Professional Documents
Culture Documents
Glycemic Index, Glycemic Load, and Blood Pressure: A Systematic Review and Meta-Analysis of Randomized Controlled Trials
Glycemic Index, Glycemic Load, and Blood Pressure: A Systematic Review and Meta-Analysis of Randomized Controlled Trials
1176 Am J Clin Nutr 2017;105:1176–90. Printed in USA. Ó 2017 American Society for Nutrition
baseline results or a change in blood pressure from baseline to 10. A broadly symmetrical funnel plot was taken to indicate an
follow-up for each arm. For the latter, we calculated the dif- absence of small-study effects.
ference in the change between groups with the use of the ap- To determine whether some heterogeneity was due to existing
propriate t test to provide the difference between groups along confounders such as age, weight loss, BMI at baseline, or energy or
with a measure of variation. If only a P value was provided for macronutrient intake, we carried out meta-regressions on available
the difference between arms or if the results were displayed in a data. To determine whether there was a dose response for the asso-
figure but not presented in a table, the SE of the difference ciation of GI on blood pressure, we converted GI difference between
between arms was estimated. groups to the glucose scale (if on the white bread scale) with the use of
We used a random-effects meta-analysis of the intervention trial the methods by Wolever et al. (28) and split the trials at the median into
data as our primary outcome. We carried out a fixed-effects meta- 2 groups, those with a high difference in GI and those with a low
analysis as a sensitivity analysis. A weighted mean difference was difference in GI, to provide the pooled estimate for each subgroup.
calculated (weighted by the inverse of the variance). Heterogeneity
was presented as the proportion of the total variation in study es- RESULTS
timates that was due to between-study heterogeneity (I2) (26). It is
common to interpret I2 as being excessive where the value is in Trial characteristics
excess of 50–75%. We chose to use 75% as our cutoff because Fourteen trials comprising 1097 participants provided data on
there is higher methodologic variability when the exposure is a the effects of high- or low-GI or -GL diets on blood pressure, all
dietary factor (27). We generated the pooled estimate together with of which were included in a meta-analysis. The main reasons for
95% CIs, but where I2 values were .75%, no pooled estimate was exclusion at the second data extraction stage in the original re-
generated. We assessed small-study effects, such as publication view were related to study design (n = 411), type of carbohydrate
bias, with the use of a funnel plot if the number of studies exceeded (n = 322), length of the trial (n = 265), and lack of healthiness of
FIGURE 1 Flowchart to indicate the number of studies included at each stage of the review. GI, glycemic index; GL, glycemic load.
on 04 August 2018
Authors, year Actual diet characteristics,
(ref), country, and Characteristics of Intervention Intervention description for each macronutrient intake, energy GI or GL value Group Weight
study name participants2 duration group intake, and fiber (scale)3 weight change change difference4
Abete et al., 2008 56% male, age: 36, 8 wk Lower GI (n = 32): energy restricted. E%: CHO 50, P 18, F 32 GI 40–45 (bread) 27.5% 2.2%
(29), Spain BMI: 32 Individually prescribed diet within Fiber: 24.9 g/d
a strict dietary framework repeated
on a 3-d rotation basis; 84% of
CHO provided by pasta and
legumes.
Higher GI (n = 32): energy restricted. E%: CHO 48, P 20, F 33 GI 60–65 (bread) 25.3%
Individually prescribed diet within Fiber: 18.5 g/d
a strict dietary framework repeated
on a 3-d rotation basis; 84% of
CHO provided by rice and
potatoes.
Bellisle et al., 0% male, age range: 12 wk Lower GI (n = 96): Weight Watchers Not reported Not reported (bread) 21.5 kg 20.2 kg
2007 (30), 20–72, BMI program5 with a focus on low-GI
France range: 25–40 foods.
Higher GI (n = 65): Weight Watchers Not reported Not reported (bread) 21.7 kg
program.
Buscemi et al., 48% male, age: 50, 3 mo Lower GI (n = 47): diet containing E%: CHO 55, P 20, F 25 GI 44, GL 96 (n/a) 28.3 kg 1.2 kg
2013 (31), Italy BMI: 34 low-GI foods such as pasta, whole Fiber: 32 g/d
Higher GI/GL (n = 24): lower-fat E%: CHO 59, P 19, F 23 GI 53, GL 77 (g/1000 26.1 kg
diet. Meal plans based on an Energy: 1409 kcal/d kcal) (glucose)
exchange system; energy deficit of Fiber: 17.8 g/d
250–500 kcal/d.
