Evaluation of Various Starchy Foods: A Systematic Review and Meta-Analysis on Chemical Properties Affecting the Glycemic Index Values Based on In Vitro and In Vivo Experiments

The chemical properties that serve as major determinants for the glycemic index (GI) of starchy food and recommended low-GI, carbohydrate-based foods have remained enigmatic. This present work performed a systematic assessment of linkages between chemical properties of foods and GI, and selected low-GI starchy foods. The data were sourced from literature published in various scientific journals. In total, 57 relevant studies and 936 data points were integrated into a database. Both in vitro and in vivo studies on GI values were included. The database was subsequently subjected to a meta-analysis. Meta-analysis from in vitro studies revealed that the two significant factors responsible for the GI of starchy foods were resistant starch and phenolic content (respectively, standardized mean difference (SMD): −2.52, 95% confidence interval (95%CI): −3.29 to −1.75, p (p-value) < 0.001; SMD: −0.72, 95%CI: −1.26 to −0.17, p = 0.005), while the lowest-GI crop type was legumes. Subgroup analysis restricted to the crop species with significant low GI found two crops, i.e., sorghum (SMD: −0.69, 95%CI: −2.33 to 0.96, p < 0.001) and red kidney bean (SMD: −0.39, 95%CI: −2.37 to 1.59, p = 0.001). Meta-analysis from in vivo studies revealed that the two significant factors responsible for the GI of starchy foods were flavonoid and phenolic content (respectively, SMD: −0.67, 95%CI: −0.87 to −0.47, p < 0.001; SMD: −0.63, 95%CI: −1.15 to −0.11, p = 0.009), while the lowest-GI crop type was fruit (banana). In conclusion, resistant starch and phenolic content may have a desirable impact on the GI of starchy food, while sorghum and red kidney bean are found to have low GI.


Introduction
Type 1 diabetes mellitus (T1DM) has become a chronic metabolic disorder worldwide, and the regulation of blood glucose at a near-normal level could best fit the goals of preventing or delaying long-term diabetes complications in T1DM [1]. Insulin treatment alone is inadequate for controlling T1DM; essentially, dietary adjustments are required for the proper regulation of blood glucose level. In addition to type 1 diabetes mellitus, the glycemic index is associated with other non-communicable diseases such as cardiovascular diseases (CVDs), type 2 diabetes, and cancer [2]. Glycemic index (GI) is defined as the blood glucose response measured as the area under the curve (AUC) in response to a test food that an individual consumes under standard conditions and is expressed as a percentage of the AUC following consumption of a reference food that the same individual

Risk of Bias Assessment
To assess the risk of bias, Cochrane risk of bias tool was employed [16]. This was performed by using the following criteria: allocation concealment; random sequence generation; participants and personnel blinding; outcome assessment blinding; selective reporting; incomplete outcome data; and other bias. Each study was evaluated and scored as having "high", "low", or "unclear" risk of bias. Patient and clinician blinding was highly difficult and generally not feasible in these tests, and we determined that the main outcome was less vulnerable to being affected by lack of blinding. Consequently, studies with a high risk of bias for any one (or more) of the key domains were considered to have a high risk of bias. Studies for all key domains except blinding were considered to have a low risk of bias; otherwise, studies were considered to have an unclear risk of bias [17].

Quality of Evidence Assessment
The quality of evidence for primary and secondary outcomes was evaluated according to GRADE (Grading of Recommendations, Assessment, Development, and Evaluation) procedure for risk of bias, inconsistency, indirectness, imprecision, and publication bias, which were categorized as high, moderate, low, or very low [18]. The results were then summarized in tables constructed using the GRADE system [18][19][20] (GRADE version 3.6) ( Table 1). Stages of the study, including literature search, data extraction, risk of bias assessment, and evidence grade assessment were performed independently by one author (Frendy Ahmad Afandi/FAA).

