Glycemic response and healtha systematic review and meta-analysis: the database, study characteristics, and macronutrient intakes15

Geoffrey Livesey, Richard Taylor, Toine Hulshof, and John Howlett

ABSTRACT Background: Reduction of dietary glycemic response has been proposed as a means of reducing the risk of diabetes and coronary heart disease. Its role in health maintenance and management, alongside unavailable carbohydrate (eg, fiber), is incompletely understood. Objective: We aimed to assess the evidence relating the glycemic impact of foods to a role in health maintenance and management of disease. Design: We searched the literature for relevant controlled dietary intervention trials on glycemic index (GI) according to inclusion and exclusion criteria, extracted the data to a database, and synthesized the evidence via meta-analyses and meta-regression models. Results: Among literature to January 2005, 45 relevant publications were identified involving 972 subjects with good health or metabolic disease. With small reductions in GI (10 units), increases in available carbohydrate, energy, and protein intakes were found in all studies combined. Falling trends in energy, available carbohydrate, and protein intakes then occurred with progressive reductions in GI. Fat intake was essentially unchanged. Unavailable carbohydrate intake was generally higher for intervention diets but showed no trend with GI (falling or rising). Among studies reporting on GI, variation in glycemic load was approximately equally explained by variation in GI and variation in available carbohydrate intake. An exchange of available and unavailable carbohydrate (1 g/g) was evident in these studies. Conclusions: Among GI studies, observed reductions in glycemic load are most often not solely due to substitution of high for low glycemic carbohydrate foods. Available carbohydrate intake is a confounding factor. The role of unavailable carbohydrate remains to be accounted for. Am J Clin Nutr 2008;87(suppl):223S36S. KEY WORDS Carbohydrate, glycemic response, glycemic index, glycemic load, meta-analysis, systematic review INTRODUCTION

Among early scientific enquiry is that showing that unavailable carbohydrate does not adversely elevate blood glucose concentrations and so is a useful nutrient source for persons with diabetes; available carbohydrate was seen as being adverse or not well tolerated (8). That some available carbohydrate might also be suitable as an energy source for persons with diabetes became evident later; this was characterized as low-GI available carbohydrate (9). The idea that unavailable carbohydrate might have a direct or indirect role in glycemic control has renewed interest, with 8 possible mechanisms proposed (10 18). Most recently, a short narrative review of replacement of ingredient available carbohydrate with ingredient unavailable carbohydrate suggested reductions in fasting blood glucose or glycated proteins in persons with diabetes but not in persons in whom fasting blood glucose was not raised (19). Among all these reports is consideration that for optimal glycemic control, unavailable carbohydrate might be used alongside low-GI available carbohydrate to limit the amount of high-glycemic carbohydrate among food choices. However, the relative importance of unavailable and low-glycemic available carbohydrate in health promotion and management is unknown. The objectives of the present study were to construct a database of randomized controlled (or similar) intervention studies that could be used to address queries about the possible role of GL and indexes of GL, as modifiable by available and unavailable carbohydrate, in the management of health and prevention of disease in respect of common metabolic conditions. In this, the first of 2 articles in this issue of the Journal, the database is
From Independent Nutrition Logic, Wymondham, Norfolk, United Kingdom (GL and RT); Kellogg Europe, Den Bosch, Netherlands (TH); and Wembley Park, Middlesex, United Kingdom (JH). 2 Presented at an ILSI Europe workshop titled Glycemic Response and Health, held in Nice, France, on 6-8 December 2006. 3 The review was commissioned by the Dietary Carbohydrates Task Force of the European Branch of the International Life Sciences Institute (ILSI Europe) and was funded by industry members Cerestar, Coca-Cola, Danisco, Groupe Danone, Kellogg, Kraft Foods, National Starch, Nestl, RHM Technology, Royal Cosun, Su dzucker, Tate & Lyle Specialty Sweeteners, and Unilever. The opinions expressed herein are those of the authors and do not necessarily represent the view of ILSI or ILSI Europe. 4 Address reprint requests to ILSI Europe. E-mail: publications@ ilsieurope.be. 5 Address correspondence to G Livesey, Independent Nutrition Logic, Pealerswell House, 21 Bellrope Lane, Wymondham, Norfolk, NR18 0QX United Kingdom. E-mail: glivesey@inlogic.co.uk.
1

Downloaded from www.ajcn.org by on February 3, 2010

Reduction of the glycemic response to foods, via either reduced glycemic index (GI) (1) or reduced glycemic load (GL) (2) has been proposed as a dietary means to help to combat diabetes mellitus and possibly coronary heart disease (CHD) (3, 4). Excessive glycemic response to carbohydrate foods and low unavailable carbohydrate intake are also implicated in stroke (5) and certain forms of cancer, in particular colorectal cancer, in some groups (6, 7).

Am J Clin Nutr 2008;87(suppl):223S36S. Printed in USA. 2008 American Society for Nutrition

223S

224S

LIVESEY ET AL
TABLE 2 Exclusion criteria used when screening titles and abstracts 1) 2) 3) 4) 5) 6) Epidemiologic studies Interventions to assess effect on any form of cancer Interventions in animal studies Interventions with a historical design Studies clearly without a control treatment Treatments targeting the use of unavailable carbohydrate, protein, or fat in place of total or available carbohydrate in the absence of a primary targeting of a reduction in glycemic index of available carbohydrate Treatments reducing the glycemic index of the diet by using inhibitors of carbohydrate digestion Change in glycemic index or load as part of a multiple intervention not reflected in the control treatment, such as with drugs or lifestyle factors Interventions specifically to study satiety, which had no relevant outcomes Publications related to food product characteristics, including determinations of glycemic index or load Methodologic publications Publications relating to exercise performance Publications on guidelines, reviews, commentaries, and abstracts Study proposals, commentaries, and reviews

described together with an assessment of the extent to which the studies achieved reductions in GI without modifying the intake of other macronutrient energy sources. Use of the database to examine the evidence on the relations between glycemic response to food and specific aspects of health is reported separately in the second article.
MATERIALS AND METHODS

The database was constructed by using 3 processes applied sequentially: 1) a literature search; 2) examination of titles, abstracts, and full articles; and 3) database construction, data analysis, and synthesis of evidence. The electronic databases searched included the Cumulative Index to Nursing & Allied Health Literature (CINAHL; Internet: www.cinahl.com) from 1982 to January 2005, the Cochrane Central Register of Controlled Trials (CENTRAL; Internet: www.mrw.interscience. wiley.com/cochrane/cochrane_clcentral_articles_fs.html) to January2005,theElsevierMedicalDatabase(EMBASE;Internet:www. embase.com) from 1980 to December 2003, the US National Library of Medicine database (MEDLINE via the PubMed portal; Internet: www.ncbi.nlm.nih.gov.80/sites/entrez/) from 1950 to January 2005; Elseviers Science-Specific Search Engine on the Internet (SCIRUS; Internet: scirus.com) to January 2005, and Blackwells Nutrition and Food Science database underpinned by the Commonwealth Agricultural Bureau International (CABI; Internet: www.nutritionandfoodsciences.org) from 1981 to January 2005. All field searches for glyc(a)emic index or glyc(a)emic indices or glyc(a)emic load alone and each together with diabetes were performed. Previous meta-analyses (4, 20 22) and reviews (23) were also examined to find citations and enabled a relative assessment of search success. Records were imported into a bibliographic database (Endnote 7, published 2003; ISI Research Soft, Berkeley, CA) to exclude duplicates automatically, with further screening manually to exclude remaining duplicates. Potentially relevant studies were identified by screening titles and abstracts by using the inclusion and exclusion criteria given in Tables 1 and 2, respectively. Full articles identified as being potentially relevant were subjected to detailed examination against these criteria and were included only if the information could be extracted and the articles were from English language publications, from a foreign language publication with English abstract for which additional information could be sought from the authors, or from unpublished work available by that time. Study quality was assessed on the basis of criteria in the Cochrane Reviewers Handbook (24). Numeric and alphanumeric (string) data were double extracted from the source publications to 2 provisional databases by RT and GL independently. Inequalities in cell contents between the provisional databases were examined by RT and GL separately and corrections made. Where
TABLE 1 Inclusion criteria used when screening titles and abstracts 1) Publications in which glycemic index or load referred to foods or diets and not to a diabetological parameter 2) Controlled intervention studies 3) Duration of treatment 1 wk 4) Studies in healthy persons, persons with either impaired glucose tolerance or hyperlipidemia or at risk of coronary artery disease or having diagnosed type 1 or 2 diabetes

7) 8)

9) 10) 11) 12) 13) 14)

Downloaded from www.ajcn.org by on February 3, 2010

inconsistencies remained, agreement was reached by joint consultation of the publication (or report) of the study in question. The agreed data were imported and stored in a StataCorp (College Station, TX) STATA SE 9.0 database file (*.dta) for query and analysis. Conversion of data reported in common units to SI units was made by GL and RT independently (Table 3) when they extracted the data. Subject groups were classified according to the following health types: those in good health, those with impaired glucose tolerance, those with type 1 diabetes mellitus, those with type 2 diabetes mellitus, and those with CHD. Subject groups were also classified according to body weight as normal, overweight, or obese according to body mass index (cutoffs in kg/m2 of 25 or 30) or the original author information. Biochemical risk factors extracted were fasting blood glucose, fasting insulin, glycated proteins (HbA1c and fructosamine), insulin sensitivity, retrospectively calculated insulin sensitivity by homeostatic model assessment (HOMA %S), pancreatic B-cell function calculated retrospectively by homeostatic model assessment (HOMA %B), cholesterol (total, LDL, and HDL), and
TABLE 3 Conversion factors Variable Glycemic index Fasting glucose Fasting insulin Serum triglycerides (triacylglycerols) Fasting capillary blood glucose Metabolizable energy1 Available carbohydrate Unavailable carbohydrate Protein Fat
1

Conversion factor % glucose 1.43 % bread 1 mg/dL 18.02 mmol/L 1 U/mL 0.143 pmol/mL 1 mg/dL 88.5 mmol/L 0.61 (0.94 fasting venous glucose) 16.7 kJ/g 8 kJ/g 16.7 kJ/g 37.7 kJ/g

kcal 4.184 kJ.

