You are on page 1of 14

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 Among early scientific enquiry is that showing that unavail-


Background: Reduction of dietary glycemic response has been able carbohydrate does not adversely elevate blood glucose con-
proposed as a means of reducing the risk of diabetes and coronary centrations and so is a useful nutrient source for persons with
heart disease. Its role in health maintenance and management, along- diabetes; available carbohydrate was seen as being adverse or not
side unavailable carbohydrate (eg, fiber), is incompletely under- well tolerated (8). That some available carbohydrate might also
stood. be suitable as an energy source for persons with diabetes became

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


Objective: We aimed to assess the evidence relating the glycemic evident later; this was characterized as low-GI available carbo-
impact of foods to a role in health maintenance and management of hydrate (9). The idea that unavailable carbohydrate might have a
disease. direct or indirect role in glycemic control has renewed interest,
Design: We searched the literature for relevant controlled dietary with 8 possible mechanisms proposed (10 18). Most recently,
intervention trials on glycemic index (GI) according to inclusion and a short narrative review of replacement of ingredient available
exclusion criteria, extracted the data to a database, and synthesized carbohydrate with ingredient unavailable carbohydrate sug-
the evidence via meta-analyses and meta-regression models. gested reductions in fasting blood glucose or glycated proteins in
Results: Among literature to January 2005, 45 relevant publications persons with diabetes but not in persons in whom fasting blood
were identified involving 972 subjects with good health or metabolic glucose was not raised (19). Among all these reports is consid-
disease. With small reductions in GI (10 units), increases in avail- eration that for optimal glycemic control, unavailable carbohy-
able carbohydrate, energy, and protein intakes were found in all drate might be used alongside low-GI available carbohydrate to
studies combined. Falling trends in energy, available carbohydrate, limit the amount of high-glycemic carbohydrate among food
and protein intakes then occurred with progressive reductions in GI. choices. However, the relative importance of unavailable and
Fat intake was essentially unchanged. Unavailable carbohydrate low-glycemic available carbohydrate in health promotion and
intake was generally higher for intervention diets but showed no management is unknown.
trend with GI (falling or rising). Among studies reporting on GI, The objectives of the present study were to construct a data-
variation in glycemic load was approximately equally explained by base of randomized controlled (or similar) intervention studies
variation in GI and variation in available carbohydrate intake. An that could be used to address queries about the possible role of GL
exchange of available and unavailable carbohydrate (1 g/g) was and indexes of GL, as modifiable by available and unavailable
evident in these studies. carbohydrate, in the management of health and prevention of
Conclusions: Among GI studies, observed reductions in glycemic disease in respect of common metabolic conditions. In this, the
load are most often not solely due to substitution of high for low first of 2 articles in this issue of the Journal, the database is
glycemic carbohydrate foods. Available carbohydrate intake is a
confounding factor. The role of unavailable carbohydrate remains to 1
From Independent Nutrition Logic, Wymondham, Norfolk, United
be accounted for. Am J Clin Nutr 2008;87(suppl):223S36S. Kingdom (GL and RT); Kellogg Europe, Den Bosch, Netherlands (TH); and
Wembley Park, Middlesex, United Kingdom (JH).
KEY WORDS Carbohydrate, glycemic response, glycemic in- 2
Presented at an ILSI Europe workshop titled Glycemic Response and
dex, glycemic load, meta-analysis, systematic review 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
INTRODUCTION Europe) and was funded by industry members Cerestar, Coca-Cola, Danisco,
Reduction of the glycemic response to foods, via either re- Groupe Danone, Kellogg, Kraft Foods, National Starch, Nestl, RHM Tech-
nology, Royal Cosun, Sudzucker, Tate & Lyle Specialty Sweeteners, and
duced glycemic index (GI) (1) or reduced glycemic load (GL) (2)
Unilever. The opinions expressed herein are those of the authors and do not
has been proposed as a dietary means to help to combat diabetes
necessarily represent the view of ILSI or ILSI Europe.
mellitus and possibly coronary heart disease (CHD) (3, 4). Ex- 4
Address reprint requests to ILSI Europe. E-mail: publications@
cessive glycemic response to carbohydrate foods and low un- ilsieurope.be.
available carbohydrate intake are also implicated in stroke (5) 5
Address correspondence to G Livesey, Independent Nutrition Logic,
and certain forms of cancer, in particular colorectal cancer, in Pealerswell House, 21 Bellrope Lane, Wymondham, Norfolk, NR18 0QX
some groups (6, 7). United Kingdom. E-mail: glivesey@inlogic.co.uk.

Am J Clin Nutr 2008;87(suppl):223S36S. Printed in USA. 2008 American Society for Nutrition 223S
224S LIVESEY ET AL

described together with an assessment of the extent to which the TABLE 2


studies achieved reductions in GI without modifying the intake of Exclusion criteria used when screening titles and abstracts
other macronutrient energy sources. Use of the database to ex- 1) Epidemiologic studies
amine the evidence on the relations between glycemic response 2) Interventions to assess effect on any form of cancer
to food and specific aspects of health is reported separately in the 3) Interventions in animal studies
second article. 4) Interventions with a historical design
5) Studies clearly without a control treatment
6) Treatments targeting the use of unavailable carbohydrate, protein, or
MATERIALS AND METHODS fat in place of total or available carbohydrate in the absence of a
primary targeting of a reduction in glycemic index of available
The database was constructed by using 3 processes applied
carbohydrate
sequentially: 1) a literature search; 2) examination of titles, ab- 7) Treatments reducing the glycemic index of the diet by using
stracts, and full articles; and 3) database construction, data anal- inhibitors of carbohydrate digestion
ysis, and synthesis of evidence. The electronic databases 8) Change in glycemic index or load as part of a multiple intervention
searched included the Cumulative Index to Nursing & Allied not reflected in the control treatment, such as with drugs or lifestyle
Health Literature (CINAHL; Internet: www.cinahl.com) from factors
1982 to January 2005, the Cochrane Central Register of Con- 9) Interventions specifically to study satiety, which had no relevant
trolled Trials (CENTRAL; Internet: www.mrw.interscience. outcomes
wiley.com/cochrane/cochrane_clcentral_articles_fs.html) to Janu- 10) Publications related to food product characteristics, including
determinations of glycemic index or load
ary2005,theElsevierMedicalDatabase(EMBASE;Internet:www.
11) Methodologic publications
embase.com) from 1980 to December 2003, the US National Li-
12) Publications relating to exercise performance