Statistical Analysis
We used weighted analysis using Hedges' d (Standard Mean Difference/SMD) for statistical methods. The data extracted from selected journals were mean, standard deviation or standard error, and the number of replicate experiments. The SMD with corresponding 95%CI values were pooled using the random-effects model. Exploration of heterogeneity across studies was carried out using the I 2 index [20] (the I 2 > 50% indicated sufficient heterogeneity), and publication bias was determined using the Begg's test and Egger's test (p < 0.05 was considered statistically significant for publication bias). We used Meta-Essentials tools for the meta-analysis process. The criterion of publication bias assessment approach using the GRADE System, consisted of selection, performance, attrition, detection, and reporting bias. The variables used for subgroup analysis were the method of study (in vivo or in vitro), type of reference food, and type of crops. erogeneity across studies was carried out using the I index [20] (the I > 50% indicated sufficient heterogeneity), and publication bias was determined using the Begg's test and Egger's test (p < 0.05 was considered statistically significant for publication bias). We used Meta-Essentials tools for the meta-analysis process. The criterion of publication bias assessment approach using the GRADE System, consisted of selection, performance, attrition, detection, and reporting bias. The variables used for subgroup analysis were the method of study (in vivo or in vitro), type of reference food, and type of crops.

Characteristics of Articles
Fifty-seven studies, involving 936 participants, were published from 2002 to 2019. Twenty-six studies [9][10][11][12]17,18,20,21,26,31,35,39,[43][44][45][46][47]49,[54][55][56][57][59][60][61][62] used an in vitro experiment, while thirty-two studies involved in vivo experiments. Only one study used both in vitro and in vivo data experiments [11]. The in vivo studies involved healthy participants, and some studies used rats [51,58]. The PICOS of this research is defined as Participants, Interventions, Comparisons, Outcomes, and Study Design. Participants in the in vivo experiment were healthy adults. Interventions were lower food chemical properties. Comparisons were higher food chemical properties. Outcomes of this research were resistant starch, dietary fibre, protein, phenolic, and flavonoid content significantly affecting GI starchy foods, while selected low-GI foods were sorghum and red kidney bean. The study design used in this research was the completely randomized design. Among fifty-seven included studies, forty-eight were selected for discussion of the relationship between chemical properties and GI value . The remaining studies were used for the selection of low-GI carbohydrate-based foods [55][56][57][58][59][60][61][62]. All studies reported changes in glycemic index, four studies [7,[17][18][19] reported changes in amylose content to GI, ten studies [9][10][11][12][13][20][21][22][23][24] reported changes in resistant starch (RS) content to GI, fifteen studies reported a change in dietary fibre content to GI, fifteen studies reported a change in fat content to GI, fifteen studies reported a change in protein content to GI, twelve studies reported a change in phenol content to GI, ten studies reported a change in flavonoid content to GI, five studies reported a change in cereal type to GI, six studies reported a change in tuber type to GI, two studies reported a change in fruit type to GI, and four studies reported a change in legume type to GI. Detailed characteristics of eligible studies are presented in Tables 2 and 3.  Thiranusornkij et al. 2019 [12] Odenigbo et al. 2012 [24] Ek et al. 2013 [45] Hidayat et al. 2017 [46] Singh et al. 2011 [47] Kumar et al. 2019 [10] Srikaeo and Sangkhiaw 2014 [11] Darandakumbura et al. 2013 [14] Ayerdi et al. 2005 [13] Oboh and Ogbebor 2010 [48] Vahini et al.

Characteristics of Articles
Fifty-seven studies, involving 936 participants, were published from 2002 to 2019. Twenty-six studies [9][10][11][12]17,18,20,21,26,31,35,39,[43][44][45][46][47]49,[54][55][56][57]59,[60][61][62] used an in vitro experiment, while thirty-two studies involved in vivo experiments. Only one study used both in vitro and in vivo data experiments [11]. The in vivo studies involved healthy participants, and some studies used rats [51,58]. The PICOS of this research is defined as Participants, Interventions, Comparisons, Outcomes, and Study Design. Participants in the in vivo experiment were healthy adults. Interventions were lower food chemical properties. Comparisons were higher food chemical properties. Outcomes of this research were resistant starch, dietary fibre, protein, phenolic, and flavonoid content significantly affecting GI starchy foods, while selected low-GI foods were sorghum and red kidney bean. The study design used in this research was the completely randomized design. Among fiftyseven included studies, forty-eight were selected for discussion of the relationship between chemical properties and GI value . The remaining studies were used for the selection of low-GI carbohydrate-based foods [55][56][57][58][59][60][61][62]. All studies reported changes in glycemic index, four studies [7,[17][18][19] reported changes in amylose content to GI, ten studies [9][10][11][12][13][20][21][22][23][24] reported changes in resistant starch (RS) content to GI, fifteen studies reported a change in dietary fibre content to GI, fifteen studies reported a change in fat content to GI, fifteen studies reported a change in protein content to GI, twelve studies reported a change in phenol content to GI, ten studies reported a change in flavonoid content to GI, five studies reported a change in cereal type to GI, six studies reported a change in tuber type to GI, two studies reported a change in fruit type to GI, and four studies reported a change in legume type to GI. Detailed characteristics of eligible studies are presented in Tables 2 and 3.
Other bias Selective reporting (reporting bias) Incomplete outcome data (attrition bias) Blinding of outcome assessment (detection bias) Blinding of participants and personnel (performance… Allocation concealment (selection bias)