Ebbeling et al., 21% male, age: 28, 6 mo Lower GL (n = 73): advice to consume — GI 45, GL 30 (glucose) — 0.5 kg
2007 (33), USA BMI: .30 low-GL foods such as nonstarchy
vegetables, legumes, and temperate
fruits, and to limit refined grains,
starchy vegetables, fruit juices, and
sweets.
Higher GL (n = 66): low-fat diet. — GI 55, GL 70 (glucose) —
Advice to consume low-fat grains,
vegetables, fruits, and legumes,
and to limit added fats, high-fat
snacks, and sweets.
(Continued)
1179
by guest
TABLE 1 (Continued )
1180
on 04 August 2018
Authors, year Actual diet characteristics,
(ref), country, and Characteristics of Intervention Intervention description for each macronutrient intake, energy GI or GL value Group Weight
study name participants2 duration group intake, and fiber (scale)3 weight change change difference4
Fava et al., 2013 49% male, age: 54, 24 wk Lower GI/GL (n = 88)—High- High-MF, low-GI—Energy: GI 54 (bread) 0.2 kg n/a
(34), UK, BMI: 29 monounsaturated, low-GI: target 2019 kcal/d
RISCK trial as E%, F 38, SF 10, MF 20, PF 6, E%: CHO 46, P 17, F 35
CHO 45, GI 53%. Fiber: 20 g/d
Lower GI/GL (n = 88)—High- High-CHO, low-GI—Energy: GI 56 (bread) 20.8 kg
carbohydrate, low-GI: target as 1854 kcal/d
E%, F 28, SF 10, MF 11, PF 6, E%, CHO 55, P 18, F 23
CHO 55, GI 51%. Fiber: 22 g/d
Higher GI/GL (n = 77)—Higher- High-MF, high-GI—Energy: GI 66 (bread) 0.4 kg
monounsaturated, high-GI: target 2056 kcal/d
as E%, F 38, SF 10, MF 20, PF 6, E%: CHO 43, P 16, F 38
CHO 45, GI 64%. Fiber: 19 g/d
Higher GI/GL (n = 77)—High- High-CHO, high-GI— GI 66 (bread) 21.8 kg
carbohydrate, high-GI: target as E%, Energy: 1645 kcal/d
F 28, SF 10, MF 11, PF 6, CHO 55, E%: CHO 51, P 20, F 27
GI 64%. Fiber: 17 g/d
Gögebakan et al., 36% male, age: 41, 26 wk Lower GI (n = 773)—Low-protein, low- Not reported Target GI 15% lower Low P, low GI: n/a
2011 (35), BMI: 34 GI: target E%, F 23–28, CHO 57–62, than high GI 0.27 kg
P 23–28.
Higher GI (n = 487)—Low protein, Not reported Target GI 15% higher Low P, high GI:
high GI: target E%, F 23–28, than low GI 1.45 kg
CHO 57–62, P 10–15. (glucose) High P, high GI:
Higher GI (n = 487)—High protein, 0.36 kg
high GI: Target E%, F 23–28,
CHO 45–50, P 23–28.
Jensen et al., 2008 0% male, age 10 wk Lower GI (n = 55): received low-GI E%: CHO 81, P 13, F 6 GI 72 (glucose) 22 kg 0.7 kg
(36) Denmark, range: 20–40, test foods in place of their usual Energy: 4860 kJ/d
The Danish GI BMI: 28 CHO-rich foods. Fiber: 29 g/d
study Higher GI (n = 44): received high-GI E%: CHO 82, P 13, F 6 GI 95 (glucose) 21.3 kg
test foods in place of their usual Energy: 4886 kJ/d
CHO-rich foods. Fiber: 32 g/d
Maki et al., 2007 33% male, age: 50, 36 wk Lower GL (n = 86): dietary advice g/d: CHO 69 P 97 F 80 GI 48, GL 8173 24.5 kg 1.9 kg
(19), USA BMI: 32 for ad libitum reduced-GL foods Energy: 1365 kcal/d (bread)
Fiber: 11 g/d
Higher GL (n = 84)—Higher GL, g/d: CHO 168, P 75, F 62 GI 51, GL 12,118 22.6 kg
lower fat: reduce fat intake, Energy: 1525 kcal/d (bread)
decrease portion sizes, target Fiber: 12 g/d
energy deficit 500–800 kcal/d.