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


brary of Medicine database (MEDLINE via the PubMed portal; 13) Publications on guidelines, reviews, commentaries, and abstracts
Internet: www.ncbi.nlm.nih.gov.80/sites/entrez/) from 1950 to Jan- 14) Study proposals, commentaries, and reviews
uary 2005; Elseviers Science-Specific Search Engine on the Inter-
net (SCIRUS; Internet: scirus.com) to January 2005, and Black-
wells Nutrition and Food Science database underpinned by the inconsistencies remained, agreement was reached by joint con-
Commonwealth Agricultural Bureau International (CABI; Internet: sultation of the publication (or report) of the study in question.
www.nutritionandfoodsciences.org) from 1981 to January 2005. The agreed data were imported and stored in a StataCorp (Col-
All field searches for glyc(a)emic index or glyc(a)emic indi- lege Station, TX) STATA SE 9.0 database file (*.dta) for query
ces or glyc(a)emic load alone and each together with diabetes and analysis. Conversion of data reported in common units to SI
were performed. Previous meta-analyses (4, 20 22) and reviews units was made by GL and RT independently (Table 3) when
(23) were also examined to find citations and enabled a relative they extracted the data.
assessment of search success. Records were imported into a biblio- Subject groups were classified according to the following
graphic database (Endnote 7, published 2003; ISI Research Soft, health types: those in good health, those with impaired glucose
Berkeley, CA) to exclude duplicates automatically, with further tolerance, those with type 1 diabetes mellitus, those with type 2
screening manually to exclude remaining duplicates. diabetes mellitus, and those with CHD. Subject groups were also
Potentially relevant studies were identified by screening titles classified according to body weight as normal, overweight, or
and abstracts by using the inclusion and exclusion criteria given obese according to body mass index (cutoffs in kg/m2 of 25 or
in Tables 1 and 2, respectively. Full articles identified as being 30) or the original author information.
potentially relevant were subjected to detailed examination Biochemical risk factors extracted were fasting blood glucose,
against these criteria and were included only if the information fasting insulin, glycated proteins (HbA1c and fructosamine), in-
could be extracted and the articles were from English language sulin sensitivity, retrospectively calculated insulin sensitivity by
publications, from a foreign language publication with English homeostatic model assessment (HOMA %S), pancreatic B-cell
abstract for which additional information could be sought from function calculated retrospectively by homeostatic model assess-
the authors, or from unpublished work available by that time. ment (HOMA %B), cholesterol (total, LDL, and HDL), and
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 TABLE 3
Conversion factors
provisional databases by RT and GL independently. Inequalities
in cell contents between the provisional databases were exam- Variable Conversion factor
ined by RT and GL separately and corrections made. Where
Glycemic index % glucose 1.43 % bread
Fasting glucose 1 mg/dL 18.02 mmol/L
Fasting insulin 1 U/mL 0.143 pmol/mL
TABLE 1
Serum triglycerides (triacylglycerols) 1 mg/dL 88.5 mmol/L
Inclusion criteria used when screening titles and abstracts
Fasting capillary blood glucose 0.61 (0.94 fasting venous
1) Publications in which glycemic index or load referred to foods or diets glucose)
and not to a diabetological parameter Metabolizable energy1
2) Controlled intervention studies Available carbohydrate 16.7 kJ/g
3) Duration of treatment 1 wk Unavailable carbohydrate 8 kJ/g
4) Studies in healthy persons, persons with either impaired glucose Protein 16.7 kJ/g
tolerance or hyperlipidemia or at risk of coronary artery disease or Fat 37.7 kJ/g
having diagnosed type 1 or 2 diabetes 1
kcal 4.184 kJ.
FOOD, GLYCEMIA, AND HEALTH: DATABASE AND DIETS 225S
plasma triacylglycerols. Body weight was the only constitutional crossover designs. Similarly, SEDs between treatments took on
risk factor extracted. Dietary risk or nutritional factors extracted methods preferences, SED1SED2SED3. SED1 was the re-
were metabolizable energy, fat, available carbohydrate, unavail- ported value. SED2 was calculated from change score SEs (com-
able carbohydrate, and protein intake, together with the potential bined treatment SED across time). SED3 was estimated from
risk factor GI with calculation of GL. Duration of treatment was pooled treatment end mean and SD values, the number of par-
extracted either as a continuous variable (weeks) or as a categor- ticipants, and rp. Preferences for SDs for start and end mean
ical variable or 12 wk. For the glycated proteins, treatment values were SD1SD2SD3. SD1 was as originally reported,
duration was often low relative to half-life in blood, and so these whereas SD2 was calculated from dependent SE of means and
were considered with and without adjustment for half-life and the number of observations. SD3 was imputed by using first or
duration of treatment. second order regressions relating SD1 to the corresponding mean
Random effects meta-analyses and meta-regressions, each for those studies imparting this information. The preferences
weighted by inverse variance, were undertaken by using meta were elaborated to maximize precision and minimize bias for
and metareg in STATA 9.2 SE (StataCorp) according to Co- combined study treatment effects and was essential to retaining
chrane guidelines (25). Meta provides combined means with the maximum number of studies in the analysis. Doing so
95% CIs, the Der Simonian and Laird estimate of between- avoided bias from dropping studies with incompletely reported
studies variance (DSL), Q-test for heterogeneity, and an asymp- statistics but included bias due to imprecision of SD3, and so
totic z test for the null hypothesis that the true effect is zero. SED3.
Metareg was (unless stated otherwise) executed by using re- Meta-analytic methods make much use of graphic material. In
stricted maximum likelihood (REML) for the fitting of coeffi- presenting the results, we give both figures and related data in
cients, Knapp & Hartungs coefficient SE for t test of signifi- tables because this is preferred to ensure provision of both per-

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


cance, likelihood-ratio estimates of between-studies variance spective and precision for the outcomes (24, 25). Where further
(Tau2), a likelihood-ratio test for heterogeneity, and an expres- analysis is made to test the significance of a particular perspec-
sion of the proportion of total variation due to heterogeneity (I2). tive, this entailed the provision of further figures and associated
Optionally, with regression models including a constant, the outcomes.
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 RESULTS
moments). Outliers were identified during sensitivity analysis by
means of the statistic 1Bij, ie, the difference between the The database and study characteristics
coefficient with and without the outlier divided by the SE with- The search for literature to January 2005 yielded 2782 poten-
out. Mean differences between dietary treatments were ex- tially relevant publication records. After the titles and abstracts
pressed either in absolute units when the metric was common to were screened and the full articles reviewed, 45 publications
all studies or as a percentage of the average of the mean dietary were identified as being relevant and of suitable quality for in-
treatments when the metrics differed among studies. clusion in the present database. Because some publications re-
Parallel and crossover studies reporting no crossover effects ported more than one study, because of repeat observations in
were combined. Studies generally reported either treatment dif- some studies (across the duration of treatment), and because
ferences in end measurements or differences in change with time some outcomes were assessed by more than one method, the 45
(change score) for each treatment or both. When an SE difference publications yielded 80 conditions contrasting the reported high
(SED) between treatments in end measures or change in mea- versus low GI diets (Table 4). All included studies had control
sures in time was not reported by the original authors, it was versus treatment comparisons and designs that were analyzable
calculated from either CIs or P values, error df, the number of as either crossover (20 studies) or parallel (25 studies). Informa-
multiple treatment comparisons reported, and the method for tion was monitored at different times within each study (see
testing the significance (Students paired t test, Tukeys test, etc). Table 4). This was used to help address questions related to the
Studies often did not report either the dependent (unpaired) or the persistence, gain, or loss of effect with time during treatment.
independent (paired) SED between treatments. These were Unless stated otherwise, independence of study observations was
therefore calculated from the dependent SED at the end of the preserved by including only the end of treatment results.
period of treatment and coefficients of correlation (rp) between The database included 972 participants per treatment arm with
dependent and independent values that were imputed from other group mean ages of 10 to 63 y, with both males (n 511) and
studies, with the use of duration-dependent values of rp when females (n 461) represented. Participants were either normal
found appropriate. weight (16 studies), overweight (18 studies), or obese (10 stud-
Calculation of differences in each study (treatment mean ies) or were unclassified by weight (1 study). Similar numbers of
control mean and independent SED for treatment effects) was participants took part in each treatment arm (770 for the high-GI
possible by several ways depending on the information available. treatment and 793 for the low-GI treatment). Study participants
The accuracy of a method and the availability of information for were either healthy (ie, no diagnosis of disease was evident; 13
the method dictated preferences. For mean treatment difference studies) or had impaired glucose tolerance (2 studies, duration of
values, the preference was m1m2m3m4. m1 was the re- impairment unknown), had type 1 diabetes (7 studies, mean
ported treatment difference (corrected for starting values). m2 duration from diagnosis of 3 to 16 y, 1 unknown duration), had
was the treatment difference calculated from reported change type 2 diabetes (17 studies, mean duration from diagnosis of 0
scores. m3 was the same as m2 but first calculating the change to 12.5 y, 7 unknown duration), were at risk of primary (4 studies,
scores from reported start and end values for each treatment. m4 duration unknown) or secondary (1 study, duration unknown)
was the difference in reported end values for each treatment in CHD, or had hyperlipidemia (1 study combining Type II a, Type
226S LIVESEY ET AL

