Dietary Glycemic Index and Load and the Risk of Type 2 Diabetes: A Systematic Review and Updated Meta-Analyses of Prospective Cohort Studies

Abstract

Published meta-analyses indicate significant but inconsistent incident type-2 diabetes(T2D)-dietary glycemic index (GI) and glycemic load (GL) risk ratios or risk relations (RR). It is nowover a decade ago that a published meta-analysis used a predefined standard to identify validstudies. Considering valid studies only, and using random effects dose-response meta-analysis(DRM) while withdrawing spurious results (p < 0.05), we ascertained whether these relationswould support nutrition guidance, specifically for an RR > 1.20 with a lower 95% confidence limit>1.10 across typical intakes (approximately 10th to 90th percentiles of population intakes). Thecombined T2D-GI RR was 1.27 (1.15-1.40) (p < 0.001, n = 10 studies) per 10 units GI, while that forthe T2D-GL RR was 1.26 (1.15-1.37) (p < 0.001, n = 15) per 80 g/d GL in a 2000 kcal (8400 kJ) diet.The corresponding global DRM using restricted cubic splines were 1.87 (1.56-2.25) (p < 0.001, n =10) and 1.89 (1.66-2.16) (p < 0.001, n = 15) from 47.6 to 76.1 units GI and 73 to 257 g/d GL in a 2000kcal diet, respectively. In conclusion, among adults initially in good health, diets higher in GI or GLwere robustly associated with incident T2D. Together with mechanistic and other data, thissupports that consideration should be given to these dietary risk factors in nutrition advice.Concerning the public health relevance at the global level, our evidence indicates that GI and GLare substantial food markers predicting the development of T2D worldwide, for persons ofEuropean ancestry and of East Asian ancestry.

Keywords: alcohol; cohort studies; dietary fiber; epidemiology; glycemic index; glycemic load; meta-analysis; protein; type 2 diabetes.

Figure 1

Process of the review and outcomes of the inclusion/exclusion criteria for the T2D–GI and GL risk relations. The literature was searched for prospective cohort studies investigating these relations, from 1946 to 6 December 2018. MEDLINE, EMBASE, and other sources (see Methods) were searched. ^a–^e: See Section 3 of the Supplementary Materials for further details. Abbreviations: GI, glycemic index; n, number of published reports; T2D, type 2 diabetes; GL, glycemic load.

Figure 2

Process of the review and outcomes of the inclusion/exclusion criteria for the T2D–GI risk relation. The literature was searched for prospective cohort studies investigating this relation, from 1946 to 6 December 2018. MEDLINE, EMBASE, and other sources (see Methods) were searched. Superscripts: a. Whether T2D is related to the dietary glycemic index (GI) and whether the risk relation is >1.20 with a lower confidence limit > 1.10 for each 10th to the 90th percentile of GI intake (which on analysis was per 10 units GI). b. Is the T2D–GI risk relation conditional on the validity of the dietary instrument, on the sex of the participants, or on the BMI (kg/m²)? c. Details of the full reports excluded are described in the associated text and in more detail in Section 3 (f and g) of the Supplementary Materials. d. Studies with dietary instrument correlations for carbohydrate ≤ 0.55 were invalid. e. When a study was suspected of being an outlier was proven to be so upon analysis (p < 0.05) and persisted to be so in several different meta-analyses (Table S15), it was excluded from the primary analysis. Abbreviations: DRM, dose-response meta-analysis; GI, glycemic index; n, number of published reports; k, number of studies (more than one may appear in one report); T2D, type 2 diabetes.