FOOD, GLYCEMIA, AND HEALTH: DATABASE AND DIETS

225S

plasma triacylglycerols. Body weight was the only constitutional risk factor extracted. Dietary risk or nutritional factors extracted were metabolizable energy, fat, available carbohydrate, unavailable carbohydrate, and protein intake, together with the potential risk factor GI with calculation of GL. Duration of treatment was extracted either as a continuous variable (weeks) or as a categorical variable or 12 wk. For the glycated proteins, treatment duration was often low relative to half-life in blood, and so these were considered with and without adjustment for half-life and duration of treatment. Random effects meta-analyses and meta-regressions, each weighted by inverse variance, were undertaken by using meta and metareg in STATA 9.2 SE (StataCorp) according to Cochrane guidelines (25). Meta provides combined means with 95% CIs, the Der Simonian and Laird estimate of betweenstudies variance (DSL), Q-test for heterogeneity, and an asymptotic z test for the null hypothesis that the true effect is zero. Metareg was (unless stated otherwise) executed by using restricted maximum likelihood (REML) for the fitting of coefficients, Knapp & Hartungs coefficient SE for t test of significance, likelihood-ratio estimates of between-studies variance (Tau2), a likelihood-ratio test for heterogeneity, and an expression of the proportion of total variation due to heterogeneity (I2). Optionally, with regression models including a constant, the Monte-Carlo (distribution free) permutations test was used to assess whether a coefficient differed significantly from zero (STATA option permut with metareg fitted by the method of moments). Outliers were identified during sensitivity analysis by means of the statistic 1Bij, ie, the difference between the coefficient with and without the outlier divided by the SE without. Mean differences between dietary treatments were expressed either in absolute units when the metric was common to all studies or as a percentage of the average of the mean dietary treatments when the metrics differed among studies. Parallel and crossover studies reporting no crossover effects were combined. Studies generally reported either treatment differences in end measurements or differences in change with time (change score) for each treatment or both. When an SE difference (SED) between treatments in end measures or change in measures in time was not reported by the original authors, it was calculated from either CIs or P values, error df, the number of multiple treatment comparisons reported, and the method for testing the significance (Students paired t test, Tukeys test, etc). Studies often did not report either the dependent (unpaired) or the independent (paired) SED between treatments. These were therefore calculated from the dependent SED at the end of the period of treatment and coefficients of correlation (rp) between dependent and independent values that were imputed from other studies, with the use of duration-dependent values of rp when found appropriate. Calculation of differences in each study (treatment mean control mean and independent SED for treatment effects) was possible by several ways depending on the information available. The accuracy of a method and the availability of information for the method dictated preferences. For mean treatment difference values, the preference was m1m2m3m4. m1 was the reported treatment difference (corrected for starting values). m2 was the treatment difference calculated from reported change scores. m3 was the same as m2 but first calculating the change scores from reported start and end values for each treatment. m4 was the difference in reported end values for each treatment in

crossover designs. Similarly, SEDs between treatments took on methods preferences, SED1SED2SED3. SED1 was the reported value. SED2 was calculated from change score SEs (combined treatment SED across time). SED3 was estimated from pooled treatment end mean and SD values, the number of participants, and rp. Preferences for SDs for start and end mean values were SD1SD2SD3. SD1 was as originally reported, whereas SD2 was calculated from dependent SE of means and the number of observations. SD3 was imputed by using first or second order regressions relating SD1 to the corresponding mean for those studies imparting this information. The preferences were elaborated to maximize precision and minimize bias for combined study treatment effects and was essential to retaining the maximum number of studies in the analysis. Doing so avoided bias from dropping studies with incompletely reported statistics but included bias due to imprecision of SD3, and so SED3. Meta-analytic methods make much use of graphic material. In presenting the results, we give both figures and related data in tables because this is preferred to ensure provision of both perspective and precision for the outcomes (24, 25). Where further analysis is made to test the significance of a particular perspective, this entailed the provision of further figures and associated outcomes.

Downloaded from www.ajcn.org by on February 3, 2010

RESULTS

The database and study characteristics The search for literature to January 2005 yielded 2782 potentially relevant publication records. After the titles and abstracts were screened and the full articles reviewed, 45 publications were identified as being relevant and of suitable quality for inclusion in the present database. Because some publications reported more than one study, because of repeat observations in some studies (across the duration of treatment), and because some outcomes were assessed by more than one method, the 45 publications yielded 80 conditions contrasting the reported high versus low GI diets (Table 4). All included studies had control versus treatment comparisons and designs that were analyzable as either crossover (20 studies) or parallel (25 studies). Information was monitored at different times within each study (see Table 4). This was used to help address questions related to the persistence, gain, or loss of effect with time during treatment. Unless stated otherwise, independence of study observations was preserved by including only the end of treatment results. The database included 972 participants per treatment arm with group mean ages of 10 to 63 y, with both males (n 511) and females (n 461) represented. Participants were either normal weight (16 studies), overweight (18 studies), or obese (10 studies) or were unclassified by weight (1 study). Similar numbers of participants took part in each treatment arm (770 for the high-GI treatment and 793 for the low-GI treatment). Study participants were either healthy (ie, no diagnosis of disease was evident; 13 studies) or had impaired glucose tolerance (2 studies, duration of impairment unknown), had type 1 diabetes (7 studies, mean duration from diagnosis of 3 to 16 y, 1 unknown duration), had type 2 diabetes (17 studies, mean duration from diagnosis of 0 to 12.5 y, 7 unknown duration), were at risk of primary (4 studies, duration unknown) or secondary (1 study, duration unknown) CHD, or had hyperlipidemia (1 study combining Type II a, Type

226S
TABLE 4 Study characteristics Duration of diagnosis y Normal healthy Frost et al 1998 (30) (n is 2) Frost et al (1998) (30) (n) Herrmann et al 2001 (31) (20F) Herrmann et al 2001 (31) (30F) Jenkins et al 1987 (32) (d1) Jenkins et al. 1987 (32) (d2) Kiens et al 1996 (26) (d1) Kiens et al 1996 (26) (d2) Kiens et al 1996 (26) (d3) Kiens et al 1996 (26) (d4 is 2) Kiens et al 1996 (26) (d4) Kurup 1992 (33) (raggi vs rice) Kurup 1992 (33) (raggi vs tapioca) Bouche et al 2002 (34) Dumesnil 2001 (35) (ph1 vs LGI out) Pereira et al 2004 (36) (out) Price (unpublished) Sloth et al 2004 (37) (d1) Sloth et al 2004 (37) (d2) Sloth et al 2004 (37) (d3) Sloth et al 2004 (37) (d4) Sloth et al 2004 (37) (d5) Agus et al 2000 (38) (out) Carels et al 2005 (39) Ebbeling et al 2003 (40) (d1) Ebbeling et al 2003 (40) (d2) Spieth et al 2000 (41) Impaired glucose tolerance Wolever & Mehling 2002 (42) (d1) Wolever & Mehling 2002 (42) (d2) Wolever & Mehling 2002 (42) (d3) Wolever & Mehling 2002 (42) (d4 is 2) Wolever & Mehling 2002 (42) (d4) Slabber et al 1994 (43) (C) Slabber et al 1994 (43) (P) Type 1 diabetes Calle-Pascual 1988 (44) (T1) Collier et al 1988 (45) (d1) Collier et al 1988 (45) (d2) Fontvieille et al 1988 (46) Fontvieille et al 1992 (47) (mT1) Giacco et al 2000 (29) Gilbertson et al 2001 (48) Lafrance et al 1998 (49) Type 2 diabetes Jenkins et al 1988 (50) Brand et al 1991 (1) Calle-Pascual 1988 (44) (T2) Fontvieille et al 1992 (47) (mT2) Frost et al 1994 (51) Jarvi et al 1999 (52) Jimenez-Cruz et al 2003 (53) Kabir et al 2002 (27) Komindr et al 2001 (54) Tsihlias et al 2000 (55) (d1) Tsihlias et al 2000 (55) (d2) Wolever et al 1992 (56) (d1) Wolever et al 1992 (56) (d2) 3 3 14.6 13.4 10.3 3.7 15.8 12.5 5 7.8 0.5 8 8 n n n n n n n n n n n n n ow ow ow ow ow ow ow ow ow ob ob ob ob ob ow ow ow ow ow ob ob n n n n n n n n n ow ow ow ow ow ow ow ow ow ow ow ow Body weight band2