TABLE 4
Study characteristics

Duration Body No. of No. of Duration


of weight Percentage Randomization Study participants participants of
Sources (reference)1 diagnosis band2 Age male reported3 design4 on high GI on low GI treatment

y y % n n w
Normal healthy
Frost et al 1998 (30) (n is 2) n 36 0 r p 6 6 3.0
Frost et al (1998) (30) (n) n 36 0 r p 6 6 3.0
Herrmann et al 2001 (31) (20F) n 27 56 r x 9 9 1.1
Herrmann et al 2001 (31) (30F) n 27 56 r x 9 9 1.1
Jenkins et al 1987 (32) (d1) n 33 100 r x 6 6 1.0
Jenkins et al. 1987 (32) (d2) n 33 100 r x 6 6 2.0
Kiens et al 1996 (26) (d1) n 24 100 r x 7 7 1.0
Kiens et al 1996 (26) (d2) n 24 100 r x 7 7 2.0
Kiens et al 1996 (26) (d3) n 24 100 r x 7 7 3.0
Kiens et al 1996 (26) (d4 is 2) n 24 100 r x 7 7 4.3
Kiens et al 1996 (26) (d4) n 24 100 r x 7 7 4.3
Kurup 1992 (33) (raggi vs rice) n 36 nr p10 35 40 2.1
Kurup 1992 (33) (raggi vs tapioca) n 36 nr p10 35 23 2.1
Bouche et al 2002 (34) ow 46 100 r x 11 11 5.0

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


Dumesnil 2001 (35) (ph1 vs LGI out) ow 47 100 r x 12 12 0.9
Pereira et al 2004 (36) (out) ow 31 23 r p 17 22 9.6
Price (unpublished) ow r p 22 22 12.0
Sloth et al 2004 (37) (d1) ow 30 0 r p 22 23 2.0
Sloth et al 2004 (37) (d2) ow 30 0 r p 22 23 4.0
Sloth et al 2004 (37) (d3) ow 30 0 r p 22 23 6.0
Sloth et al 2004 (37) (d4) ow 30 0 r p 22 23 8.0
Sloth et al 2004 (37) (d5) ow 30 0 r p 22 23 10.0
Agus et al 2000 (38) (out) ob 28 100 r x 10 10 0.9
Carels et al 2005 (39) ob 44 0 r p 23 21 20.0
Ebbeling et al 2003 (40) (d1) ob 16 31 r p 7 7 26.0
Ebbeling et al 2003 (40) (d2) ob 16 31 r p 7 7 52.0
Spieth et al 2000 (41) ob 10 46 nr p 43 64 18.4
Impaired glucose tolerance
Wolever & Mehling 2002 (42) (d1) ow 57 21 r p 11 13 4.0
Wolever & Mehling 2002 (42) (d2) ow 57 21 r p 11 13 8.0
Wolever & Mehling 2002 (42) (d3) ow 57 21 r p 11 13 12.0
Wolever & Mehling 2002 (42) (d4 is 2) ow 57 21 r p 11 13 16.0
Wolever & Mehling 2002 (42) (d4) ow 57 21 r p 11 13 16.0
Slabber et al 1994 (43) (C) ob 35 0 m x 16 16 12.0
Slabber et al 1994 (43) (P) ob 35 0 m p 15 15 12.0
Type 1 diabetes
Calle-Pascual 1988 (44) (T1) n 26 nr p 12 12 4.0
Collier et al 1988 (45) (d1) 3 n 12 86 r x 7 7 3.0
Collier et al 1988 (45) (d2) 3 n 12 86 r x 7 7 6.0
Fontvieille et al 1988 (46) 14.6 n 44 50 r x 8 8 3.0
Fontvieille et al 1992 (47) (mT1) 13.4 n 43 67 r x 12 12 5.0
Giacco et al 2000 (29) 10.3 n 28 37 r p 22 24 24.0
Gilbertson et al 2001 (48) 3.7 n 11 50 r p 49 55 52.0
Lafrance et al 1998 (49) 15.8 n 78 r x 9 9 1.7
Type 2 diabetes
Jenkins et al 1988 (50) 12.5 n 65 25 r x 8 8 2.0
Brand et al 1991 (1) 5 ow 62 63 r x 16 16 12.0
Calle-Pascual 1988 (44) (T2) ow 59 nr p 12 12 4.0
Fontvieille et al 1992 (47) (mT2) 7.8 ow 56 67 r x 6 6 5.0
Frost et al 1994 (51) 0.5 ow 55 71 r p 26 25 12.0
Jarvi et al 1999 (52) 8 ow 66 75 r x 20 20 3.4
Jimenez-Cruz et al 2003 (53) 8 ow 59 43 r x 14 14 6.0
Kabir et al 2002 (27) ow 59 100 r x 13 13 4.0
Komindr et al 2001 (54) ow 46 0 r x 10 10 4.0
Tsihlias et al 2000 (55) (d1) ow 62 59 r p 29 30 13.0
Tsihlias et al 2000 (55) (d2) ow 62 59 r p 29 30 26.0
Wolever et al 1992 (56) (d1) ow 67 47 r x 15 15 1.0
Wolever et al 1992 (56) (d2) ow 67 47 r x 15 15 2.0
(Continued to the right)
FOOD, GLYCEMIA, AND HEALTH: DATABASE AND DIETS 227S

TABLE 4 (Continued)

No. of meals Food Energy Available Protein Unavailable Change in Glycemic Change in
with intake intake CHO intake Fat intake intake CHO intake unavailable index Change Glycemic glycemic
treatment control5 (high GI)6 (high GI)7 (high GI)7 (high GI)7 (high GI) CHO intake (high GI)8 in GI load8 load

n/d kJ/d %E %E %E g/d g/d % % g eq/d g eq/d

0 lc 8368 48 39 13 23.8 0.0 64 13 153 31


0 lc 8368 48 39 13 23.8 0.0 64 13 153 31
3 lc 7467 65 20 15 669 22 1919 78
3 lc 7333 55 30 15 669 22 1599 64
3 c 11 506 62 20 19 57.8 13.8 74 29 317 134
3 c 11 506 62 20 19 57.8 13.8 74 29 317 134
3 c 10 309 47 41 13 7.3 11.7 64 17 186 51
3 c 10 309 47 41 13 7.3 11.7 64 17 186 51
3 c 10 309 47 41 13 7.3 11.7 64 17 186 51
3 c 10 309 47 41 13 7.3 11.7 64 17 186 51
3 c 10 309 47 41 13 7.3 11.7 64 17 186 51
1 c 8335 59 27 14 17.8 0.9 6811,12 5 20111,12 29
1 c 8335 59 27 14 17.8 3.4 6811,12 5 20111,12 8
0 lc 10 301 42 37 18 19.0 12.0 71 30 184 96

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


3 a 11 695 55 30 15 26.0 3.0 60 7 231 127
3 c 6276 65 18 17 20.0 12.0 59 23 143 85
0 lc
0 a 9800 58 21 17 36.0 1.0 6811 13 22911 55
0 a 9800 58 21 17 36.0 1.0 6811 13 22911 55
0 a 9800 58 21 17 36.0 1.0 6811 13 22911 55
0 a 9650 58 22 17 35.0 2.0 6811 13 22411 56
0 a 9650 58 22 17 35.0 2.0 6811 13 22411 56
3 c 6247 67 18 15 25.6 3.4
3 c 6941 53 32 17 57 5 125 8
3 a 6711 55 28 18 9.0 1.0 56 3 124 21
3 a 6021 55 29 18 9.5 0.5 56 3 111 0
3 a 58 28 18

3 a 7170 53 28 17 22.7 13.5 59 5 134 8


3 a 7170 53 28 17 22.7 13.5 59 5 134 8
3 a 7170 53 28 17 22.7 13.5 59 5 134 8
3 a 7170 53 28 17 22.7 13.5 59 5 134 8
3 a 7170 53 28 17 22.7 13.5 59 5 134 8
3 lc 5124 50 30 20
3 lc 5124 50 30 20