Figure 3

Forest plot of the dose-response T2D–GI risk relation by validity of the dietary instrument. The box size is proportional to the weight contributed by the study to the combined study mean. Horizontal lines span individual study 95% confidence intervals. Arrowheads indicate truncations. Diamonds represent the combined study mean RR values per 10 GI units (based on the glucose standard), and the corresponding 95% CI values. Analysis was undertaken on the natural logarithms and are shown untransformed. p-Values were calculated using the z test for RR and the Chi² test for I². Eligible studies were: Barclay et al. (2007) [68], Bhupathiraju et al. (2014) (NHS II) [10], Bhupathiraju et al. (2014) (HPS) [10], Hodge et al. (2004) [69], Krishnan et al. (2007) [43], Mekary et al. (2011) [40], Meyer et al. (2000) [64] (p = 0.011), Mosdol et al. (2007) [65], Oba et al. (2013) (men) [44], Oba et al. (2013) (women) [44], Sahyoun (2008) [70], Sakurai et al. (2012) [71], Sluijs et al. (2013) [67], Stevens et al. (2002) (AA) [66], Stevens et al. (2002) (EA) [66], van Woudenbergh et al. (2011) [72], and Villegas et al. (2007) [46]. An outlier study was: Simila et al. (2011) [41] (p = 0.033). Abbreviations: AA, African-American; EA, European-American; CI, confidence interval; HPS, Health Professionals Study; I²; inconsistency; NHS, Nurses’ Health Study; p, probability; RR, risk relation. Footnotes: a, CORR refers to the dietary instrument’s correlation coefficient for carbohydrate or carbohydrate foods; b, percentage weights are from random effects; c, the study of Sluijs et al. [67] reported no overall correlation for their combined regions, some regions had correlations <0.55.

Figure 4

Funnel plot of the study results for the T2D–GI risk relation for valid studies using valid dietary instruments. Valid studies were those with valid dietary instruments (correlation coefficient for carbohydrate or carbohydrate foods >0.55, deemed a prior) and not including an outlier study [38]) (Table S15) (p < 0.05, deemed a priori in the random effects model. The inverted funnel shows the study-level 95% confidence limits with fixed effects, the outer funnel shows the study-level 95% confidence limits with random effects, and the vertical dashed line shows the combined studies random effects trim-and-fill mean, which exponentiated was 1.27 (1.15–1.40) (p < 0.001, n = 10 studies).

Figure 5

Global dose-response meta-analysis of the T2D–GI risk relation. The larger plot shows the mean increment in the log risk relation and 95% CI values. The 95% CI values were based on random effects analysis, the inner (black dashed) lines show the conventional CI values based on (τ² + σ²) whereas the outer (grey dashed) lines were based on an additional forecasting term, σ_f², so that the outer 95% CI was based on (τ² + σ² + σ_f²). Insert: Shows the corresponding unlogged RR (black, continuous), the lower 95% confidence limit when accounting for σ_f² (black dashed), and the corresponding log-linear prediction (red).

Figure 6

Process of the review and outcomes of the inclusion and exclusion criteria for the T2D–GL risk relation. Literature searches were for prospective cohort studies investigating this relation published in the period of 1946 to 6 December 2018 via MEDLINE and EMBASE and other sources—see the Methods section and Figure 1. Superscripts note: a. Whether T2D is related to the dietary glycemic load (GL) and whether the risk relation is > 1.20 with a lower confidence limit > 1.10 for each 10th to the 90th percentile of GL intake (which, on analysis, was per 80 g GL in 2000 kcal (8400 kJ)? b. Is the T2D–GL risk relation conditional on the validity of the dietary instrument for carbohydrate, on the sex of the participants, or on the BMI (kg/m²), and was the T2D–GI risk relation conditional on protein intake [10] and/or alcohol (ALC) intake [40]? c. Details of the full reports excluded are described in the associated text and in more detail in Section 3 (h to l) of the Supplementary Materials. d. Studies used (n = 22) in the meta-analysis with multiple covariates. e. Studies were invalid if the dietary instrument had a correlation coefficient for carbohydrate foods ≤ 0.55. f. When a study suspected of being an outlier in a forest plot was proven to be so upon meta-regression analysis (p < 0.05) and persisted to be so in several different meta-analyses (Table S15), it was excluded. Abbreviations: DRM, dose-response meta-analysis; GL, glycemic load; n, number of published reports; k, number of studies (more than one may appear in one report); T2D, type 2 diabetes.