LIVESEY ET AL

Sources (reference)1

Age y 36 36 27 27 33 33 24 24 24 24 24 36 36 46 47 31 30 30 30 30 30 28 44 16 16 10 57 57 57 57 57 35 35 26 12 12 44 43 28 11 65 62 59 56 55 66 59 59 46 62 62 67 67

Percentage male % 0 0 56 56 100 100 100 100 100 100 100 100 100 23 0 0 0 0 0 100 0 31 31 46 21 21 21 21 21 0 0 86 86 50 67 37 50 78 25 63 67 71 75 43 100 0 59 59 47 47

Randomization reported3

Study design4

No. of participants on high GI n

No. of Duration participants of on low GI treatment n 6 6 9 9 6 6 7 7 7 7 7 40 23 11 12 22 22 23 23 23 23 23 10 21 7 7 64 13 13 13 13 13 16 15 12 7 7 8 12 24 55 9 8 16 12 6 25 20 14 13 10 30 30 15 15 w 3.0 3.0 1.1 1.1 1.0 2.0 1.0 2.0 3.0 4.3 4.3 2.1 2.1 5.0 0.9 9.6 12.0 2.0 4.0 6.0 8.0 10.0 0.9 20.0 26.0 52.0 18.4 4.0 8.0 12.0 16.0 16.0 12.0 12.0 4.0 3.0 6.0 3.0 5.0 24.0 52.0 1.7 2.0 12.0 4.0 5.0 12.0 3.4 6.0 4.0 4.0 13.0 26.0 1.0 2.0

r r r r r r r r r r r nr nr r r r r r r r r r r r r r nr r r r r r m m nr r r r r r r r r r nr r r r r r r r r r r

p p x x x x x x x x x p10 p10 x x p p p p p p p x p p p p p p p p p x p p x x x x p p x x x p x p x x x x p p x x

6 6 9 9 6 6 7 7 7 7 7 35 35 11 12 17 22 22 22 22 22 22 10 23 7 7 43 11 11 11 11 11 16 15 12 7 7 8 12 22 49 9 8 16 12 6 26 20 14 13 10 29 29 15 15

Downloaded from www.ajcn.org by on February 3, 2010

(Continued to the right)

FOOD, GLYCEMIA, AND HEALTH: DATABASE AND DIETS

TABLE 4 (Continued) No. of meals with treatment n/d 0 0 3 3 3 3 3 3 3 3 3 1 1 0 3 3 0 0 0 0 0 0 3 3 3 3 3 3 3 3 3 3 3 3 1 0 0 0 3 2 0 3 3 3 1 3 3 3 0 1 3 1 1 3 3 lc lc lc lc c c c c c c c c c lc a c lc a a a a a c c a a a a a a a a lc lc c lc lc lc c lc a lc lc c c c lc c lc c lc lc lc c c Food intake control5 Energy intake (high GI)6 kJ/d 8368 8368 7467 7333 11 506 11 506 10 309 10 309 10 309 10 309 10 309 8335 8335 10 301 11 695 6276 9800 9800 9800 9650 9650 6247 6941 6711 6021 7170 7170 7170 7170 7170 5124 5124 8304 11 452 11 071 8862 7477 7653 7704 5385 6790 8531 7477 7531 7615 6530 6168 8400 8400 6017 6017 Available CHO intake (high GI)7 %E 48 48 65 55 62 62 47 47 47 47 47 59 59 42 55 65 58 58 58 58 58 67 53 55 55 58 53 53 53 53 53 50 50 60 46 47 45 51 54 49 53 53 46 60 51 44 54 64 58 54 54 59 59 Fat intake (high GI)7 %E 39 39 20 30 20 20 41 41 41 41 41 27 27 37 30 18 21 21 21 22 22 18 32 28 29 28 28 28 28 28 28 30 30 20 37 37 36 39 29 34 29 26 31 20 39 32 29 20 30 29 29 21 21 Protein intake (high GI)7 %E 13 13 15 15 19 19 13 13 13 13 13 14 14 18 15 17 17 17 17 17 17 15 17 18 18 18 17 17 17 17 17 20 20 20 17 16 17 21 17 17 18 21 19 20 21 18 18 12 17 17 20 20 Unavailable CHO intake (high GI) g/d 23.8 23.8 57.8 57.8 7.3 7.3 7.3 7.3 7.3 17.8 17.8 19.0 26.0 20.0 36.0 36.0 36.0 35.0 35.0 25.6 9.0 9.5 22.7 22.7 22.7 22.7 22.7 12.9 22.5 23.5 33.6 27.0 15.0 17.1 28.0 26.0 12.9 27.0 19.6 34.0 25.0 23.1 23.1 33.4 33.4 Change in unavailable CHO intake g/d 0.0 0.0 13.8 13.8 11.7 11.7 11.7 11.7 11.7 0.9 3.4 12.0 3.0 12.0 1.0 1.0 1.0 2.0 2.0 3.4 1.0 0.5 13.5 13.5 13.5 13.5 13.5 13.7 16.2 15.5 6.1 0.0 24.1 1.8 6.0 0.0 13.7 0.0 9.8 4.0 9.0 27.2 27.2 2.4 2.4 Glycemic index (high GI)8 % 64 64 669 669 74 74 64 64 64 64 64 6811,12 6811,12 71 60 59 6811 6811 6811 6811 6811 57 56 56 59 59 59 59 59 6211,12 58 58 60 64 64 57 66 65 64 6211,12 64 59 59 56 61 6811,12 62 62 62 62 Change in GI % 13 13 22 22 29 29 17 17 17 17 17 5 5 30 7 23 13 13 13 13 13 5 3 3 5 5 5 5 5 7 9 9 14 26 14 1 19 17 9 7 26 4 19 12 4 18 7 7 19 19 Glycemic load8 g eq/d 153 153 1919 1599 317 317 186 186 186 186 186 20111,12 20111,12 184 231 143 22911 22911 22911 22411 22411 125 124 111 134 134 134 134 134 18511,12 183 180 144 147 160 161 110 120 19011,12 147 116 145 139 14611,12 165 165 131 131

227S

Change in glycemic load g eq/d 31 31 78 64 134 134 51 51 51 51 51 29 8 96 127 85 55 55 55 56 56 8 21 0 8 8 8 8 8 21 5 2 29 56 40 36 21 20 22 56 5 40 45 39 39 39 41 41

Downloaded from www.ajcn.org by on February 3, 2010

(Continued on next page)

228S
TABLE 4 (Continued) Duration of diagnosis y Heilbronn 2002 (57) (d1) Heilbronn 2002 (57) (d2) Jimenez-Cruz et al 2004 (58) Luscombe et al 1999 (59) Rizkalla et al 2004 (60) Rizkalla et al 2004 (60) (is 2) Wolever et al 1992 (61) (d1) Wolever et al 1992 (61) (d2) Wolever et al 1992 (61) (d3) Zhang et al 2003 (62) Hyperlipidemia Jenkins et al 1987 (63) (all) Jenkins et al 1987 (63) (fam) Jenkins et al 1987 (63) (TIIa) Jenkins et al 1987 (63) (TIIb) Jenkins et al 1987 (63) (TIII) Jenkins et al 1987 (63) (TIV) Coronary heart disease risk Frost et al 1998 (30) (chdr) Frost et al 1998 (30) (chdr is 2) Frost et al 1996 (64) Frost et al 1996 (64) (is 2) Brynes et al 2003 (28) (st is 2) Brynes et al 2003 (28) (st) Brynes et al 2003 (28) (su is 2) Brynes et al 2003 (28) (su) Frost et al 2004 (65)
1

LIVESEY ET AL

Sources (reference)1

Body weight band2

Age y

Percentage male % 51 51 67 100 100 50 50 50 73 73 33 86 100 81 0 0 77 77 100 100 100 100 87

Randomization reported3

Study design4

No. of participants on high GI n

No. of Duration participants of on low GI treatment n 24 24 8 21 12 12 6 6 6 36 30 6 6 7 1 16 8 8 15 15 17 17 17 17 26 w 4.0 8.0 3.0 4.0 4.0 4.0 2.0 4.0 5.7 21.0 4.0 4.0 4.0 4.0 4.0 4.0 3.0 3.0 4.0 4.0 3.4 3.4 3.4 3.4 12.0