1 c 8304 60 20 20 12.9 13.7 6211,12 7 18511,12 21


0 lc 11 452 46 37 17 22.5 16.2 58 9 183 5
0 lc 11 071 47 37 16 23.5 15.5 58 9 180 2
0 lc 8862 45 36 17 33.6 6.1 60 14 144 29
3 c 7477 51 39 21 27.0 0.0 64 26 147 56
2 lc 7653 54 29 17 15.0 24.1 64 14 160 40
0 a 49 34 17 57 1
3 lc 7704 53 29 18 17.1 1.8 66 19 161 36

3 lc 5385 53 26 21 28.0 6.0 65 17 110 21


3 c 6790 46 31 19 26.0 0.0 64 9 120 20
1 c 8531 60 20 20 12.9 13.7 6211,12 7 19011,12 22
3 c 7477 51 39 21 27.0 0.0 64 26 147 56
3 lc 7531 44 32 19.6 9.8 59 4 116 5
3 c 7615 54 29 18 34.0 4.0 59 19 145 40
0 lc 6530 64 20 18 25.0 9.0 56 12 139 45
1 c 61 4
3 lc 6168 58 30 12 6811,12 18 14611,12 39
1 lc 8400 54 29 17 23.1 27.2 62 7 165 39
1 lc 8400 54 29 17 23.1 27.2 62 7 165 39
3 c 6017 59 21 20 33.4 2.4 62 19 131 41
3 c 6017 59 21 20 33.4 2.4 62 19 131 41
(Continued on next page)
228S LIVESEY ET AL

TABLE 4 (Continued)

Duration Body No. of No. of Duration


of weight Percentage Randomization Study participants participants of
Sources (reference)1 diagnosis band2 Age male reported3 design4 on high GI on low GI treatment

y y % n n w
Heilbronn 2002 (57) (d1) 5 ob 57 51 r p 21 24 4.0
Heilbronn 2002 (57) (d2) 5 ob 57 51 r p 21 24 8.0
Jimenez-Cruz et al 2004 (58) 7 ob 51 r x 8 8 3.0
Luscombe et al 1999 (59) 6.3 ob 57 67 r x 21 21 4.0
Rizkalla et al 2004 (60) ob 54 100 r x 12 12 4.0
Rizkalla et al 2004 (60) (is 2) ob 54 100 r x 12 12 4.0
Wolever et al 1992 (61) (d1) ob 63 50 r x 6 6 2.0
Wolever et al 1992 (61) (d2) ob 63 50 r x 6 6 4.0
Wolever et al 1992 (61) (d3) ob 63 50 r x 6 6 5.7
Zhang et al 2003 (62) 3 51 r p 36 36 21.0
Hyperlipidemia
Jenkins et al 1987 (63) (all) 4.7 n 53 73 nr x13 30 30 4.0
Jenkins et al 1987 (63) (fam) 4.7 n 53 73 nr x13 6 6 4.0
Jenkins et al 1987 (63) (TIIa) 3.8 n 47 33 nr x13 6 6 4.0
Jenkins et al 1987 (63) (TIIb) 6.2 n 51 86 nr x13 7 7 4.0

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


Jenkins et al 1987 (63) (TIII) 5.3 n 57 100 nr x13 1 1 4.0
Jenkins et al 1987 (63) (TIV) 4.8 n 55 81 nr x13 16 16 4.0
Coronary heart disease risk
Frost et al 1998 (30) (chdr) n 36 0 r p 8 8 3.0
Frost et al 1998 (30) (chdr is 2) n 36 0 r p 8 8 3.0
Frost et al 1996 (64) ow 63 77 r p 15 15 4.0
Frost et al 1996 (64) (is 2) ow 63 77 r p 15 15 4.0
Brynes et al 2003 (28) (st is 2) ow 45 100 r x 17 17 3.4
Brynes et al 2003 (28) (st) ow 45 100 r x 17 17 3.4
Brynes et al 2003 (28) (su is 2) ow 45 100 r x 17 17 3.4
Brynes et al 2003 (28) (su) ow 45 100 r x 17 17 3.4
Frost et al 2004 (65) ow 63 87 r p 29 26 12.0
(Continued to the right)
1
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 229S

TABLE 4 (Continued)

No. of meals Food Energy Available Protein Unavailable Change in Glycemic Change in
with intake intake CHO intake Fat intake intake CHO intake unavailable index Change Glycemic glycemic
treatment control5 (high GI)6 (high GI)7 (high GI)7 (high GI)7 (high GI) CHO intake (high GI)8 in GI load8 load

n/d kJ/d %E %E %E g/d g/d % % g eq/d g eq/d


0 c 6008 61 17 22 29.8 0.5 75 32 164 72
0 c 6008 61 17 22 29.8 0.5 75 32 164 72
3 lc 8360 54 27 18 30.0 23.0 72 12 194 46
0 lc 7569 53 21 23 30.0 0.0 63 20 151 47
2 lc 9586 38 37 20 21.0 13.0 71 32 155 77
2 lc 9586 38 37 20 21.0 13.0 71 32 155 77
0 c 5807 57 23 19 33.0 1.0 61 20 122 40
0 c 5807 57 23 19 33.0 1.0 61 20 122 40
0 c 5807 57 23 19 33.0 1.0 61 20 122 40
3 lc 59 9

3 lc 7176 49 29 19 21.8 5.9 60 8 125 18


3 lc 7176 49 29 19 21.8 5.9 60 8 125 18
3 lc 7176 49 29 19 21.8 5.9 60 8 125 18
3 lc 7176 49 29 19 21.8 5.9 60 8 125 18

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


3 lc 7176 49 29 19 21.8 5.9 60 8 125 18
3 lc 7176 49 29 19 21.8 5.9 60 8 125 18

0 lc 8786 51 37 12 29.4 2.8 62 14 166 27


0 lc 8786 51 37 12 29.4 2.8 62 14 166 27
3 lc 8544 44 36 17 18.2 0.7 65 12 145 37
3 lc 8544 44 36 17 18.2 0.7 65 12 145 37
3 a 9020 46 36 18 19.0 12.0 69 20 173 65
3 a 9020 46 36 18 19.0 12.0 69 20 173 65
3 a 9900 51 33 16 19.0 0.0 69 6 209 19
3 a 9900 51 33 16 19.0 0.0 69 6 209 19
3 lc 7360 47 32 18 21.0 6.0 58 7 120 7