(Continued)
by guest
on 04 August 2018
TABLE 1 (Continued )
Melanson et al., 12% male, age: 39, 12 wk Lower GI (n = 157): whole-grain E%: CHO 49, P 23, F 30 GI 42, GL 45 (bread) 23.4 kg 20.3 kg
2012 (37), USA BMI: 31 foods such as whole-grain cereals, Energy: 5878 kJ/d
whole-grain pasta, oatmeal, and Fiber: 14 g/d
whole-grain bread, with refined
grains used sparingly.
Higher GI (n = 85): dietary advice to E%: CHO 47, P 20, F 31 GI: 47 GL: 42 (bread) 23.7 kg
follow Weight Watchers diet based Energy: 5772 kJ/d
on points aiming to control Fiber: 13 g/d
portions rather than food types.
Pereira et al., 2004 23.7% male, age: Low GL: 65 d Lower GI (n = 46): energy-restricted E%: CHO 43, P 27, F 30 GI: 50, GL 82 (bread) 21.1 kg/wk 0.1 kg
(18), USA 31, BMI: Low fat: 69 d low-GL diet (60% of predicted Energy: 1500 kcal/d
overweight or requirements). Fiber: 32 g/d
obese Higher GI, lower fat (n = 34): E%: CHO 65, P 17, F 18 GI 82, GL 205 (bread) 21.0 kg/wk
energy-restricted low-fat diet Energy: 1500 kcal/d
(60% of predicted requirements). Fiber: 20 g/d
NCEP Step 1 diet.
Philippou et al., 100% male, age 6 mo Lower GI (n = 56): carbohydrate g/d: CHO 224 GI 51, GL 114 0.7 kg
change difference4
CHO, carbohydrate; DiOGenes, Diet, Obesity and Genes; E%, percentage of energy; F, fat; GI, glycemic index; GL, glycemic load; MF, monounsaturated fat; n/a, not available; NCEP, National
available in the Department of Health report (22). Eight trials
Cholesterol Education Program; P, protein; PF, polyunsaturated fat; ref, reference; RISCK, Reading, Imperial, Surrey, Cambridge, King’s; SF, saturated fat; UK, United Kingdom; USA, United States.
Weight
0.2 kg
were included in the original search. In the updated search, 482
studies were obtained with the use of the same search criteria,
of which 29 were identified as potentially relevant. The reasons
for excluding potentially relevant articles included the follow-
ing: blood pressure not reported at both baseline and follow-up
(n = 6), participants not healthy (n = 5), not a relevant GI or GL
weight change
26 kg
Group
GI 47%, GL 92 g
(bread)
France (1), Germany (1), Italy (1), New Zealand (1), and Spain (1).
The value denotes a difference in weight between the low-GI and the high-GI diet. A positive value indicates a greater loss in low GI. All the trials used a parallel group design that ranged in duration
from 2 to 12 mo. The first results reported after the end of the
intervention were used in the analyses. All the studies included
macronutrient intake, energy
Fiber: 21 g/d
centages, but these were not transformed (35, 40). The median
Intervention description for each
total energy (18, 19, 32, 38). For the 10 studies that reported actual
nutrients (not targets), the median differences in energy, pro-
tein, total fat, carbohydrate, and fiber between groups were as
follows: median energy was 12 kcal higher in the low-GI diets;
14% male, age: 42,
Characteristics of
in the low-GI diets; the median difference in total fat was the
BMI: 35
(40), New
Zealand
2
3
4
5
6
Abete et al., 2008 (29) Unclear Unclear Unclear Unclear Bias No bias No bias
Bellisle et al., 2007 (30) Unclear Unclear Bias Bias No bias No bias No bias
Buscemi et al., 2013 (31) No bias Unclear Bias No bias No bias No bias No bias
Ebbeling et al., 2007 (33) No bias No bias Bias No bias No bias No bias No bias
Ebbeling et al., 2005 (32) Unclear Unclear Unclear Unclear No bias No bias No bias
Fava et al., 2013 (34) Unclear Unclear Bias Unclear Bias No bias No bias
Gögebakan et al., 2011 (35) Unclear Unclear Bias Unclear Bias No bias No bias
Jensen et al., 2008 (36) No bias Unclear No bias No bias No bias No bias No bias
Maki et al., 2007 (19) Unclear Unclear Bias Unclear No bias No bias No bias
Melanson et al., 2012 (37) Unclear Unclear Bias Unclear Bias No bias No bias
Pereira et al., 2004 (18) Unclear Unclear Unclear Unclear Bias No bias No bias
Philippou et al., 2009 (38) Unclear Unclear Bias Unclear Bias No bias No bias
Randolph et al., 2014 (39) Unclear Unclear Bias Unclear No bias No bias No bias
Venn et al., 2010 (40) Unclear Unclear Bias Unclear Bias No bias No bias
All studies reported body weight, either in kilograms or BMI, subjects were mostly overweight or obese men and
percentage of weight change at follow-up compared with women who followed a hypoenergetic diet (generally in the
baseline. Body weight decreased in the majority of trials in both region of 1500 kcal/d).