5 5 7 6.3 3 4.7 4.7 3.8 6.2 5.3 4.8

ob ob ob ob ob ob ob ob ob n n n n n n n n ow ow ow ow ow ow ow

57 57 51 57 54 54 63 63 63 51 53 53 47 51 57 55 36 36 63 63 45 45 45 45 63

r r r r r r r r r r nr nr nr nr nr nr r r r r r r r r r

p p x x x x x x x p x13 x13 x13 x13 x13 x13 p p p p x x x x p

21 21 8 21 12 12 6 6 6 36 30 6 6 7 1 16 8 8 15 15 17 17 17 17 29

Downloaded from www.ajcn.org by on February 3, 2010

(Continued to the right)

Abbreviations in parentheses distinguish data from within the same publication, which might carry more than one study or note an outlying study: out, study is outlying, ie, the change in glycemic load is not explained by the change in carbohydrate intake or glycemic index. T1, persons with type 1 diabetes; T2, persons with type 2 diabetes; mT1, mT2, observations presented from a mixture of types of diabetes, results applied to each type with appropriate numbers of participants; d1, observations are for the first duration of treatment, d2 for the second, and so on; chdr, those participants at coronary heart disease risk; n, normal participants; 30F, 20F, distinguishes the level of fat intake in two studies in the same publication; fam, refers to a subgroup on which fructosamine concentrations were measured; all, TIIa, TLIb, TIV, refers to the type of hyperlipidemia; P, C, observations representing the same subjects initially in a parallel design and eventually for a subgroup in the cross-over design; ph1, phase 1 diet of the American Heart Foundation; raggi vs rice, study was analyzed in the publication as 2 sequential studies, here analysis was made as a 1 parallel design; raggi vs tapioca, study was analyzed in the publication as 2 sequential studies, here analysis was made as 1 parallel design; st, starch as the major modification; su, sucrose as the major modification; is 2, insulin sensitivity measured by a second method on the whole or a subsample of the study populations. 2 n, ow, ob, normal, overweight, and obese as classified by original authors or at present by BMI with cutoffs at 25 and 30 (kg/m2). 3 r, randomization reported; nr, randomization not reported. 4 x, p, crossover or parallel study design. 5 c, lc, a, controlled, limited controlled, and ad libitum food intake controlled. 6 kJ metabolizable energy (not inclusive of energy from unavailable carbohydrate), kilojoules (1 cal 4.184 kJ), and percentage thereof. 7 Percentage of metabolizable energy, sum values 99% correspond to alcohol consumption 102% (two cases) unexplained. 8 Glycemic index (GI) and load are based on glucose 100 g eq/100 g. GI values are original author values with conversion from bread equivalents where appropriate, or in the case of relatively simple food exchanges are calculated by using information from relevant food tables and international look-up tables for GI. 9 In the case of Herrmann et al 2001 (31), GI values were given as 65 (high-GI diets) and 45 (low-GI diet) for which values of 66 and 44 were taken to give a minimum estimate of treatment difference in GI. 10 Two studies by Kurup 1992 (33) presented by authors in one publication as control-treatment sequence designs had multiple treatments analyzable and are included here with parallel design. 11 Values required assumptions to be made about the underlying diet, which may affect the accuracy of dietary estimates without influence on the treatment difference estimates. 12 GI values of treatment and control food sources estimated from look-up tables, which may affect the accuracy of both the dietary and the difference estimates. 13 A study by Jenkins et al 1987b (63) was presented by the authors as a control-treatment-control design and is presented here by combining controls and analyzed as a crossover design.

FOOD, GLYCEMIA, AND HEALTH: DATABASE AND DIETS

TABLE 4 (Continued) No. of meals with treatment n/d 0 0 3 0 2 2 0 0 0 3 3 3 3 3 3 3 0 0 3 3 3 3 3 3 3 c c lc lc lc lc c c c lc lc lc lc lc lc lc lc lc lc lc a a a a lc Food intake control5 Energy intake (high GI)6 kJ/d 6008 6008 8360 7569 9586 9586 5807 5807 5807 7176 7176 7176 7176 7176 7176 8786 8786 8544 8544 9020 9020 9900 9900 7360 Available CHO intake (high GI)7 %E 61 61 54 53 38 38 57 57 57 49 49 49 49 49 49 51 51 44 44 46 46 51 51 47 Fat intake (high GI)7 %E 17 17 27 21 37 37 23 23 23 29 29 29 29 29 29 37 37 36 36 36 36 33 33 32 Protein intake (high GI)7 %E 22 22 18 23 20 20 19 19 19 19 19 19 19 19 19 12 12 17 17 18 18 16 16 18 Unavailable CHO intake (high GI) g/d 29.8 29.8 30.0 30.0 21.0 21.0 33.0 33.0 33.0 21.8 21.8 21.8 21.8 21.8 21.8 29.4 29.4 18.2 18.2 19.0 19.0 19.0 19.0 21.0 Change in unavailable CHO intake g/d 0.5 0.5 23.0 0.0 13.0 13.0 1.0 1.0 1.0 5.9 5.9 5.9 5.9 5.9 5.9 2.8 2.8 0.7 0.7 12.0 12.0 0.0 0.0 6.0 Glycemic index (high GI)8 % 75 75 72 63 71 71 61 61 61 59 60 60 60 60 60 60 62 62 65 65 69 69 69 69 58 Change in GI % 32 32 12 20 32 32 20 20 20 9 8 8 8 8 8 8 14 14 12 12 20 20 6 6 7 Glycemic load8 g eq/d 164 164 194 151 155 155 122 122 122 125 125 125 125 125 125 166 166 145 145 173 173 209 209 120

229S

Change in glycemic load g eq/d 72 72 46 47 77 77 40 40 40 18 18 18 18 18 18 27 27 37 37 65 65 19 19 7

Downloaded from www.ajcn.org by on February 3, 2010

II b, Type III, mean duration of diagnosis of 4.7 y). The health types were assigned by the authors of the original publications and reports. It should not be implied from the use of the health types that particular participants fit a health type uniquely. For example, all those with diabetes might also be considered at risk of CHD. Studies included participants taking medication. Insulin dosage was reported in all 7 studies of persons with type 1 diabetes, in 2 of 17 studies of persons with type 2 diabetes, and in 1 of 1 study of participants at risk of secondary CHD. All but 1 study of persons with type 2 diabetes included noninsulin medication for glycemic control (hypoglycemic agents). Interventions were by diet, with intention to exchange the form of available carbohydrate (high versus low GI). The extent to which this was achieved varied between studies. In many cases, the dietary changes were accompanied by variably higher quantities of unavailable carbohydrate. Thus, the diets are sometimes qualified collectively as variably lower glycemic and variably higher unavailable carbohydrate. Three studies were identified statistically as outliers, ie, belonging to a population of studies differing from the remainder. These were Agus et al (38), Dumesnil et al (35), and Pereira et al (36) and were identified because of the exceptional replacement of available carbohydrate with either protein or fat. Treatment durations ranged from 1 to 52 wk. Studies included interventions with food exchanges at 1 or 2 or 3 (and more) or probably 3 (and more) meals per day (6, 2, 25, and 12 studies, respectively). This enabled queries concerning meal numbers

and dose dependency of outcomes in relation to dietary GL or any of its indexes. Diets were intended to meet maintenance and growth requirements (37 studies) or submaintenance requirements (8 studies); no overfeeding studies were encountered. Control of food intake was categorized as ad libitum (7 studies), limited controlled food intake (22 studies), and controlled food intake (16 studies). All studies were free-living (no studies of hospitalized patients or subjects housed in metabolic wards or centers of human nutrition). Studies were from all continents: Asia (4 studies), Australasia (6 studies), Europe (19 studies), South America (2 studies), Africa (1 study), and North America (13 studies). Mortality and morbidity scores were not collected, because few studies were of sufficient duration to encounter reliable responses. No data were reported on mortality, cardiovascular events, or diabetic complications. Changes in the severity of risk factors were the primary outcomes: fasting glucose, glycated proteins (HbA1c, fructosamine), fasting insulin, insulin sensitivity, calculated HOMA %S, calculated HOMA %B, calculated HOMA %D (the product of HOMA %B HOMA %S), fasting plasma or serum triacylglycerols, and body weight. Data on fasting total, HDL, and LDL cholesterol were extracted, appeared to be confounded, and were not investigated further at this time. Other outcomes were macronutrient intakes, physical activity, and adverse events. Insufficient numbers of studies monitored physical activity to accumulate information in the database; 3 that did so provided no data (2729). The following intake