II b, Type III, mean duration of diagnosis of 4.7 y). The health and dose dependency of outcomes in relation to dietary GL or
types were assigned by the authors of the original publications any of its indexes. Diets were intended to meet maintenance and
and reports. It should not be implied from the use of the health growth requirements (37 studies) or submaintenance require-
types that particular participants fit a health type uniquely. For ments (8 studies); no overfeeding studies were encountered.
example, all those with diabetes might also be considered at risk Control of food intake was categorized as ad libitum (7 studies),
of CHD. Studies included participants taking medication. Insulin limited controlled food intake (22 studies), and controlled food
dosage was reported in all 7 studies of persons with type 1 intake (16 studies).
diabetes, in 2 of 17 studies of persons with type 2 diabetes, and All studies were free-living (no studies of hospitalized patients
in 1 of 1 study of participants at risk of secondary CHD. All but or subjects housed in metabolic wards or centers of human nu-
1 study of persons with type 2 diabetes included noninsulin trition). Studies were from all continents: Asia (4 studies), Aus-
medication for glycemic control (hypoglycemic agents). tralasia (6 studies), Europe (19 studies), South America (2 stud-
Interventions were by diet, with intention to exchange the form ies), Africa (1 study), and North America (13 studies).
of available carbohydrate (high versus low GI). The extent to Mortality and morbidity scores were not collected, because
which this was achieved varied between studies. In many cases, few studies were of sufficient duration to encounter reliable
the dietary changes were accompanied by variably higher quan- responses. No data were reported on mortality, cardiovascular
tities of unavailable carbohydrate. Thus, the diets are sometimes events, or diabetic complications. Changes in the severity of risk
qualified collectively as variably lower glycemic and variably factors were the primary outcomes: fasting glucose, glycated
higher unavailable carbohydrate. Three studies were identified proteins (HbA1c, fructosamine), fasting insulin, insulin sensitiv-
statistically as outliers, ie, belonging to a population of studies ity, calculated HOMA %S, calculated HOMA %B, calculated
differing from the remainder. These were Agus et al (38), HOMA %D (the product of HOMA %B HOMA %S), fasting
Dumesnil et al (35), and Pereira et al (36) and were identified plasma or serum triacylglycerols, and body weight. Data on
because of the exceptional replacement of available carbohy- fasting total, HDL, and LDL cholesterol were extracted, ap-
drate with either protein or fat. peared to be confounded, and were not investigated further at this
Treatment durations ranged from 1 to 52 wk. Studies included time. Other outcomes were macronutrient intakes, physical ac-
interventions with food exchanges at 1 or 2 or 3 (and more) or tivity, and adverse events. Insufficient numbers of studies mon-
probably 3 (and more) meals per day (6, 2, 25, and 12 studies, itored physical activity to accumulate information in the data-
respectively). This enabled queries concerning meal numbers base; 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 there were no dropouts in 23 studies, and 1 study (30) with
had freedom to respond, ie, ad libitum plus limited controlled dropouts presented both intention-to-treat and compliant-to-diet
food intake; the other 15 had controlled, ie, fixed, food intake), information. Individual study dropouts ranged from 0% to
available carbohydrate intake (41 studies, 26 with freedom to 61% of study entrants with an unweighted combined study
respond); 2) unavailable carbohydrate intake (36 studies, 22 with mean of 8%.
freedom to respond); 3) protein intake (39 studies, 24 with free- The number of participants per treatment arm per study was
dom to respond); 4) fat intake (40 studies, 26 with freedom to generally small. There were 15 studies for which this number was
respond); 5) calculated GL (ie, diet GI times available carbohy- 10 units, 15 studies with 10 to 20, 11 studies with 20 to
drate intake/100; 38 studies, 24 with freedom to respond); 6) 30, and 4 studies with 30 to 60. Power calculations were
dietary GI (as % glucose) ostensibly of available carbohydrate seldom given, and then only for few of the measurements made.
(41 studies, 26 with freedom to respond; 6 studies did not fully Forty of the 45 studies reported macronutrient intakes and
report GI values but this was recoverable approximately from changes in intake between high to low glycemic carbohydrate
other information available (see footnotes to Table 4); 7) calcu- diets. Of these, 38 reported GI values (or information on foods
lated GI of total carbohydrate (35 studies, 22 with freedom to enabling GI to be estimated). Although many studies reported
respond) were estimated as 100 times the dietary GL divided by designs intended to have similar macronutrient intakes in the
the sum weight of available and unavailable carbohydrate. treatment and control arms, the precision with which this was
Two publications (28, 31) included more than one treatment achieved varied between studies. The possibility that difference
(versus control) per study. In one case, dietary fat intake differed in treatment outcomes covary with unintended differences in
between treatments (30); in the other, the source of carbohydrate macronutrient intakes was examined, because such differences
modifying GI differed (sucrose versus starch; 28). This informa- between treatments might confound the interpretation of a treat-

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


tion was retained in the database because it was particularly ment effect as being due to GI if not accounted for by, for ex-
useful where study results were heterogeneous. One study re- ample, covariance.
porting an effect on body weight (43) informed about the first part To investigate the relation between GI and the energy and
of a crossover study as a parallel design. In this case, analyses macronutrient intakes from the intervention diets, the studies
were performed on the crossover design, and the parallel study were either combined or were divided into 3 categories (Table
was used only for comparison of combined results by study 5). Thus, foods were either available ad libitum or were available
design. in wholly controlled amounts, or were available subject to lim-
Common to each of the 45 studies entered into the database ited (or partial) control. In the ad libitum category, food intake
was information for participants that were compliant to diet could vary but diet composition was intended to be fixed (food
(ie, dropouts and noncompliant participants discarded). Twenty- was provided or largely provided). In the wholly controlled food
one studies had reported outcomes on the basis of compliance to intake category, both food intake and composition were fixed
diet analyses, whereas 24 studies reported information that was (food was provided or largely provided and may have been ad-
both intention-to-treat and compliant-to-diet. Among these, justed in amount if body weight changed). In the intermediate,

TABLE 5
Studies included by food intake control category1

Controlled food intake Limited controlled food intake Ad libitum food intake
2
Agus et al 2000 (38) Bouche et al 2002 (34) Brynes et al 2003 (28) (su)
Brand et al 1991 (1) Collier et al 1988 (45) (d2) Brynes et al 2003 (28) (st)
Calle-Pascual 1988 (44) (exp1_T2) Fontvieille et al 1988 (46) Ebbeling et al 2003 (40) (d2)
Calle-Pascual 1988 (44) (exp2_T2) Frost et al 1994 (52) Sloth et al 2004 (37) (d5)
Carels et al 2005 (39) Frost et al 1996 (65) Wolever & Mehling 2002a (42) (d4)
Fontvieille et al 1992 (47) (exp1) Frost et al 1998 (30) (exp1)
Fontvieille et al 1992 (47) (exp2) Frost et al 1998 (30) (exp2)
Heilbronn 2002 (57) (d2) Frost et al 2004 (64)
Jarvi et al 1999 (52) Giacco et al 2000 (29)
Jenkins et al 1987a (32) (d2) Herrmann et al 2001 (31) (expt2)
Kiens et al 1996 (26) (d4) Jenkins et al 1987b (63) (all)
Kurup 1992 (33) (raggi vs rice) Jenkins et al 1988 (50)
Kurup 1992 (33) (expt2 raggi vs tapioca) Jimenez-Cruz et al 2003 (53)
Wolever et al 1992a (61) (d3) Jimenez-Cruz et al 2004 (58)
Wolever et al 1992a (61) (d2) 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)
1
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 FIGURE 3. Difference in the intake of available carbohydrate associated
glycemic index achieved. Observations are grouped by the category of con- with difference in glycemic index achieved. Observations are grouped by the
trol over food intake (where total is all categories combined). Each study is category of control over food intake (where total is all categories combined).
represented by a bubble proportional to the inverse (Tau2 SE2). Large Each study is represented by a bubble proportional to the inverse (Tau2 SE2).
bubbles indicate studies with large weight. AC, available carbohydrate; ME, Large bubbles indicate studies with large weight. AC, available carbohydrate.

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


metabolizable energy.

toward greater such effect as control over food intake is pro-


partially controlled category, scope existed for both intakes and gressively relaxed, from controlled to limited controlled to ad
composition to vary (the diet was advised or largely advised). libitum. Multivariate meta-regressions (legend to Figure 4)
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 meta-
regression (Figure 1, Figure 2, and Figure 3) indicate elevated
intakes of metabolizable energy, protein, and available carbohy-
drate 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

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 macronu-
trient 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 correspond-
FIGURE 2. Difference in protein intake associated with difference in ing values in the figures table). Univariate model: intake constant slope
glycemic index achieved. Observations are grouped by the category of con- GI Tau SE. Multivariate model: intake overall constant*
trol over food intake (where total is all categories combined). Each study is trend** (category -1control, 0limited control, 1ad-lib) slope1 GIad-lib
represented by a bubble proportional to the inverse (Tau2 SE2). Large slope2 GIlimited control slope3 GI control Tau SE. Contr, controlled;
bubbles indicate studies with small errors. AC, available carbohydrate. ltd contr., limited controlled; ad-lib, ad libitum.
232S LIVESEY ET AL

including categorical variables for controlled, limited con-


trolled, and ad libitum food intake categories support that
elevations in intake of metabolizable energy, available car-
bohydrate, 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 occur-
ring already in the controlled food intake category. Neverthe-
less, 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 multivar-
iate analysis using categorical variables for the level of food FIGURE 6. Difference in fat intake associated with difference in glyce-
intake control indicated that a fall in metabolizable energy, avail- mic index achieved. Observations are grouped by the category of control over
able carbohydrate, and protein intake occurs with progressive food intake (where total is all categories combined). Each study is repre-
reduction in GI, each being very significant (Figure 5). Further- sented by a bubble proportional to the inverse (Tau2 SE2). Large bubbles
indicate studies with large weight. AC, available carbohydrate.