groups (Table 1), although weight loss was slightly more
pronounced in the low-GI diets, with a median difference in
weight loss of 0.5 kg between groups. The difference between Quality of trials
groups was not statistically significant for the majority of The results of the quality check are reported in Table 2. The
studies, although the study by Abete et al. (29) reported sig- majority of the trials reported that subjects and researchers were
nificant differences between groups with higher weight loss on not blinded to the nature of the intervention, although 1 study was
the low-GI diet. Of the 10 studies that reported mean baseline double-blind (36). Ten of the studies stated that there was no
FIGURE 2 Difference in SBP (expressed in mm Hg) between low-GI diet and high-GI diet. The forest plot displays the weighted difference in means,
95% CI for difference in means, and unit difference in GI index between groups for each study. CHO, carbohydrate; D+L, DerSimonian and Laird random-
effects estimate; GI, glycemic index; ID, identification; I-V, inverse variance fixed-effects estimate; SBP, systolic blood pressure.
blinding of participants (19, 31, 33–35, 37–40) or both researchers high-GI diets. Two studies reported data from 4 arms: the study
and participants (30). Some trials did not provide sufficient in- by Fava et al. (34) included results on differences between low-
formation, particularly on blinding of researchers. Furthermore, and high-GI diets from participants on high-carbohydrate and
only one trial clearly described good allocation concealment (33), high-monounsaturated diets, and the study by Gögebakan et al.
whereas the remaining trials did not provide enough information (35) included results on differences from participants on low-
to make a definite decision. In addition, none of the trials mea- protein and high-protein diets. The total number of data points
sured blood pressure as the primary outcome. In all studies, the possible in the forest plots was therefore 16. Thirteen of the 14
primary outcome was a measure of body fatness. Nevertheless, trials reported the difference in GI between groups in GI units;
some studies stated that a protocol was followed for measuring however, one trial did not report this information (30). Nine of
blood pressure (30, 35), or they provided details on length of time the trials reported the difference in GL between groups, and 8 of
participants were at rest before measurement (18, 19, 29, 33, 37, these 9 trials reported that the low-GI diet was also the low-GL
38), whereas the remaining studies did not provide any informa- diet. However, 1 study reported that the high-GI diet was lower
tion. A further indicator of trial quality was the degree to which in GL (37).
