amounts. This can be reflective of various dietary assessment tools that lead to various measurement error amounts in each study or can be because of contrast in populations, phenotypes, and various diets.

Generally speaking, the diet’s GL has the likelihood to be related partly to the content of dietary fiber. This essentially means that there is difficulty in dissociating the GL effects from the fiber content. In the studies that are included in this paper, the fiber adjustment has the tendency to have association with estimates that are larger where this has been done (Gupta et al., 2011). This suggests that other studies may be subject to underestimation with regards to the association, and there may be underestimation of the pooled estimate. In a similar way, GL and GI may have the reflection of dietary quality’s other aspects, such as intake of saturated fat, with the findings have the part reflection of characteristics other dietary aspects. There is a high likelihood that intakes of other carbohydrate may be the substitute of protein or fat, while the intake of constant energy is maintained. This has been another example where there is inability for the observational studies in assigning casualty and this also goes with their meta-analysis. The results are laden with inconsistencies for the type 2 diabetes and the total carbohydrate that can be because of the main sources’ differences and the types of the consumption of the carbohydrate’s other differences in the practices of dietary between the cohorts of Australia, China, US, and Europe. This may also be reflective of the healthier possibility with consumption of carbohydrates of more active people. It can be alternatively explained in relation to the differences between the carbohydrate type eaten and the carbohydrate amount consumed, with different cohorts that have different proportions of women and men. This may be accounting for any dose-response plot’s nonlinear appearance with studies that reports to the total carbohydrates’ higher intakes that have various carbohydrate sources in the diet compared to the ones that reports lower intakes. The dose-response curves of nonlinear nature have susceptibility to cohorts with different intake ranges that leads to the nonlinear curve’s appearance. In a situation like this, differences in population or design can be the causation of the nonlinearity appearance. However, there have been spread of reasonable degree of carbohydrate intake over several studies inclusive of the meta-analysis.

The specific type of sugar can also be impacting the GI. Glucose is contained with high GI (Appendix 5) and there is low glucose in fructose (Appendix 6). The high fructose corn syrup and sucrose have around 50 percent fructose and 50 percent glucose. This resulted in the GIs reflecting these two monosaccharides’ mix and has been moderate to the GI foods. The foods’ macronutrients composition has led the GI to be affected. In the food, protein or fat can change the absorption and starch digestion rate and lowering the GI (Kaufman, 2012). Appendix 7 has provided the selected food their fat and GI content. The higher fiber food, the soluble fiber in particular, has often been having low GI. Therefore, most studies have the inability in determining whether the effect has been because of low GL/GI or the fiber in the diet. A study conducted recently had the design of addressing this issue. While this well controlled study’s authors have been suggestive of the low GI having greater importance compared to the dietary food, the study failed in equalizing the intake of fiber between the low and high GI groups (Sisson and Cornell, 2011). The findings of this study have the consistency with, and contribution to, a growing body of evidence to protect the associations with low dietary GL and GI. The results have the quantification for the range of exposure for the first time associated with lowered risk and the risk reduction is quantified that has the association with the differences specified in GL and GI. The carbohydrate related results, more generically, have less clear, and the focus of the future research could be in greater details on the composition and source of carbohydrate having association with the greatest risk.

References

Dong, J. Y., Zhang, L., Zhang, Y. H. and Qin, L. Q. (2011) ‘Dietary glycaemic index and glycaemic load in relation to the risk of type 2 diabetes: a meta-analysis of prospective cohort studies’, Br J Nutr, 106:1649–1654.

Flint, A., Møller, B. K., Raben, A., et al. (2004) ‘The use of glycaemic index tables to predict glycaemic index of composite breakfast meals’, Br J Nutr, 91:979–989.

Greenland, S. and Longnecker, M. P. (1992) ‘Methods for trend estimation from summarized dose-response data, with applications to meta-analysis’, Am J Epidemiol, 135: 1301–1309.

Hopping, B. N., Erber, E., Grandinetti, A., Verheus, M., Kolonel, L. N. and Maskarinec, G. (2010) ‘Dietary fiber, magnesium, and glycemic load alter risk of type 2 diabetes in a multiethnic cohort in Hawaii’, J Nutr, 140: 68–74.