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


more, 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. change (Figure 6). Nevertheless, statistically significant effects
In contrast with metabolizable energy, protein, and available are not excluded because in the control food intake category a
carbohydrate, the intake of fat appears essentially unchanged by meta-regression constant equal to 8.3 g fat/d (P kh-t
the interventions, or at best fat intake appears not so sensitive to 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 con-
firmed (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 FIGURE 7. Difference in unavailable carbohydrate intake associated
given to the trend by the permutations test (0.003, 0.12, and 0.005 in place of with difference in glycemic index achieved. Observations are grouped by the
corresponding values in the figures table). Univariate model: intake category of control over food intake (where total is all categories combined).
constant slope GI Tau SE. Multivariate model: intake Each study is represented by a bubble proportional to the inverse
constant control constantlimited control constant overall slope* GI (Tau2SE2). Large bubbles indicate studies with large weight. AC, available
trend** ( GI category -1control, 0limited control, 1ad-lib) Tau SE. 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 un-
available 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 (regres-
sion 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 libi-
tum 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 inter-
ventions; this is independent of the absolute decrease in GI.
Heterogeneity (2 - Tau2, the between-studies variance) was
highly significant among the study observations. For metaboliz- FIGURE 9. Difference in glycemic load associated with difference in
able energy, 2 was 157 reduced to 117 kJ2 after accounting for glycemic index achieved. Observations are grouped by the category of con-
the variation in both GI and category of food intake (ie, 25% of trol over food intake (where total is all categories combined). Each study is
represented by a bubble proportional to the inverse (Tau2 SE2). Large
the variance was explained). For available carbohydrate, heter- bubbles indicate studies with large weight. AC, available carbohydrate.

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


ogeneity 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 than can be expected from a change in GI alone for carbohydrate
0.001). No reliable (df not low) or significant (P 0.05) patterns intakes 400 g/d. In view of the overlap of the observations and
emerged among these data within each health type separately meta-regression lines associating GL and GI for the different
(healthy, CHD risk, type 1 diabetes, and type 2 diabetes) to food intake control categories (Figure 9) and the limitations of
explain these variations. Nor was there evidence of significant the study numbers, the statistical significance of the differences
residuals for studies combined by health type separately and between the slopes of the met-regression lines for the different
which could have suggested that a health type departs from these food consumption groups has not been assessed.
general trends. In all studies combined, treatment differences in available
carbohydrate intake contributed as much to variation in GL as
Glycemic load did GI (Figure 10). Heterogeneity in GL among all studies
The GL of the intervention diets decreased with progressive combined was 892 g glucose equivalent (2) and was reduced by
reduction in the GI significantly for all studies combined (P 54% after accounting for differences in available carbohydrate
kh-t 0.001) and for studies combined by food intake cate- intake univariately and by 93% after further accounting for GI
gories separately (controlled, P kh-t 0.001; limited con-
trolled, P kh-t 0.001; ad libitum, P kh-t 0.05; TABLE 6
Figure 9). Each had a corresponding slope on the meta- Difference in glycemic load (GL) in intervention studies related to
regression lines in the range of from 2.9 to 4.5 g glucose equiv- difference in glycemic index (GI) achieved, for each category of food
alent per unit GI (% glucose) (Table 6). The combined average intake1
rate of fall was 3.11 g glucose equivalent per GI % glucose, more
All studies Controlled Limited
combined intake control Ad libitum

k (df) 37 (35) 14 (12) 19 (17) 5 (3)


Slope (g/GI%glu) 3.1 3.2 2.9 4.5
SEE 0.4 0.6 0.50 1.3
P kh-t 0.001 0.001 0.001 0.046
3P 0.001 0.001 0.138
Constant (g/d) 10.2 13.2 7 21.9
SEE 6.4 12 9.3 16.1
P kh-t 0.122 0.298 0.439 0.268
3P 0.894 1.317 0.804
I2 (g2/g2) 0.67 0.24 0.78 0.31
Tau (g/d) 11 10 12 16
P Q 0.001 0.20 0.001 0.31
P X 0.001 0.23 0.001 0.26
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
FIGURE 8. Difference in available carbohydrate intake associated with total variation (Tau2 SE2). P Q, probability of significant heterogene-
unavailable carbohydrate intake found. ity; 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 meta-
regression 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 univari-
ately, with a marginal exception for fat in the controlled food
intake category.

DISCUSSION
A reduction in GI achieved by dietary intervention is com-
monly accompanied by changes in available carbohydrate in-
take. 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 carbohy-
FIGURE 10. Contribution of difference in glycemic index (GI; E) and drate intake that accompanies variance in GI (Figure 10). Also,
difference in available carbohydrate (AC; F) intake to the difference in available carbohydrate intake varies by g/g exchange of available
glycemic load (GL) achieved in the intervention studies. Data are (F, slope1 and unavailable carbohydrate. In large part, the wider range of
AC) and (E, slope2 GI) from the model GL slope1 AC

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


GL observed than expected from the change in GI alone is ex-
slope2 GI constant Tau SE. g eq, glucose equivalents.
plained by elevated intakes of available carbohydrate for low
reductions in GI followed by a trend toward progressive reduc-
bivariately. The overall effect on GL was additive, although not tion in the available carbohydrate intake with progressive reduc-
simply additive, because small reductions in GL due to reduced tion in GI. Reduction in the intake of available carbohydrate, and
GI were accompanied (confounded) by the intake of more avail- with it metabolizable energy, occurs when GI is reduced by 10
able carbohydrate, which would elevate GL (Figure 10). Such GI units on average or for GL 30 g glucose equivalent (Table 8).
confounding may be either circumstantial or a physiologic re- Overall, such reduction in metabolizable energy is contributed to
sponse to low GI. also by reductions in protein intake.
Reductions in metabolizable energy and available carbohy- There is no evidence even when relaxing control over food
drate intake associated with the reductions in GL achieved in 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
TABLE 7 from available carbohydrate. The generality of an inverse rela-
Difference in metabolizable energy, protein, available carbohydrates, fat, tion between available carbohydrate and fat intake is therefore
and unavailable carbohydrate intakes in intervention studies related to broken in this instance.
difference in glycemic load (GL) achieved, for each category of food
intake1
TABLE 8
All studies Limited Ad Reduction in glycemic index and load required before metabolizable
combined Controlled controlled libitum energy and carbohydrate intake are reduced1
Metabolizable energy Upper
k 38 14 19 5
x SE P kh-t 95% CI
Slope (kJ/g eq) 8.01 0.1 15.21 23.42
Constant (kJ/d) 2452 67 4722 420 Glycemic index (% glucose)
Protein Metabolizable energy
k 37 14 18 5 Limited control 10 6 0.140 24
Slope (g/g eq) 0.07 0.04 0.2 0.24 Ad libitum 9 2 0.002 15
Constant (g/d) 6.21 0.1 141 4.9 Available carbohydrate
Available carbohydrates Limited control 13 4 0.005 23
k 38 14 19 5 Ad libitum 10 4 0.061 22
Slope (g/g eq) 0.543 0.281 0.803 0.862 Glycemic load (g glucose equivalents)
Constant (g/d) 213 151 283 222 Metabolizable energy
Fat Limited control 31 6 0.001 44
k 38 14 19 5 Ad libitum 18 6 0.049 36
Slope (g/g eq) 0.02 0.092 0.01 0.17 Available carbohydrate
Constant (g/d) 3.3 6.2 2.3 0.8 Limited control 34 5 0.001 45
Unavailable carbohydrates Ad libitum 26 7 0.023 46
k 35 13 17 5 1
For glycemic index, data are intercepts on the x axis in Figures 1 and
Slope (g/g eq) 0.06 0.07 0.07 0.04
3, and for glycemic load they are the corresponding figures for similar
Constant (g/d) 4.1 1.3 5.9 3.4
analyses summarized in Table 7. Data are not shown for protein, which
1
P kh-t 0.01. yielded less consistent information, or for control treatments. Values were
2
P kh-t 0.05. calculated by using nonlinear combination of regression coefficients (ratio:
3
P kh-t 0.001. constant to slope).
FOOD, GLYCEMIA, AND HEALTH: DATABASE AND DIETS 235S
The present findings derive from random effects inverse vari- 12. Livesey G. Health potential of polyols as sugar replacers, with emphasis
ance meta-analysis and REML meta-regression across studies. on low glycaemic properties. Nutr Res Rev 2003;16:16391.
13. Robertson MD, Livesey G, Hampton SM, Mathers JC. Evidence for
Publication bias is not excluded among the present studies, altered control of glucose disposal after total colectomy. Br J Nutr 2000;
though no analyses were indicative. However, this is strictly 84:8139.
assessable only when fixed-effects analysis is applicable. The 14. Hofman Z, van Drunen JD, de Later C, Kuipers H. The effect of different
observed effects apply to combined health and body weight types nutritional feeds on the postprandial glucose response in healthy volun-
(healthy, impaired glucose tolerance, hyperlipidemia, type 1 di- teers 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
abetes, type 2 diabetes, primary and secondary CHD risk, and Food Res 2005;49:560 70.
normal, overweight, and obese weight for height). Generally, 16. Livesey G. Non-digestible carbohydrates and glycaemic control. In:
there is too little information of a consistent nature or just too Vorster H, Blaauw R, Dhansay M, Kuzwayo P, Moend L, Wentzel-
little information to establish the present findings in any one Viljoen E, eds. Nutrition Safari for Inovative Solutions, Proceedings of
the 18th International Congress of Nutrition. Durban, South Africa: ICC,
health or body weight type separately or for interventions of 2005.
12 wk duration. 17. Robertson MD, Bickerton AS, Dennis AL, Vidal H, Frayn KN. Insulin-
Because a rise in available carbohydrate intake with small sensitizing effects of dietary resistant starch and effects on skeletal
reductions in GI occurred in all 3 food intake control categories muscle and adipose tissue metabolism. Am J Clin Nutr 2005;82:559 67.
(ad libitum, limited controlled, and controlled), it may prove 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
difficult to avoid when using GI alone as an intervention tool. dietary fibre improved insulin economy in young women with impaired
Should this be so, a revised approach to interventions for the glucose tolerance. Eur J Clin Nutr 2006;60:334 41.
control of the postprandial glycemic response could be needed. 19. Livesey G. Glycaemic responses and toleration. In: Mitchell H, ed.
This is particularly so wherever small reductions in GI can be Sweeteners and sugar alternatives in food technology. Oxford, United
Kingdom: Blackwell, 2006:118.