adherence to the diet was monitored and encouraged. One study The summary estimate for all 14 trials (16 comparisons)
used urinary nitrogen to measure adherence (35), and 2 studies when using random-effects methods indicated that SBP was
provided food and checked adherence through food diaries (18, 1.13 mm Hg (95% CI: 20.25, 2.51 mm Hg; P = 0.11) lower with
36). The most common method of assessment was through diaries consumption of a lower-GI diet (Figure 2). The results for fixed-
(30–33, 41, 42). The remaining trials did not provide any infor- effects methods indicated that SBP was 1.10 mm Hg (95% CI:
mation on adherence. The quality of the studies was, therefore, 20.20, 2.40 mm Hg; P = 0.10) lower with consumption of a
generally poor, with most studies categorized as prone to bias or lower-GI diet. The estimates for individual studies ranged from
unclear. This is a common problem with dietary studies because of 24.9 mm Hg to 16.0 mm Hg. The proportion of variation from
the difficulties inherent in the conduct of double-blind dietary real effects rather than sampling error was low (I2 = 9%). The
studies. For this reason, we did not exclude studies from the re- summary estimate for all 14 trials (16 comparisons) when using
view based on our quality check. random-effects methods indicated that DBP was 1.26 mm Hg
(95% CI: 0.22, 2.30 mm Hg; P = 0.02) lower with consumption
of a lower GI diet (Figure 3). The results for fixed-effects
Blood pressure methods indicated that DBP was 1.18 mm Hg (95% CI: 0.29,
All 14 studies provided information on, or permitted esti- 2.08 mm Hg; P = 0.01) lower with consumption of a lower-GI
mation of, differences in both SBP and DBP between low- and diet. The estimates for individual studies ranged from 22.9 to
FIGURE 3 Difference in DBP (expressed in mm Hg) between low-GI diet and high-GI diet. The forest plot displays the weighted difference in means,
95% CI for difference in means, and unit difference in GI index between groups for each study. CHO, carbohydrate; DBP, diastolic blood pressure; D+L,
DerSimonian and Laird random-effects estimate; GI, glycemic index; ID, identification; I-V, inverse variance fixed-effects estimate.
Mechanisms this may promote energy intake regulation, which often leads
The mechanisms for the effect of low-GI diets on blood lipid to weight loss. Weight loss has been identified as a strong pre-
profiles and blood pressure readings are not clear. A low-GI diet dictor of lower blood pressure, and, therefore, it is a proba-
is usually high in some types of dietary fiber, in particular ble confounder. However, as noted above, the authors think
soluble fiber, but it is not necessarily low in carbohydrate, this is unlikely to be the cause in this review, because most of
whereas a low-GL diet is low in total carbohydrate. Low-GI the trials were comparing a low-energy, low-GI, or low-GL
and -GL diets both tend to have a low energy density, and diet with a low-energy diet of higher GI or GL. In addition,
FIGURE 6 Difference in SBP (expressed in mm Hg) between low-GL diet and high-GL diet. The forest plot displays the weighted difference in means,
95% CI for difference in means, and grams difference in GL index between groups for each study. D+L, DerSimonian and Laird random-effects estimate; GL,
glycemic load; ID, identification; I-V, inverse variance fixed-effects estimate; SBP, systolic blood pressure.
the trials reported similar differences in body weight between amounts of weight, but in trials of short duration, we may be
baseline and follow-up between the control and intervention unable to detect weight-loss differences that would emerge in
groups. In most cases, participants in both groups lost similar the long term.
FIGURE 7 Difference in DBP (expressed in mm Hg) between low-GL diet and high-GL diet. The forest plot displays the weighted difference in means,
95% CI for difference in means, and grams difference in GL index between groups for each study. DBP, diastolic blood pressure; D+L, DerSimonian and Laird
random-effects estimate; GL, glycemic load; ID, identification; I-V, inverse variance fixed-effects estimate.
Given that changes to blood pressure are a composite of altered carbohydrate fraction, including dietary protein and fiber con-
sympathovagal balance, leading to increases in heart rate and tent, energy density, and sensory quality. The review may have
stroke volume accompanied by changes to arteriolar tone, the excluded informative studies ,6 wk in duration, such as the
impact of high-GI diets on both sympathetic tone and endothelial OmniCarb Trial (54); however, there is no universally agreed
function should be considered. The decrease in blood pressure on length of follow-up, and the pragmatic length of 6 wk was
may be a consequence of lower sugars acting on sympathetic tone selected in advance for this review. In addition, the review did
and epithelial function, rather than just a function of slowly not include children and adolescents, and therefore the re-
digested starch. Indeed, diets high in fructose are associated with sults from this review cannot be extrapolated to younger age
elevated blood pressure (47) and increases in sympathetic tone groups.