230S

LIVESEY ET AL

data were noted: 1) metabolizable energy intake (41 studies, 26 had freedom to respond, ie, ad libitum plus limited controlled food intake; the other 15 had controlled, ie, fixed, food intake), available carbohydrate intake (41 studies, 26 with freedom to respond); 2) unavailable carbohydrate intake (36 studies, 22 with freedom to respond); 3) protein intake (39 studies, 24 with freedom to respond); 4) fat intake (40 studies, 26 with freedom to respond); 5) calculated GL (ie, diet GI times available carbohydrate intake/100; 38 studies, 24 with freedom to respond); 6) dietary GI (as % glucose) ostensibly of available carbohydrate (41 studies, 26 with freedom to respond; 6 studies did not fully report GI values but this was recoverable approximately from other information available (see footnotes to Table 4); 7) calculated GI of total carbohydrate (35 studies, 22 with freedom to respond) were estimated as 100 times the dietary GL divided by the sum weight of available and unavailable carbohydrate. Two publications (28, 31) included more than one treatment (versus control) per study. In one case, dietary fat intake differed between treatments (30); in the other, the source of carbohydrate modifying GI differed (sucrose versus starch; 28). This information was retained in the database because it was particularly useful where study results were heterogeneous. One study reporting an effect on body weight (43) informed about the first part of a crossover study as a parallel design. In this case, analyses were performed on the crossover design, and the parallel study was used only for comparison of combined results by study design. Common to each of the 45 studies entered into the database was information for participants that were compliant to diet (ie, dropouts and noncompliant participants discarded). Twentyone studies had reported outcomes on the basis of compliance to diet analyses, whereas 24 studies reported information that was both intention-to-treat and compliant-to-diet. Among these,

there were no dropouts in 23 studies, and 1 study (30) with dropouts presented both intention-to-treat and compliant-to-diet information. Individual study dropouts ranged from 0% to 61% of study entrants with an unweighted combined study mean of 8%. The number of participants per treatment arm per study was generally small. There were 15 studies for which this number was 10 units, 15 studies with 10 to 20, 11 studies with 20 to 30, and 4 studies with 30 to 60. Power calculations were seldom given, and then only for few of the measurements made. Forty of the 45 studies reported macronutrient intakes and changes in intake between high to low glycemic carbohydrate diets. Of these, 38 reported GI values (or information on foods enabling GI to be estimated). Although many studies reported designs intended to have similar macronutrient intakes in the treatment and control arms, the precision with which this was achieved varied between studies. The possibility that difference in treatment outcomes covary with unintended differences in macronutrient intakes was examined, because such differences between treatments might confound the interpretation of a treatment effect as being due to GI if not accounted for by, for example, covariance. To investigate the relation between GI and the energy and macronutrient intakes from the intervention diets, the studies were either combined or were divided into 3 categories (Table 5). Thus, foods were either available ad libitum or were available in wholly controlled amounts, or were available subject to limited (or partial) control. In the ad libitum category, food intake could vary but diet composition was intended to be fixed (food was provided or largely provided). In the wholly controlled food intake category, both food intake and composition were fixed (food was provided or largely provided and may have been adjusted in amount if body weight changed). In the intermediate,

Downloaded from www.ajcn.org by on February 3, 2010

TABLE 5 Studies included by food intake control category1 Controlled food intake Agus et al 2000 (38) Brand et al 1991 (1) Calle-Pascual 1988 (44) (exp1_T2) Calle-Pascual 1988 (44) (exp2_T2) Carels et al 2005 (39) Fontvieille et al 1992 (47) (exp1) Fontvieille et al 1992 (47) (exp2) Heilbronn 2002 (57) (d2) Jarvi et al 1999 (52) Jenkins et al 1987a (32) (d2) Kiens et al 1996 (26) (d4) Kurup 1992 (33) (raggi vs rice) Kurup 1992 (33) (expt2 raggi vs tapioca) Wolever et al 1992a (61) (d3) Wolever et al 1992a (61) (d2)
2

Limited controlled food intake Bouche et al 2002 (34) Collier et al 1988 (45) (d2) Fontvieille et al 1988 (46) Frost et al 1994 (52) Frost et al 1996 (65) Frost et al 1998 (30) (exp1) Frost et al 1998 (30) (exp2) Frost et al 2004 (64) Giacco et al 2000 (29) Herrmann et al 2001 (31) (expt2) Jenkins et al 1987b (63) (all) Jenkins et al 1988 (50) Jimenez-Cruz et al 2003 (53) Jimenez-Cruz et al 2004 (58) Komindr et al 2001 (54) Lafrance et al 1998 (49) Luscombe et al 1999 (59) Rizkalla et al 2004 (60) Tsihlias et al 2000 (55) (d2)

Ad libitum food intake Brynes et al 2003 (28) (su) Brynes et al 2003 (28) (st) Ebbeling et al 2003 (40) (d2) Sloth et al 2004 (37) (d5) Wolever & Mehling 2002a (42) (d4)

For abbreviations in parentheses, see Table 4. Controlled food intake, both food intake and composition were fixed (food was provided or largely provided and may have been adjusted in amount if body weight changed); limited controlled food intake, scope existed for both intakes and composition to vary (the diet was advised or largely advised, usually with supervision); ad libitum intake, food intake could vary but in these cases diet composition was intended to be fixed (food was provided or largely provided). 2 Outlier in some analyses.

FOOD, GLYCEMIA, AND HEALTH: DATABASE AND DIETS

231S

FIGURE 1. Difference in energy intake associated with difference in glycemic index achieved. Observations are grouped by the category of control over food intake (where total is all categories combined). Each study is represented by a bubble proportional to the inverse (Tau2 SE2). Large bubbles indicate studies with large weight. AC, available carbohydrate; ME, metabolizable energy.

FIGURE 3. Difference in the intake of available carbohydrate associated with difference in glycemic index achieved. Observations are grouped by the category of control over food intake (where total is all categories combined). Each study is represented by a bubble proportional to the inverse (Tau2 SE2). Large bubbles indicate studies with large weight. AC, available carbohydrate. Downloaded from www.ajcn.org by on February 3, 2010

partially controlled category, scope existed for both intakes and composition to vary (the diet was advised or largely advised). Inequalities between treatment and control diets could arise in all 3 categories according to the food choices and application of the dietary prescriptions. Glycemic index and intake of food components Univariate analyses using REML random effects metaregression (Figure 1, Figure 2, and Figure 3) indicate elevated intakes of metabolizable energy, protein, and available carbohydrate for small reductions in GI when comparing treatments with controls. Thus, regression constants at a hypothetical zero change in GI are positive in each food intake category for energy (48 to 1009 kJ/d), protein (1.5 to 15.7g/d), and available carbohydrate (14.9 to 34.6 g/d). Visual inspection of this information collected together (Figure 4) suggests a trend

toward greater such effect as control over food intake is progressively relaxed, from controlled to limited controlled to ad libitum. Multivariate meta-regressions (legend to Figure 4)

FIGURE 2. Difference in protein intake associated with difference in glycemic index achieved. Observations are grouped by the category of control over food intake (where total is all categories combined). Each study is represented by a bubble proportional to the inverse (Tau2 SE2). Large bubbles indicate studies with small errors. AC, available carbohydrate.

FIGURE 4. Confounding increments in energy and macronutrient intakes in studies on glycemic index. Top panel: For small reductions in glycemic index (GI), there is a jump in intake, which is at a theoretical maximum represented by a regression constant in the univariate model shown. The constant occurs at a hypothetical zero reduction in GI. Results are shown for each of the 3 macronutrient and energy intakes, and these for each of the 3 categories of food intake control, providing results for 9 univariate models. The table shows the levels of statistical significance attained for each of the 9 results from the univariate regressions. Also shown in the table are the levels of significance of effects overall* for food intake control categories and the trends** across the food intake control categories, as obtained by application of the multivariate model. Results are converted to units of metabolizable energy for the purpose of display within the same axis. Almost identical levels of statistical significance were given to the trend by the permutations test (0.058, 0.34, and 0.041 in place of the corresponding values in the figures table). Univariate model: intake constant slope GI Tau SE. Multivariate model: intake overall constant* trend** (category -1control, 0limited control, 1ad-lib) slope1 GIad-lib slope2 GIlimited control slope3 GI control Tau SE. Contr, controlled; ltd contr., limited controlled; ad-lib, ad libitum.