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


expected. Likewise, it may be that maximum reduction in GL
20. Brand-Miller JC, Petocz P, Colagiuri S. Meta-analysis of low-glycemic
through contemporary strategies of GI reduction might not op- index diets in the management of diabetes: response to Franz. Diabetes
timally reach the intended goal. One consideration would be a Care 2003;26:3363 4, author reply 3364 5.
more direct attempt that reduces the GL of diets while controlling 21. Opperman AM, Venter CS, Oosthuizen W, Thompson RL, Vorster HH.
fat intake. Before doing so, evidence relating GL to health is Meta-analysis of the health effects of using the glycaemic index in
meal-planning. Br J Nutr 2004;92:367 81.
necessary and is addressed in the follow-up paper.
22. Anderson JW, Randles KM, Kendall CW, Jenkins DJ. Carbohydrate and
The authors are grateful to G Janecek, University of East Anglia, Norwich, fiber recommendations for individuals with diabetes: a quantitative as-
UK, for preliminary discussion of the statistical approach. sessment and meta-analysis of the evidence. J Am Coll Nutr 2004;23:
517.
The contributions of the authors were as followsGL and RT: data col-
23. Raben A. Should obese patients be counselled to follow a low-glycaemic
lection; GL: analysis; GL and JH: writing; and TH: comment. GL and RT had index diet? No. Obes Rev 2002;3:24556.
no financial or personal conflicts of interest. JH is currently advising an 24. Higgins JPT, Green S, eds. Cochrane handbook for systematic reviews
industry group comprising food manufacturers and retailers who are prepar- of interventions 4.1.6. Chichester, United Kingdom: Wiley & Sons, Ltd,
ing a submission to the authorities in Europe supporting the case for the use 2003.
of GI in the labeling of food products. TH works for Kellogg. 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;
REFERENCES 63:4753.
1. Brand JC, Colagiuri S, Crossman S, Allen A, Roberts DC, Truswell AS. 27. Kabir M, Oppert JM, Vidal H, et al. Four-week low-glycemic index
Low-glycemic index foods improve long-term glycemic control in breakfast with a modest amount of soluble fibers in type 2 diabetic men.
NIDDM. Diabetes Care 1991;14:95101. Metabolism 2002;51:819 26.
2. Salmeron J, Ascherio A, Rimm EB, et al. Dietary fiber, glycemic load, 28. Brynes AE, Mark Edwards C, Ghatei MA, et al. A randomised four-
and risk of NIDDM in men. Diabetes Care 1997;20:54550. intervention crossover study investigating the effect of carbohydrates on
3. Liu S, Willett WC, Stampfer MJ, et al. A prospective study of dietary daytime profiles of insulin, glucose, non-esterified fatty acids and tria-
glycemic load, carbohydrate intake, and risk of coronary heart disease in cylglycerols in middle-aged men. Br J Nutr 2003;89:20718.
US women. Am J Clin Nutr 2000;71:1455 61. 29. Giacco R, Parillo M, Rivellese AA, et al. Long-term dietary treatment
4. Kelly S, Frost G, Whittaker V, Summerbell C. Low glycaemic index with increased amounts of fiber-rich low-glycemic index natural foods
diets for coronary heart disease. Cochrane Database Syst Rev 2004: improves blood glucose control and reduces the number of hypoglyce-
CD004467. mic events in type 1 diabetic patients. Diabetes Care 2000;23:1461 6.
5. Oh K, Hu FB, Cho E, et al. Carbohydrate intake, glycemic index, gly- 30. Frost G, Leeds A, Trew G, Margara R, Dornhorst A. Insulin sensitivity
cemic load, and dietary fiber in relation to risk of stroke in women. Am J in women at risk of coronary heart disease and the effect of a low
Epidemiol 2005;161:1619. glycemic diet. Metabolism 1998;47:124551.
6. Franceschi S, Dal Maso L, Augustin L, et al. Dietary glycemic load and 31. Herrmann TS, Bean ML, Black TM, Wang P, Coleman RA. High gly-
colorectal cancer risk. Ann Oncol 2001;12:173 8. cemic index carbohydrate diet alters the diurnal rhythm of leptin but not
7. Giovannucci E. Insulin, insulin-like growth factors and colon cancer: a insulin concentrations. Exp Biol Med (Maywood) 2001;226:1037 44.
review of the evidence. J Nutr 2001;131:3109S20S. 32. Jenkins DJ, Wolever TM, Collier GR, et al. Metabolic effects of a
8. McCance RA, Lawrence RD. The carbohydrate content of foods. Med- low-glycemic-index diet. Am J Clin Nutr 1987;46:968 75.
ical Research Council special report series no. 135. London, United 33. Kurup PG, Krishnamurthy S. Glycemic response and lipemic index of
Kingdom: Her Majestys Stationery Office, 1929. rice, raggi and tapioca as compared to wheat diet in human. Indian J Exp
9. Jenkins DJ, Wolever TM, Taylor RH, et al. Glycemic index of foods: a Biol 1993;31:2913.
physiological basis for carbohydrate exchange. Am J Clin Nutr 1981; 34. Bouche C, Rizkalla SW, Luo J, et al. Five-week, low-glycemic index diet
34:362 6. decreases total fat mass and improves plasma lipid profile in moderately
10. Jenkins DJ, Axelsen M, Kendall CW, Augustin LS, Vuksan V, Smith U. overweight nondiabetic men. Diabetes Care 2002;25:822 8.
Dietary fibre, lente carbohydrates and the insulin-resistant diseases. Br J 35. Dumesnil JG, Turgeon J, Tremblay A, et al. Effect of a low-glycaemic
Nutr 2000;83(suppl):S157 63. indexlow-fat high protein diet on the atherogenic metabolic risk pro-
11. Bjorck I, Elmstahl HL. The glycaemic index: importance of dietary fibre file of abdominally obese men. Br J Nutr 2001;86:557 68.
and other food properties. Proc Nutr Soc 2003;62:201 6. 36. Pereira MA, Swain J, Goldfine AB, Rifai N, Ludwig DS. Effects of a
236S LIVESEY ET AL