Random Sequence generations (Selection bias)
low risk of bias unclear risk of bias high risk of bias

Secondary Outcomes
The contribution of six chemical properties to GI and five source types of carbohydrates to GI can be seen in Tables 4 and 5. Forty-eight studies  reported on the relationship between chemical properties and GI, while nine studies reported on the source types of carbohydrates to GI, respectively. Compared to other chemical properties of starchy foods or the source of carbohydrate-based foods, fat content did not reduce GI (SMD: 0.05, 95%CI: −0.16 to 0.27, p = 0.312; I 2 = 93.64%), as was also found in tuber type (SMD: −0.25, 95%CI: −0.93 to 0.43, p = 0.233; I 2 = 79.4%) or fruit type (SMD: 0.25, 95%CI: −0.60 to 1.10, p = 0.284; I 2 = 89.3%).        [58]. This means that the data used in the control are the data that have a proximate or bioactive compound analysis result (like fat, protein, phenolic, flavonoid, and others) that is higher than the data used in the intervention, such as in, e.g., Ramdath et al. (2004), who compared tannia and green banana. Green banana had higher protein content (2.7 g per serving food for glycemic index (GI) test) than tannia (2.6 g per serving food for GI test). The GI value of green banana is 109.00 ± 44.27, while that of tannia is 88.00 ± 28.46. Therefore, we input the GI value of green banana as the control data and the GI value of tannia as intervention data.

Strength of Evidence and Publication Bias
The quality of evidence was evaluated by the GRADE system. The level of evidence of RS content was at level A and highly recommended. Phenol content and legume type were at level B and moderately recommended. All evidence profiles for the primary and secondary outcomes are provided in Table 5. For the meta-analysis of RS content to GI food, any publication bias was observed by Begg's test and Egger's test (Begg's, p = 0.020; Egger's, p = 0.004) (Figure 5a). For phenol content on GI food, any publication bias was observed by Begg's test and Egger's test (Begg's, p = 0.001; Egger's, p = 0.008). For the meta-analysis of legume type on GI food, any publication bias was observed by Begg's test and Egger's test (Begg's, p = 0.087; Egger's, p = 0.077).