Jenkins, D. J., Kendall, C. W., McKeown-Eyssen, G., et al. (2008) ‘Effect of a low-glycemic index or a high-cereal fiber diet on type 2 diabetes: a randomized trial’, JAMA, 300:2742-53.

Krishnan, S., Rosenberg, L., Singer, M., et al. (2007) ‘Glycemic index, glycemic load, and cereal fiber intake and risk of type 2 diabetes in US black women’, Arch Intern Med, 167:2304–2309.

Mekary, R. A., Rimm, E. B., Giovannucci, E., et al. (2011) ‘Joint association of glycemic load and alcohol intake with type 2 diabetes incidence in women’, Am J Clin Nutr, 94:1525–1532.

Montonen, J., Järvinen, R., Knekt, P., Heliövaara, M, and Reunanen, A. (2007) ‘Consumption of sweetened beverages and intakes of fructose and glucose predict type 2 diabetes occurrence’, J Nutr, 137:1447–1454.

Schulz, M., Liese, A. D., Fang, F., Gilliard, T. S. and Karter, A. J. (2006) ‘Is the association between dietary glycemic index and type 2 diabetes modified by waist circumference?’ Diabetes Care, 29:1102–1104.

Sonestedt, E., Lyssenko, V., Ericson, U., et al. (2012) ‘Genetic variation in the glucose-dependent insulinotropic polypeptide receptor modifies the association between carbohydrate and fat intake and risk of type 2 diabetes in the Malmo Diet and Cancer cohort’, J Clin Endocrinol Metab, 97:E810–E818.

van Bakel, M. M., Slimani, N., Feskens, E. J, et al. (2009) ‘Methodological challenges in the application of the glycemic index in epidemiological studies using data from the European Prospective Investigation into Cancer and Nutrition’, J Nutr., 139:568-75.

Villegas, R., Liu, S. Gao, Y. T., et al. (2007) ‘Prospective study of dietary carbohydrates, glycemic index, glycemic load, and incidence of type 2 diabetes mellitus in middle-aged Chinese women’, Arch Intern Med, 167:2310–2316.

Buscemi, S., Nicolucci, A., Mattina, A., et al. (2013) ‘Association of dietary patterns with insulin resistance and clinically silent carotid atherosclerosis in apparently healthy people’, Eur J Clin Nutr., 67:1284–90.

Elmazar, H. M., Essa, A. B., Ojurongbe, O., Oyesiji, K., et al. (2014) ‘Environmental and dietary factors affecting the progression of Type 2 diabetic retinopathy in Aljabal Algharby, Libya’, Int Res J Med Med Sci., 2:1–5.

Zheng, Y., He, M. and Congdon, N. (2012) ‘The worldwide epidemic of diabetic retinopathy’, Indian J Ophthalmol., 60:428–31.

Evert, A.B. Boucher, J.L. et al. (2013) ‘Nutrition Therapy Recommendations For the Management of Adults with Diabetes’, Position Statement by the ADA, Diabetes Care, 363821-42.

Franz, J. et al. (2014) ‘Evidence-based diabetes nutrition therapy recommendations are effective: the key is individualization’, Diabetes, Metabolic Syndrome and Obesity: Targets and Therapy, 7 65–72.

Cozma, A. I., Sievenpiper, J. L., de Souza, R. J., et al. (2012) ‘Effect of fructose on glycemic control in diabetes: a systematic review and meta-analysis of controlled feeding trials’, Diabetes Care, 35:1611–1620.

Kaufman, F. R., ed. (2012) Medical Management of Type 1 Diabetes [6th ed.], Alexandria, VA: American Diabetes Association.

Sisson, E. M. and Cornell, S. (2011) Pharmacotherapy for glucose management. In: Mensing C, ed. The Art and Science of Diabetes Self-Management Education Desk Reference [2nd ed.], Chicago, IL: American Association of Diabetes Educators, pp. 417-458.