(48), whereas increased glucose intake is also associated with
increases in basal heart rate (49). Furthermore, high plasma uric
acid concentrations, which are associated with both increased Policy implications
fructose and glucose consumption, are also associated with en- Systematic reviews of the associations between fiber and blood
dothelial dysfunction (50, 51). Dietary protein is also thought to pressure and lipids report similar or smaller effects on health than
have an impact on blood pressure (52). It is not clear to what this review (55–57); therefore, it is possible that lower-glycemic
extent the improvements in blood pressure are due to individual diets do offer a further beneficial effect over and above a high-
components of a low-GI diet, including lower amounts of overall fiber and low-fat diet by encompassing benefits from many
carbohydrates and sugars and higher amounts of plant proteins components. However, it is not clear exactly which compo-
and soluble fibers. nents of a low-GI diet are responsible for the improvements in
blood pressure. Indeed, there was no strong evidence of a dose
response. Advising on a low-GI diet to healthy individuals is
Strengths and limitations of this review
more complex than describing a high-fiber diet, but lower-
To our knowledge, this is the first comprehensive review of the glycemic diets are generally rich in high–soluble fiber foods,
effects of GI and GL on blood pressure. It included RCTs, which such as oats, beans, legumes, vegetables, and whole fruits, and
are considered to be the highest quality of study, of $6 wk in low in sweetened drinks. High-quality research in normal-
duration, as well as meta-analysis. The review was carried out weight individuals is needed to enable the contribution of
with the use of Preferred Reporting Items for Systematic Re- dietary manipulation to markers of CVD to be established,
views and Meta-Analyses guidelines while using an established independent of weight changes before inclusion of a low-GI
and published protocol. diet in nutrition policy.
However, many of the trials included low numbers of par-
ticipants in each group, far below the sample of many hundreds of
participants needed to detect differences in blood pressure of Conclusions
2 mm Hg with reasonable power. This resulted in large SEs for In relatively healthy individuals, lower-glycemic diets are
most of the individual estimates, and although the values for I2 associated with significantly better profiles of blood pressure,
were low, there was still a wide range of estimates, with over- although no clear dose response was apparent in these analy-
lapping CIs indicating high levels of heterogeneity. Limitations ses. Furthermore, many of the trials included in the review aimed
of using I2 as a measure of heterogeneity are discussed in detail to reduce weight in participants, making it difficult to isolate
by Borenstein et al. (53). Blood pressure was not the primary the impact of diet on blood pressure. The trials were also subject
outcome for any of the trials, and, therefore, the quality of the to considerable sources of bias, as is often the case in trials
data on blood pressure could be below the standard expected, as involving food-based interventions. Before lower-glycemic diets
well as not being powered to detect differences in these sec- are universally recommended by health professionals, high-quality
ondary outcomes. Even so, a review of this type, with pooled trials in healthy normal-weight populations are needed to deter-
estimates from a meta-analysis of .1000 participants in total, is mine the effects of GI on blood pressure independent of weight
able to detect small consistent differences. Many of the studies change.
did provide information on how blood pressure was measured,
and some followed a published protocol, so it is unlikely that We thank Charlotte Woodhead and Camilla Nykjaer for their contributions
blood pressure was poorly measured. to data extraction and checking, James Thomas for his work on developing the
database into which all articles were extracted, and David Haughton for his
Because many of the markers of CVD are related to weight, it
advice on mechanisms of glycemic index on blood pressure.
was difficult to isolate the contribution of the type of diet, in this The authors’ responsibilities were as follows—VJB: was the project
case the GI, as separate from changes in weight. It cannot be ruled lead for the main systematic review of dietary carbohydrates and cardi-
out that weight loss, at least in the short term, is explaining some ometabolic health and developed the research plan and had primary
of the beneficial effects of a low GI diet on blood pressure. responsibility for the final content of the manuscript; VJB, DET, and
Within the trials, there was some variation in the methods CLC: searched the databases; CPG: helped develop search strategies;
used to calculate GI and GL. Accordingly, the individual au- VJB, DET, CLC, and CELE: undertook article screening; VJB, DET,
thor definitions of high and low GI and GL have been adopted DCG, CLC, and CELE: conducted data extraction; DET and CLC: con-
ducted quality of data extraction and checking; CELE: conducted the
to compare studies, even when the apparent differences be-
statistical analysis and wrote the first draft of the manuscript; DCG:
tween trial arms appear to be quite small or not in accord with oversaw the statistical analysis; and all authors: reviewed the manu-
notions of what may be viewed as high or low. Unless tightly script, contributed to manuscript revisions, and read and approved the
controlled in an experimental situation, in most cases, high- final manuscript. None of the authors reported a conflict of interest re-
and low-GI and -GL diets differ in many ways other than the lated to the study.