Relationship between Food Chemical Properties and GI
This work systematically reviewed the current accessible literature and found that, in general, among the various chemical properties, the presence of resistant starch and phenolic compounds exerted significant impacts on the GI of starchy foods. It is noteworthy that evidence of this finding was consistent with the previous study. Moreover, some chemical properties show an essential impact on the GI beyond those mentioned, i.e., flavonoid, protein, dietary fibre, and amylose, which were negatively correlated with GI. Further, we found that the crop type with the lowest GI value was legumes. Subgroup analysis revealed that significant low-GI crop species were sorghum and red kidney bean, while the subgroup analysis was restricted to trials that compared some crop species. This may relate to a larger quantity of RS and phenolic compounds in the crop species.
No meta-analysis has been conducted on the effect of resistant starch levels on starchy food's GI. Previous researchers stated that the relationship between resistant starch levels and GI is a negative correlation [19]. Resistant starch is starch that cannot be digested by the small intestine within 120 min after consumption but that the large intestine can ferment. Resistant starch is a linear molecule of α-1,4-D-glucan, which is obtained mainly from the retrogradation of the amylose fraction and has a relatively low molecular weight (1.2 × 10 5 Da) [74]. The results of a previous meta-analysis showed that resistant starch can reduce blood sugar levels and fasting insulin [9]. The same thing can be seen from the results of the meta-regression between resistant starch levels and effect size (Figure 5b), which has a negative slope (−0.09). The mechanism for decreasing GI is that resistant starch cannot be digested by digestive enzymes due to its compact molecular structure, such that there is no increase in blood sugar levels [74].
Furthermore, meta-analysis has not been conducted on the effect of dietary fibre levels on starchy foods' GI. Previous researchers stated that the relationship between dietary fibre levels and GI is negatively correlated [6,35]. The same thing can be seen from the result of the meta-regression between dietary fibre content and effect size, which has a negative slope equal to −0.11. The mechanism of action of dietary fibre to reduce GI is by slowing down the rate of digestion of starch and increasing the duration of intestinal transit so that dietary fibre serves as a physical barrier in digestion in the intestine, thus slowing down the interaction between enzymes and substrates. In addition, the degree of viscosity of the dietary fibre is positively related to the extent of the flattening of the postprandial glucose response [75].
Meta-analysis of the effect of fat content on starchy foods' GI has not been carried out. Previous researchers stated that the relationship of fat content to GI is negatively correlated [76]. The same thing can be seen from the results of the meta-regression between fat content and effect size, which has a negative slope as much as −0.07. The mechanism of GI reduction is that fat slows the rate at which the stomach empties, creates a steric hindrance for the enzyme [77], and interacts with amylose to form a very strong matrix, named amylo-lipid, which digestive enzymes have trouble digesting [75].
In addition, meta-analysis of the effect of protein levels on starchy foods' GI has not been done. Previous researchers stated that the relationship between protein levels and GI is negatively correlated [55]. The mechanism for decreasing GI is that a protein allegedly stimulates insulin secretion so that blood glucose is not excessive and is under control [78]. However, the meta-regression outcome between protein levels and effect size shows a different result, which has a positive slope of 0.02.
No meta-analysis has been conducted on the effect of phenol levels on starchy foods' GI. Previous researchers stated that the relationship between phenol levels and GI is negatively correlated [79]. The same can be seen from the result of the meta-regression between phenol levels and effect size, which has a negative slope, as much as −0.0001. The mechanism of GI reduction is that phenol inhibits the α-amylase enzyme and the α-glucosidase enzyme [80].
No meta-analysis has been conducted on the effect of flavonoid levels on starchy foods' GI. Previous researchers stated that the relationship between flavonoid levels and GI is negatively correlated [66]. The same can be seen from the result of the meta-regression between fat content and effect size, which has a negative slope, as much as −0.0002. The mechanism of GI reduction is that flavonoids inhibit the α-amylase and α-glucosidase enzymes [80].

Comparability of In Vitro Results to Forecast In Vivo Correlations to Chemical Properties and GI
This study used both the in vivo and in vitro methods. This consideration was based on previous studies in which in vitro test results had the same trend as in vivo tests, though in vitro tests tended to have absolute values (overestimation) about 5-25 higher than those of in vivo tests [11,81,82]. That gap was strengthened by the results of the absolute value of SMD chemical properties towards the GI of food, which, in vitro, has a higher value compared to in vivo (Tables 4 and 6). Other considerations are several theories stating that at least 10 studies must be carried out in a meta-analysis. In the analysis, subgroup analysis and sensitivity analysis are performed. Subgroup analysis was performed based on the variable type of study and reference food. This can provide an overview of two aspects if both methods are used and if only the in vivo method with 50 g glucose reference food os used (Table 6). For in vivo tests, we used and compared only those using 50 g glucose reference food. This is done because the results show that if the reference food is 50 g of white bread or white rice, it is necessary to first convert the GI value with multiplier factors, respectively, 0.77 and 0.69 [83]. If the reference food used is 25 g glucose, different results would be obtained, which would need to be multiplied by 0.67 as a conversion factor [84].
For in vivo food model simulation, we suggest some recommendations. First, our study found that RS and phenolic content had positive effects on the reduction of the GI value of starchy foods. Such a correlation is stable and reliable. Thus, RS and phenolic content should be recommended for the chemical properties of starchy foods determined to affect the GI value. Second, to date, little attention has been paid to the study of the chemical properties determinant that affects the glycemic index of starchy foods and selected low-GI carbohydrates using meta-analysis. The determinant factor discussed in this study can be a meaningful direction for further research. Finally, comprehensive in vivo trials are warranted to validate [17] the positive impact of these findings.