low-glycemic load diet on resting energy expenditure and heart disease index improves dietary profile and metabolic control in type 2 diabetic
risk factors during weight loss. JAMA 2004;292:248290. patients. Diabet Med 1994;11:397 401.
37. Sloth B, Krog-Mikkelsen I, Flint A, et al. No difference in body weight 52. Jarvi AE, Karlstrom BE, Granfeldt YE, Bjorck IE, Asp NG, Vessby BO.
decrease between a low-glycemic-index and a high-glycemic-index diet Improved glycemic control and lipid profile and normalized fibrinolytic
but reduced LDL cholesterol after 10-wk ad libitum intake of the low- activity on a low-glycemic index diet in type 2 diabetic patients. Diabetes
glycemic-index diet. Am J Clin Nutr 2004;80:337 47. Care 1999;22:10 8.
38. Agus MS, Swain JF, Larson CL, Eckert EA, Ludwig DS. Dietary com- 53. Jimnez-Cruz A, Bacardi-Gascon M, Turnbull WH, Rosales-Garay P,
position and physiologic adaptations to energy restriction. Am J Clin Severino-Lugo I. A flexible, low-glycemic index Mexican-style diet in
Nutr 2000;71:9017. overweight and obese subjects with type 2 diabetes improves metabolic
39. Carels RA, Darby LA, Douglass OM, Cacciapaglia HM, Rydin S. Ed- parameters during a 6-week treatment period. Diabetes Care 2003;26:
ucation on the glycemic index of foods fails to improve treatment out- 196770.
comes in a behavioral weight loss program. Eat Behav 2005;6:14550. 54. Komindr S, Ingsriswang S, Lerdvuthisopon N, Boontawee A. Effect of
40. Ebbeling CB, Leidig MM, Sinclair KB, Hangen JP, Ludwig DS. A long-term intake of Asian food with different glycemic indices on dia-
reduced-glycemic load diet in the treatment of adolescent obesity. Arch betic control and protein conservation in type 2 diabetic patients. J Med
Pediatr Adol Med 2003;157:7739. Assoc Thai 2001;84:8597.
41. Spieth LE, Harnish JD, Lenders CM, et al. A low-glycemic index diet in 55. Tsihlias EB, Gibbs AL, McBurney MI, Wolever TM. Comparison of
the treatment of pediatric obesity. Arch Pediatr Adol Med 2000;154: high- and low-glycemic-index breakfast cereals with monounsaturated
94751. fat in the long-term dietary management of type 2 diabetes. Am J Clin
42. Wolever TM, Mehling C. High-carbohydrate-low-glycaemic index di- Nutr 2000;72:439 49.
etary advice improves glucose disposition index in subjects with im- 56. Wolever TM, Jenkins DJ, Vuksan V, et al. Beneficial effect of a low
paired glucose tolerance. Br J Nutr 2002;87:477 87. glycaemic index diet in type 2 diabetes. Diabet Med 1992;9:451 8.
57. Heilbronn LK, Noakes M, Clifton PM. The effect of high- and low-
43. Slabber M, Barnard HC, Kuyl JM, Dannhauser A, Schall R. Effects of a
glycemic index energy restricted diets on plasma lipid and glucose pro-
low-insulin-response, energy-restricted diet on weight loss and plasma
files in type 2 diabetic subjects with varying glycemic control. J Am Coll
insulin concentrations in hyperinsulinemic obese females. Am J Clin
Nutr 2002;21:120 7.

Downloaded from ajcn.nutrition.org by guest on January 30, 2017


Nutr 1994;60:48 53.
58. Jimnez-Cruz A, Turnbull WH, Bacardi-Gascon M, Rosales-Garay P. A
44. Calle-Pascual AL, Gomez V, Leon E, Bordiu E. Foods with a low
high-fiber, moderate-glycemic-index, Mexican style diet improves dys-
glycemic index do not improve glycemic control of both type 1 and type
lipidemia in individuals with type 2 diabetes Nutr Res 2004;24:19-27.
2 diabetic patients after one month of therapy. Diabetes Metab 1988;14:
59. Luscombe ND, Noakes M, Clifton PM. Diets high and low in glycemic
629 33.
index versus high monounsaturated fat diets: effects on glucose and lipid
45. Collier GR, Giudici S, Kalmusky J, et al. Low glycaemic index starchy metabolism in NIDDM. Eur J Clin Nutr 1999;53:473 8.
foods improve glucose control and lower serum cholesterol in diabetic 60. Rizkalla SW, Bellisle F, Slama G. Health benefits of low glycaemic
children. Diab Nutr Metab 1988;1:1119. index foods, such as pulses, in diabetic patients and healthy individuals.
46. Fontvieille AM, Acosta M, Rizkalla SW, et al. A moderate switch from Br J Nutr 2002;88(suppl):S255 62.
high to low glycaemic-index foods for three weeks improves the meta- 61. Wolever TM, Jenkins DJ, Vuksan V, Jenkins AL, Wong GS, Josse RG.
bolic control of Type 1 (IDDM) diabetic subjects. Diab Nutr Metab Beneficial effect of low-glycemic index diet in overweight NIDDM
1988;1:139 43. subjects. Diabetes Care 1992;15:562 4.
47. Fontvieille AM, Rizkalla SW, Penfornis A, Acosta M, Bornet FR, Slama 62. Zhang Y-F, Yang Y-X, Ma Z-L, Han J-H, Wang Z. The research on
G. The use of low glycaemic index foods improves metabolic control of application in the education with glycemic index in diabetic patients.
diabetic patients over five weeks. Diabet Med 1992;9:444 50. Acta Nutrimenta Sinica 2003;25:248 52.
48. Gilbertson HR, Brand-Miller JC, Thorburn AW, Evans S, Chondros P, 63. Jenkins DJ, Wolever TM, Kalmusky J, et al. Low-glycemic index diet in
Werther GA. The effect of flexible low glycemic index dietary advice hyperlipidemia: use of traditional starchy foods. Am J Clin Nutr 1987;
versus measured carbohydrate exchange diets on glycemic control in 46:66 71.
children with type 1 diabetes. Diabetes Care 2001;24:1137 43. 64. Frost G, Keogh B, Smith D, Akinsanya K, Leeds A. The effect of
49. Lafrance L, Rabasa-Lhoret R, Poisson D, Ducros F, Chiasson JL. Effects low-glycemic carbohydrate on insulin and glucose response in vivo and
of different glycaemic index foods and dietary fibre intake on glycaemic in vitro in patients with coronary heart disease. Metabolism 1996;45:
control in type 1 diabetic patients on intensive insulin therapy. Diabet 669 72.
Med 1998;15:972 8. 65. Frost GS, Brynes AE, Bovill-Taylor C, Dornhorst A. A prospective
50. Jenkins DJ, Wolever TM, Buckley G, et al. Low-glycemic-index starchy randomised trial to determine the efficacy of a low glycaemic index diet
foods in the diabetic diet. Am J Clin Nutr 1988;48:248 54. given in addition to healthy eating and weight loss advice in patients with
51. Frost G, Wilding J, Beecham J. Dietary advice based on the glycaemic coronary heart disease. Eur J Clin Nutr 2004;58:1217.

You might also like