232S

LIVESEY ET AL

including categorical variables for controlled, limited controlled, and ad libitum food intake categories support that elevations in intake of metabolizable energy, available carbohydrate, and protein are each highly statistically significant overall, and that each show a trend toward greater elevation as food intake control is relaxed. Such a trend was significant for metabolizable energy and protein, although not for available carbohydrate (Figure 4), likely because of an elevation occurring already in the controlled food intake category. Nevertheless, the visual inspection and the P value suggest a trend is plausible for available carbohydrate. With a progressive reduction in GI, metabolizable energy, available carbohydrate, and protein intake then each decrease in the limited control and ad libitum control food intake categories (Figures 13). Although univariate analysis indicates this to be significant for protein intake only, more sophisticated multivariate analysis using categorical variables for the level of food intake control indicated that a fall in metabolizable energy, available carbohydrate, and protein intake occurs with progressive reduction in GI, each being very significant (Figure 5). Furthermore, relaxation of control over food intake is associated with a trend toward greater rates of reduction in energy and protein intakes, whereas for available carbohydrate such a trend would appear plausible and more probable than not. In contrast with metabolizable energy, protein, and available carbohydrate, the intake of fat appears essentially unchanged by the interventions, or at best fat intake appears not so sensitive to

FIGURE 6. Difference in fat intake associated with difference in glycemic index achieved. Observations are grouped by the category of control over food intake (where total is all categories combined). Each study is represented by a bubble proportional to the inverse (Tau2 SE2). Large bubbles indicate studies with large weight. AC, available carbohydrate. Downloaded from www.ajcn.org by on February 3, 2010

change (Figure 6). Nevertheless, statistically significant effects are not excluded because in the control food intake category a meta-regression constant equal to 8.3 g fat/d (P kh-t 0.004) occurs with a rising slope on the regression line with progressive reduction in GI (0.36 g/GI%glucose, P kh-t 0.013). In the ad libitum food intake group, however, such a trend is possibly reversed, with fat intake becoming reduced along with intake of food energy generally, though this remains to be confirmed (k 5). Unavailable carbohydrate intakes after these lower glycemic interventions are also generally elevated above the intake for control treatments (Figure 7); the increase achieves statistical significance only when all food intake control categories are combined. Thus, the meta-regression constant is 6.9 g/d at the hypothetical zero difference in GI (P kh-t 0.025). No

FIGURE 5. Confounding falls in energy and macronutrient intakes with progressive reduction in glycemic index. Top panel: The progressive fall in intake with reduction in glycemic index (GI) is represented by the slope in the regression models. Results are shown for each of the 3 macronutrient and energy intakes, and these for each of the 3 categories of food intake control, providing results for 9 univariate models. The table shows the levels of statistical significance attained for each of the 9 results from the univariate regressions. Also shown in the table are the levels of significance of effects overall* for food intake control categories and the trends** across the food intake control categories, as obtained by application of the multivariate model. Results are converted to units of metabolizable energy for the purpose of display within the same axis. Almost identical levels of significance were given to the trend by the permutations test (0.003, 0.12, and 0.005 in place of corresponding values in the figures table). Univariate model: intake constant slope GI Tau SE. Multivariate model: intake constant control constantlimited control constant overall slope* GI trend** ( GI category -1control, 0limited control, 1ad-lib) Tau SE.

FIGURE 7. Difference in unavailable carbohydrate intake associated with difference in glycemic index achieved. Observations are grouped by the category of control over food intake (where total is all categories combined). Each study is represented by a bubble proportional to the inverse (Tau2SE2). Large bubbles indicate studies with large weight. AC, available carbohydrate.

FOOD, GLYCEMIA, AND HEALTH: DATABASE AND DIETS

233S

subsequent fall in intake was evident as the glycemic index got progressively lower, such as seen for available carbohydrate. Of possible interest, meta-regression does indicate a falling trend for available carbohydrate intakes with increases in unavailable carbohydrate intake among these studies (Figure 8), which would contribute toward a reduction in GL. The relation is statistically significant when all studies are combined (regression slope 1.09, P kh-t 0.019) and of somewhat consistent sensitivity across categories (1.15 g/g, 1.19 g/g, and 1.37 g/g in the controlled, limited controlled, and ad libitum foods intake groups, respectively), although in no individual category was the effect statistically significant. These slopes are consistent with an exchange of 1 g available carbohydrate with 1 g unavailable carbohydrate as a result of the lower GI interventions; this is independent of the absolute decrease in GI. Heterogeneity (2 - Tau2, the between-studies variance) was highly significant among the study observations. For metabolizable energy, 2 was 157 reduced to 117 kJ2 after accounting for the variation in both GI and category of food intake (ie, 25% of the variance was explained). For available carbohydrate, heterogeneity was reduced 36% from 726 to 462 (g/d)2, whereas for protein it was reduced 52% from 118 to 56 (g/d)2. However, in each case, heterogeneity remained significant (P X 0.001). No reliable (df not low) or significant (P 0.05) patterns emerged among these data within each health type separately (healthy, CHD risk, type 1 diabetes, and type 2 diabetes) to explain these variations. Nor was there evidence of significant residuals for studies combined by health type separately and which could have suggested that a health type departs from these general trends. Glycemic load The GL of the intervention diets decreased with progressive reduction in the GI significantly for all studies combined (P kh-t 0.001) and for studies combined by food intake categories separately (controlled, P kh-t 0.001; limited controlled, P kh-t 0.001; ad libitum, P kh-t 0.05; Figure 9). Each had a corresponding slope on the metaregression lines in the range of from 2.9 to 4.5 g glucose equivalent per unit GI (% glucose) (Table 6). The combined average rate of fall was 3.11 g glucose equivalent per GI % glucose, more

FIGURE 9. Difference in glycemic load associated with difference in glycemic index achieved. Observations are grouped by the category of control over food intake (where total is all categories combined). Each study is represented by a bubble proportional to the inverse (Tau2 SE2). Large bubbles indicate studies with large weight. AC, available carbohydrate. Downloaded from www.ajcn.org by on February 3, 2010

than can be expected from a change in GI alone for carbohydrate intakes 400 g/d. In view of the overlap of the observations and meta-regression lines associating GL and GI for the different food intake control categories (Figure 9) and the limitations of the study numbers, the statistical significance of the differences between the slopes of the met-regression lines for the different food consumption groups has not been assessed. In all studies combined, treatment differences in available carbohydrate intake contributed as much to variation in GL as did GI (Figure 10). Heterogeneity in GL among all studies combined was 892 g glucose equivalent (2) and was reduced by 54% after accounting for differences in available carbohydrate intake univariately and by 93% after further accounting for GI
TABLE 6 Difference in glycemic load (GL) in intervention studies related to difference in glycemic index (GI) achieved, for each category of food intake1 All studies combined k (df) Slope (g/GI%glu) SEE P kh-t 3P Constant (g/d) SEE P kh-t 3P I2 (g2/g2) Tau (g/d) P Q P X 37 (35) 3.1 0.4 0.001 10.2 6.4 0.122 0.67 11 0.001 0.001 Controlled intake 14 (12) 3.2 0.6 0.001 0.001 13.2 12 0.298 0.894 0.24 10 0.20 0.23 Limited control 19 (17) 2.9 0.50 0.001 0.001 7 9.3 0.439 1.317 0.78 12 0.001 0.001 Ad libitum 5 (3) 4.5 1.3 0.046 0.138 21.9 16.1 0.268 0.804 0.31 16 0.31 0.26

FIGURE 8. Difference in available carbohydrate intake associated with unavailable carbohydrate intake found.

1 Model: difference in glycemic load slope (SEE) difference in GI constant (SEE) Tau SE. k, number of studies; df, degrees of freedom; P kh-t statistical significance based on knapphartung t; Tau, standard error among studies; I2, variation among studies (Tau2) as a proportion of total variation (Tau2 SE2). P Q, probability of significant heterogeneity; P X, likelihood-ratio test of probability that Tau is 0.

234S

LIVESEY ET AL

these intervention studies as evident in univariate analysis of all studies combined (Table 7). This comprised elevations of energy and available carbohydrate intakes (ie, significant positive metaregression constants) followed by a slope falling toward lower GLs. Unavailable carbohydrate intake and fat intake each differ little or nonsignificantly, because GL reduces with GI univariately, with a marginal exception for fat in the controlled food intake category.
DISCUSSION

FIGURE 10. Contribution of difference in glycemic index (GI; E) and difference in available carbohydrate (AC; F) intake to the difference in glycemic load (GL) achieved in the intervention studies. Data are (F, slope1 AC) and (E, slope2 GI) from the model GL slope1 AC slope2 GI constant Tau SE. g eq, glucose equivalents.

bivariately. The overall effect on GL was additive, although not simply additive, because small reductions in GL due to reduced GI were accompanied (confounded) by the intake of more available carbohydrate, which would elevate GL (Figure 10). Such confounding may be either circumstantial or a physiologic response to low GI. Reductions in metabolizable energy and available carbohydrate intake associated with the reductions in GL achieved in
TABLE 7 Difference in metabolizable energy, protein, available carbohydrates, fat, and unavailable carbohydrate intakes in intervention studies related to difference in glycemic load (GL) achieved, for each category of food intake1 All studies Limited Ad combined Controlled controlled libitum Metabolizable energy k Slope (kJ/g eq) Constant (kJ/d) Protein k Slope (g/g eq) Constant (g/d) Available carbohydrates k Slope (g/g eq) Constant (g/d) Fat k Slope (g/g eq) Constant (g/d) Unavailable carbohydrates k Slope (g/g eq) Constant (g/d) P kh-t 0.01. P kh-t 0.05. 3 P kh-t 0.001.
1 2