Determinant of Chemical Properties Affecting the GI and Low-GI Carbohydrate Foods
In our study, the determinant effect of RS or phenolic content on GI and legume, as the selected low-GI carbohydrate food, was in accordance with the previous meta-analysis [9]. Nevertheless, differences between our study and the previous analysis should be noted. First, the previous meta-analysis included thirteen trials with the involvement of 428 participants for the RS effect, while for phenolic content to GI, no meta-analysis research was found until now. However, Ramdath et al. (2014) [62] asserted that there was a significant inverse correlation between polyphenol content and the GI of potatoes (r = −0.825; p < 0.05; n = 4). In the case of the relation between legume and GI, the previous study using meta-analysis included forty trials totaling 253 participants. We included ninety-eight trials, and added subgroup analysis based on chemical properties and the source of starchy foods according to the control group, enabling us to reach a more robust conclusion by eliminating interference factors. Our meta-analysis found that heterogeneity among trials was due mainly to the design of different control group, rather than population. In addition, we assessed the quality of the evidence and the strength of the recommendations. Thus, our work was the latest and most comprehensive one.
Low-GI foods, such as legumes, had the lowest GI compared to other carbohydrate foods because legumes have relatively higher levels of resistant starch and bioactive compounds. Beans' resistant starch levels were 24.7% [74]. In the cereal group, red sorghum had the lowest GI (Table 5), apparently due to the higher content of bioactive compounds and resistant starch compared to other cereal groups. Red sorghum's resistant starch levels ranged from 3.34 to 65.36 g/100 g [85]. The phenol compound contained in sorghum is 445-2850 µg/g [86]. In the legumes, red beans had the lowest GI (Table 5), presumably due to the higher content of bioactive compounds and resistant starch compared to other legumes. The resistant starch content of red bean starch is 21.27% [87], while the total phenol content can reach 4871 mg gallic acid equivalents (GAE)/100 g dry weight [88]. Previous studies [89] only determined the glycemic index of various staple carbohydrate-rich foods in the UK diet in five groups, such as breakfast cereals, breads, pastas, and potatoes using ten subjects.
Our study also has limitations. Though this meta-analysis includes high-quality studies, the sample sizes are small. Additionally, there is enough heterogeneity among studies, which could alter the reliability of the results. Trials with higher-quality samples and larger sample sizes are required to confirm the current results.

Critical Review GI as Indicator for Classifying Healthy Foods and the Alternative Concept
Although the concept of GI is widely used in explaining the causes of diabetes, some scientists consider that GI is not accurate enough to explain this. The concept of GI is considered inappropriate to classify a food as healthy or not or to describe its impact on human health. Several aspects of criticism of the GI concept include reproducibility, its impact on physiological effects, and levels and standards of the reference food used [90]. Therefore, it is necessary to use other indicators related to the character of carbohydrates besides digestibility, such as the types of food fibre found in foodstuffs and the levels of bioactive compounds contained therein [91]. Substitutes for the concept of GI proposed by health nutrition researchers include a new method for classifying starch digestion by modeling amylolysis of plant foods using first-order enzyme kinetic principles. This research opens new horizons and supports the relationship between levels of resistant starch, dietary fibre, phenolic, flavonoids, and the value of food GI.
The results evidenced that resistant starch and phenolic content reduce the GI value of starchy foods. As regards in vitro studies, it is well known that resistant starch is negatively correlated to GI, but the results obtained for the phenolic content in this systematic review were not obvious, even though part of phenolics were bound to fibre compounds known to reduce the digestion in the flattening of postprandial glucose response. Moreover, phenols inhibit the α-amylase and α-glucosidase enzymes. Among cereals, sorghum-a gluten free cereal-is the only one that reveals high-resistant starch and low GI. This is a very interesting result due to the fact that gluten-free foods generally are low in fibre and high in GI. This finding could be useful to investigate the potential of sorghum as gluten-free products.

Conclusions
The present work successfully elucidated that resistant starch, phenolic, flavonoid, protein, and dietary fibre content exerted crucial roles in reducing the glycemic index of starchy foods. Among the starchy foods, sorghum and red kidney bean were identified to have low-GI properties. Sorghum and kidney beans have a low GI because both contain relatively high resistant starch and phenolic compounds. The relationship between levels of resistant starch, phenolic, flavonoids, protein, and fibre to GI was a negatively correlation. Resistant starch causes steric hindrance in the molecular structure of the starch, while phenolic compounds (including flavonoids) are capable of inhibiting the α-amylase and α-glucosidase enzymes. The mode of action of resistant starch in reducing GI is making the enzyme unable to hydrolyze and disrupting the hydrolysis on non-resistant starch (which creates steric hindrance). Nevertheless, microbes ferment the resistant starch in the colon so that the body will not absorb it as glucose. Protein is supposed to stimulate the secretion of insulin so that blood glucose is not excessive and can be controlled. The fibre functions as an inhibitor of physical digestion in the intestine, thereby slowing down the interactions between enzymes with the substrates.