Figure 7

Forest plot of the dose-response T2D–GL risk relation by the validity of the dietary instrument. The box sizes are proportional to the weight contributed by the study to the combined-study mean. Horizontal lines span the individual study 95% confidence intervals. Arrowheads indicate truncations. Diamonds represent the combined-study mean RR values per 80 g GL in 2000 kcal diet (based on the glucose standard), and the corresponding 95% CI values. Analysis was undertaken on the natural logarithms and shown untransformed. p-values were calculated using the z test for RR and the Chi² test for I². Eligible studies were: Krishnan et al. (2007) [43], Stevens et al. (2002) [66] (two studies), Meyer et al. (2000) [64], Mosdol et al. (2007) [65], Hopping et al. (2010) [42] (six studies), Hodge et al. (2004) [69], Mekary et al. (2011) [40], Halton et al. (2008) [4] pre-combined as “HaltMeka”, Sakurai et al. (2012) [71], Patel et al. (2007), [61], Schulze et al. (2004) [45], Villegas et al. (2007) [46], Salmeron et al. (1997) [60] in men, and Sluijs et al. (2010) [62]. For studies with CORR > 0.55 and considering the relation with CORR (p = 0.002), the study of Simila et al. (2011) [41] was an outlier (p = 0.013) whereas the study of Sluijs et al. (2010) was not (p = 0.052). Abbreviations: AA, African-American; EA, European-American; CI, confidence interval; HPS, Health Professionals’ Study; I²; inconsistency; NHS, Nurses’ Health Study; p, probability; RR, relative risk. Superscripts: a, CORR refers to the dietary instrument’s correlation coefficient for carbohydrate or carbohydrate foods; b, percentage weights are from random effects.

Figure 8

Funnel plot of the study results for the T2D–GL risk relation for valid studies. Valid studies were those with a correlation coefficient for carbohydrate or carbohydrate foods (CORR > 0.55, deemed a priori) and not including one outlier study [41] (Table 4 footnote g) (p < 0.05, deemed a priori) in the random effects model. Observations were adjusted for variation in CORR > 0.55 since this was a significant covariate (p < 0.001) deemed to be taken into control a priori [13]. The combined studies’ random effects trim-and-fill mean, when exponentiated, was 1.36 (1.28–1.45) (p < 0.001, n = 15 studies) per 80 g/d GL in a 2000 kcal (8400 kJ) diet. Trim-and-fill analysis identified missing studies = 0 to obtain a symmetrical distribution of observations. Adjustment for CORR was centered on a correlation coefficient of 0.7 (deemed a priori).

Figure 9

Possible dependence of the size of the T2D–GL risk-relation on the average population ALC consumption. The T2D–GL risk relation shown is per 80 g/d GL in a 2000 kcal (8400 kJ) diet. The negative slope was significant (p = 0.039, n = 11 studies). Circles represent individual study relative risks and have a size proportional to the study weight (larger circles: Greater weight). Lines show the trend and 95% CIs. Risk relations were adjusted for covariates (CORR, ethnicity (ETH), and duration of follow-up in years (FUY) defined in Section 3.3.14) while the population mean ALC intake was centered on 7g/d: These intakes were: Villegas et al. (2007) (2.3 g/d) [46], Schulze et al. (2003) (3.3 g/d) [45], Hodge et al. (2004) (4.3 g/d) [69], Sluijs et al. (2010) (4.6 g/d) [62], Halton et al. (2008) at 20-year follow-up [4] pre-combined with Mekary et al. (2011) at 26-year follow-up from the same study [40] without increasing the weight of the study (6.5 g/d), Krishnan et al. (2007) (6.5 g/d) [43], van Woudenbergh et al. (2011) (7 g/d) [43], Meyer et al. (2000) (7 g/d) [64], Salmeron et al. (1997) in men (11.9 g/d) [60], Similar et al. (2011) (11.9 g/d) [41], and Mosdol et al. (12.9 g/d) [65].

Figure 10

Global dose response meta-analysis of the T2D–glycemic load risk relation. The larger plot shows the mean increment in log risk relation and 95% CI values. Curves are the mean cubic-spline log risk relation (black, continuous) and the 95% CI values based on random effects analysis. The inner (black dashed) shows the conventional CI values based on (τ² + σ²) whereas the outer (grey dashed) were based on an additional, forecasting term, σ_f², thus (τ² + σ² + σ_f²), so that the 95% CIs were wider. Insert: Shows the RR (black, continuous line), the lower 95% confidence limit accounting for the three sources of error (black dashed line), and the corresponding log-linear prediction (red). The risk relation (RR) was per 80 g/d GL in a 2000 kcal (8400 kJ) diet.