A reduction in GI achieved by dietary intervention is commonly accompanied by changes in available carbohydrate intake. The resulting change in dietary GL is therefore not solely the result of a substitution of higher GI carbohydrates by lower GI carbohydrates. This is reflected in the fact that variance in GL is explained almost equally by the variance in available carbohydrate intake that accompanies variance in GI (Figure 10). Also, available carbohydrate intake varies by g/g exchange of available and unavailable carbohydrate. In large part, the wider range of GL observed than expected from the change in GI alone is explained by elevated intakes of available carbohydrate for low reductions in GI followed by a trend toward progressive reduction in the available carbohydrate intake with progressive reduction in GI. Reduction in the intake of available carbohydrate, and with it metabolizable energy, occurs when GI is reduced by 10 GI units on average or for GL 30 g glucose equivalent (Table 8). Overall, such reduction in metabolizable energy is contributed to also by reductions in protein intake. There is no evidence even when relaxing control over food intake and composition that aiming for a low-GI diet results in a rise in energy from fat intake to compensate for the fall in energy from available carbohydrate. The generality of an inverse relation between available carbohydrate and fat intake is therefore broken in this instance.
TABLE 8 Reduction in glycemic index and load required before metabolizable energy and carbohydrate intake are reduced1 x Glycemic index (% glucose) Metabolizable energy Limited control Ad libitum Available carbohydrate Limited control Ad libitum Glycemic load (g glucose equivalents) Metabolizable energy Limited control Ad libitum Available carbohydrate Limited control Ad libitum SE P kh-t Upper 95% CI

Downloaded from www.ajcn.org by on February 3, 2010

38 8.01 2452 37 0.07 6.21 38 0.543 213 38 0.02 3.3 35 0.06 4.1

14 0.1 67 14 0.04 0.1 14 0.281 151 14 0.092 6.2 13 0.07 1.3

19 15.21 4722 18 0.2 141 19 0.803 283 19 0.01 2.3 17 0.07 5.9

5 23.42 420 5 0.24 4.9 5 0.862 222 5 0.17 0.8 5 0.04 3.4

10 9 13 10

6 2 4 4

0.140 0.002 0.005 0.061

24 15 23 22

31 18 34 26

6 6 5 7

0.001 0.049 0.001 0.023

44 36 45 46

1 For glycemic index, data are intercepts on the x axis in Figures 1 and 3, and for glycemic load they are the corresponding figures for similar analyses summarized in Table 7. Data are not shown for protein, which yielded less consistent information, or for control treatments. Values were calculated by using nonlinear combination of regression coefficients (ratio: constant to slope).

FOOD, GLYCEMIA, AND HEALTH: DATABASE AND DIETS

235S

The present findings derive from random effects inverse variance meta-analysis and REML meta-regression across studies. Publication bias is not excluded among the present studies, though no analyses were indicative. However, this is strictly assessable only when fixed-effects analysis is applicable. The observed effects apply to combined health and body weight types (healthy, impaired glucose tolerance, hyperlipidemia, type 1 diabetes, type 2 diabetes, primary and secondary CHD risk, and normal, overweight, and obese weight for height). Generally, there is too little information of a consistent nature or just too little information to establish the present findings in any one health or body weight type separately or for interventions of 12 wk duration. Because a rise in available carbohydrate intake with small reductions in GI occurred in all 3 food intake control categories (ad libitum, limited controlled, and controlled), it may prove difficult to avoid when using GI alone as an intervention tool. Should this be so, a revised approach to interventions for the control of the postprandial glycemic response could be needed. This is particularly so wherever small reductions in GI can be expected. Likewise, it may be that maximum reduction in GL through contemporary strategies of GI reduction might not optimally reach the intended goal. One consideration would be a more direct attempt that reduces the GL of diets while controlling fat intake. Before doing so, evidence relating GL to health is necessary and is addressed in the follow-up paper.
The authors are grateful to G Janecek, University of East Anglia, Norwich, UK, for preliminary discussion of the statistical approach. The contributions of the authors were as followsGL and RT: data collection; GL: analysis; GL and JH: writing; and TH: comment. GL and RT had no financial or personal conflicts of interest. JH is currently advising an industry group comprising food manufacturers and retailers who are preparing a submission to the authorities in Europe supporting the case for the use of GI in the labeling of food products. TH works for Kellogg.

REFERENCES
1. Brand JC, Colagiuri S, Crossman S, Allen A, Roberts DC, Truswell AS. Low-glycemic index foods improve long-term glycemic control in NIDDM. Diabetes Care 1991;14:95101. 2. Salmeron J, Ascherio A, Rimm EB, et al. Dietary fiber, glycemic load, and risk of NIDDM in men. Diabetes Care 1997;20:54550. 3. Liu S, Willett WC, Stampfer MJ, et al. A prospective study of dietary glycemic load, carbohydrate intake, and risk of coronary heart disease in US women. Am J Clin Nutr 2000;71:1455 61. 4. Kelly S, Frost G, Whittaker V, Summerbell C. Low glycaemic index diets for coronary heart disease. Cochrane Database Syst Rev 2004: CD004467. 5. Oh K, Hu FB, Cho E, et al. Carbohydrate intake, glycemic index, glycemic load, and dietary fiber in relation to risk of stroke in women. Am J Epidemiol 2005;161:1619. 6. Franceschi S, Dal Maso L, Augustin L, et al. Dietary glycemic load and colorectal cancer risk. Ann Oncol 2001;12:173 8. 7. Giovannucci E. Insulin, insulin-like growth factors and colon cancer: a review of the evidence. J Nutr 2001;131:3109S20S. 8. McCance RA, Lawrence RD. The carbohydrate content of foods. Medical Research Council special report series no. 135. London, United Kingdom: Her Majestys Stationery Office, 1929. 9. Jenkins DJ, Wolever TM, Taylor RH, et al. Glycemic index of foods: a physiological basis for carbohydrate exchange. Am J Clin Nutr 1981; 34:362 6. 10. Jenkins DJ, Axelsen M, Kendall CW, Augustin LS, Vuksan V, Smith U. Dietary fibre, lente carbohydrates and the insulin-resistant diseases. Br J Nutr 2000;83(suppl):S157 63. 11. Bjorck I, Elmstahl HL. The glycaemic index: importance of dietary fibre and other food properties. Proc Nutr Soc 2003;62:201 6.

12. Livesey G. Health potential of polyols as sugar replacers, with emphasis on low glycaemic properties. Nutr Res Rev 2003;16:16391. 13. Robertson MD, Livesey G, Hampton SM, Mathers JC. Evidence for altered control of glucose disposal after total colectomy. Br J Nutr 2000; 84:8139. 14. Hofman Z, van Drunen JD, de Later C, Kuipers H. The effect of different nutritional feeds on the postprandial glucose response in healthy volunteers and patients with type II diabetes. Eur J Clin Nutr 2004;58:1553 6. 15. Brennan CS. Dietary fibre, glycaemic response, and diabetes. Mol Nutr Food Res 2005;49:560 70. 16. Livesey G. Non-digestible carbohydrates and glycaemic control. In: Vorster H, Blaauw R, Dhansay M, Kuzwayo P, Moend L, WentzelViljoen E, eds. Nutrition Safari for Inovative Solutions, Proceedings of the 18th International Congress of Nutrition. Durban, South Africa: ICC, 2005. 17. Robertson MD, Bickerton AS, Dennis AL, Vidal H, Frayn KN. Insulinsensitizing effects of dietary resistant starch and effects on skeletal muscle and adipose tissue metabolism. Am J Clin Nutr 2005;82:559 67. 18. Ostman EM, Frid AH, Groop LC, Bjorck IM. A dietary exchange of common bread for tailored bread of low glycaemic index and rich in dietary fibre improved insulin economy in young women with impaired glucose tolerance. Eur J Clin Nutr 2006;60:334 41. 19. Livesey G. Glycaemic responses and toleration. In: Mitchell H, ed. Sweeteners and sugar alternatives in food technology. Oxford, United Kingdom: Blackwell, 2006:118. 20. Brand-Miller JC, Petocz P, Colagiuri S. Meta-analysis of low-glycemic index diets in the management of diabetes: response to Franz. Diabetes Care 2003;26:3363 4, author reply 3364 5. 21. Opperman AM, Venter CS, Oosthuizen W, Thompson RL, Vorster HH. Meta-analysis of the health effects of using the glycaemic index in meal-planning. Br J Nutr 2004;92:367 81. 22. Anderson JW, Randles KM, Kendall CW, Jenkins DJ. Carbohydrate and fiber recommendations for individuals with diabetes: a quantitative assessment and meta-analysis of the evidence. J Am Coll Nutr 2004;23: 517. 23. Raben A. Should obese patients be counselled to follow a low-glycaemic index diet? No. Obes Rev 2002;3:24556. 24. Higgins JPT, Green S, eds. Cochrane handbook for systematic reviews of interventions 4.1.6. Chichester, United Kingdom: Wiley & Sons, Ltd, 2003. 25. Deeks J, Higgins JDA. Section 8: analysis and presentation of results. Cochrane Reviewers Handbook 2002;4.2.2. 26. Kiens B, Richter EA. Types of carbohydrate in an ordinary diet affect insulin action and muscle substrates in humans. Am J Clin Nutr 1996; 63:4753. 27. Kabir M, Oppert JM, Vidal H, et al. Four-week low-glycemic index breakfast with a modest amount of soluble fibers in type 2 diabetic men. Metabolism 2002;51:819 26. 28. Brynes AE, Mark Edwards C, Ghatei MA, et al. A randomised fourintervention crossover study investigating the effect of carbohydrates on daytime profiles of insulin, glucose, non-esterified fatty acids and triacylglycerols in middle-aged men. Br J Nutr 2003;89:20718. 29. Giacco R, Parillo M, Rivellese AA, et al. Long-term dietary treatment with increased amounts of fiber-rich low-glycemic index natural foods improves blood glucose control and reduces the number of hypoglycemic events in type 1 diabetic patients. Diabetes Care 2000;23:1461 6. 30. Frost G, Leeds A, Trew G, Margara R, Dornhorst A. Insulin sensitivity in women at risk of coronary heart disease and the effect of a low glycemic diet. Metabolism 1998;47:124551. 31. Herrmann TS, Bean ML, Black TM, Wang P, Coleman RA. High glycemic index carbohydrate diet alters the diurnal rhythm of leptin but not insulin concentrations. Exp Biol Med (Maywood) 2001;226:1037 44. 32. Jenkins DJ, Wolever TM, Collier GR, et al. Metabolic effects of a low-glycemic-index diet. Am J Clin Nutr 1987;46:968 75. 33. Kurup PG, Krishnamurthy S. Glycemic response and lipemic index of rice, raggi and tapioca as compared to wheat diet in human. Indian J Exp Biol 1993;31:2913. 34. Bouche C, Rizkalla SW, Luo J, et al. Five-week, low-glycemic index diet decreases total fat mass and improves plasma lipid profile in moderately overweight nondiabetic men. Diabetes Care 2002;25:822 8. 35. Dumesnil JG, Turgeon J, Tremblay A, et al. Effect of a low-glycaemic indexlow-fat high protein diet on the atherogenic metabolic risk profile of abdominally obese men. Br J Nutr 2001;86:557 68. 36. Pereira MA, Swain J, Goldfine AB, Rifai N, Ludwig DS. Effects of a

Downloaded from www.ajcn.org by on February 3, 2010

236S

LIVESEY ET AL
index improves dietary profile and metabolic control in type 2 diabetic patients. Diabet Med 1994;11:397 401. Jarvi AE, Karlstrom BE, Granfeldt YE, Bjorck IE, Asp NG, Vessby BO. Improved glycemic control and lipid profile and normalized fibrinolytic activity on a low-glycemic index diet in type 2 diabetic patients. Diabetes Care 1999;22:10 8. Jimnez-Cruz A, Bacardi-Gascon M, Turnbull WH, Rosales-Garay P, Severino-Lugo I. A flexible, low-glycemic index Mexican-style diet in overweight and obese subjects with type 2 diabetes improves metabolic parameters during a 6-week treatment period. Diabetes Care 2003;26: 196770. Komindr S, Ingsriswang S, Lerdvuthisopon N, Boontawee A. Effect of long-term intake of Asian food with different glycemic indices on diabetic control and protein conservation in type 2 diabetic patients. J Med Assoc Thai 2001;84:8597. Tsihlias EB, Gibbs AL, McBurney MI, Wolever TM. Comparison of high- and low-glycemic-index breakfast cereals with monounsaturated fat in the long-term dietary management of type 2 diabetes. Am J Clin Nutr 2000;72:439 49. Wolever TM, Jenkins DJ, Vuksan V, et al. Beneficial effect of a low glycaemic index diet in type 2 diabetes. Diabet Med 1992;9:451 8. Heilbronn LK, Noakes M, Clifton PM. The effect of high- and lowglycemic index energy restricted diets on plasma lipid and glucose profiles in type 2 diabetic subjects with varying glycemic control. J Am Coll Nutr 2002;21:120 7. Jimnez-Cruz A, Turnbull WH, Bacardi-Gasco n M, Rosales-Garay P. A high-fiber, moderate-glycemic-index, Mexican style diet improves dyslipidemia in individuals with type 2 diabetes Nutr Res 2004;24:19-27. Luscombe ND, Noakes M, Clifton PM. Diets high and low in glycemic index versus high monounsaturated fat diets: effects on glucose and lipid metabolism in NIDDM. Eur J Clin Nutr 1999;53:473 8. Rizkalla SW, Bellisle F, Slama G. Health benefits of low glycaemic index foods, such as pulses, in diabetic patients and healthy individuals. Br J Nutr 2002;88(suppl):S255 62. Wolever TM, Jenkins DJ, Vuksan V, Jenkins AL, Wong GS, Josse RG. Beneficial effect of low-glycemic index diet in overweight NIDDM subjects. Diabetes Care 1992;15:562 4. Zhang Y-F, Yang Y-X, Ma Z-L, Han J-H, Wang Z. The research on application in the education with glycemic index in diabetic patients. Acta Nutrimenta Sinica 2003;25:248 52. Jenkins DJ, Wolever TM, Kalmusky J, et al. Low-glycemic index diet in hyperlipidemia: use of traditional starchy foods. Am J Clin Nutr 1987; 46:66 71. Frost G, Keogh B, Smith D, Akinsanya K, Leeds A. The effect of low-glycemic carbohydrate on insulin and glucose response in vivo and in vitro in patients with coronary heart disease. Metabolism 1996;45: 669 72. Frost GS, Brynes AE, Bovill-Taylor C, Dornhorst A. A prospective randomised trial to determine the efficacy of a low glycaemic index diet given in addition to healthy eating and weight loss advice in patients with coronary heart disease. Eur J Clin Nutr 2004;58:1217.

37.

38.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

50. 51.

low-glycemic load diet on resting energy expenditure and heart disease risk factors during weight loss. JAMA 2004;292:248290. Sloth B, Krog-Mikkelsen I, Flint A, et al. No difference in body weight decrease between a low-glycemic-index and a high-glycemic-index diet but reduced LDL cholesterol after 10-wk ad libitum intake of the lowglycemic-index diet. Am J Clin Nutr 2004;80:337 47. Agus MS, Swain JF, Larson CL, Eckert EA, Ludwig DS. Dietary composition and physiologic adaptations to energy restriction. Am J Clin Nutr 2000;71:9017. Carels RA, Darby LA, Douglass OM, Cacciapaglia HM, Rydin S. Education on the glycemic index of foods fails to improve treatment outcomes in a behavioral weight loss program. Eat Behav 2005;6:14550. Ebbeling CB, Leidig MM, Sinclair KB, Hangen JP, Ludwig DS. A reduced-glycemic load diet in the treatment of adolescent obesity. Arch Pediatr Adol Med 2003;157:7739. Spieth LE, Harnish JD, Lenders CM, et al. A low-glycemic index diet in the treatment of pediatric obesity. Arch Pediatr Adol Med 2000;154: 94751. Wolever TM, Mehling C. High-carbohydrate-low-glycaemic index dietary advice improves glucose disposition index in subjects with impaired glucose tolerance. Br J Nutr 2002;87:477 87. Slabber M, Barnard HC, Kuyl JM, Dannhauser A, Schall R. Effects of a low-insulin-response, energy-restricted diet on weight loss and plasma insulin concentrations in hyperinsulinemic obese females. Am J Clin Nutr 1994;60:48 53. Calle-Pascual AL, Gomez V, Leon E, Bordiu E. Foods with a low glycemic index do not improve glycemic control of both type 1 and type 2 diabetic patients after one month of therapy. Diabetes Metab 1988;14: 629 33. Collier GR, Giudici S, Kalmusky J, et al. Low glycaemic index starchy foods improve glucose control and lower serum cholesterol in diabetic children. Diab Nutr Metab 1988;1:1119. Fontvieille AM, Acosta M, Rizkalla SW, et al. A moderate switch from high to low glycaemic-index foods for three weeks improves the metabolic control of Type 1 (IDDM) diabetic subjects. Diab Nutr Metab 1988;1:139 43. Fontvieille AM, Rizkalla SW, Penfornis A, Acosta M, Bornet FR, Slama G. The use of low glycaemic index foods improves metabolic control of diabetic patients over five weeks. Diabet Med 1992;9:444 50. Gilbertson HR, Brand-Miller JC, Thorburn AW, Evans S, Chondros P, Werther GA. The effect of flexible low glycemic index dietary advice versus measured carbohydrate exchange diets on glycemic control in children with type 1 diabetes. Diabetes Care 2001;24:1137 43. Lafrance L, Rabasa-Lhoret R, Poisson D, Ducros F, Chiasson JL. Effects of different glycaemic index foods and dietary fibre intake on glycaemic control in type 1 diabetic patients on intensive insulin therapy. Diabet Med 1998;15:972 8. Jenkins DJ, Wolever TM, Buckley G, et al. Low-glycemic-index starchy foods in the diabetic diet. Am J Clin Nutr 1988;48:248 54. Frost G, Wilding J, Beecham J. Dietary advice based on the glycaemic

52.

53.

54.

55.

56. 57.

58. 59. 60. 61. 62. 63. 64.

Downloaded from www.ajcn.org by on February 3, 2010

65.