PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptNIH Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Nutrition. Author manuscript; available in PMC Nov 1, 2007.
Published in final edited form as:
PMCID: PMC1989668
NIHMSID: NIHMS13644
Methodology for adding glycemic index and glycemic load values to 24-hour dietary recall database
Barbara C. Olendzki, R.D., M.P.H.,a* Yunsheng Ma, M.D., Ph.D.,a Annie L. Culver, B.S.Pharm,b Ira S. Ockene, M.D.,c Jennifer A. Griffith, M.S.,a Andrea R. Hafner, B.S.,a and James R. Hebert, Sc.D.d
a Division of Preventive and Behavioral Medicine, University of Massachusetts Medical School, Worcester, Massachusetts, USA
b On Call Clinicians, Redwing Fairview Medical Center Pharmacy, Redwing, Minnesota, USA
c Division of Cardiology, University of Massachusetts Medical School, Worcester, Massachusetts, USA
d South Carolina Statewide Cancer Prevention and Control Program, University of South Carolina, Columbia, South Carolina, USA
* Corresponding author. Tel.: +508-856-5195; fax: +508-856-2022. E-mail address:Barbara.olendzki/at/umassmed.edu (B. C. Olendzki).
Objectives
We describe a method of adding the glycemic index (GI) and glycemic load (GL) values to the nutrient database of the 24-hour dietary recall interview (24HR), a widely used dietary assessment. We also calculated daily GI and GL values from the 24HR.
Methods
Subjects were 641 healthy adults from central Massachusetts who completed 9067 24HRs. The 24HR-derived food data were matched to the International Table of Glycemic Index and Glycemic Load Values. The GI values for specific foods not in the table were estimated against similar foods according to physical and chemical factors that determine GI. Mixed foods were disaggregated into individual ingredients.
Results
Of 1261 carbohydrate-containing foods in the database, GI values of 602 foods were obtained from a direct match (47.7%), accounting for 22.36% of dietary carbohydrate. GI values from 656 foods (52.1%) were estimated, contributing to 77.64% of dietary carbohydrate. The GI values from three unknown foods (0.2%) could not be assigned. The average daily GI was 84 (SD 5.1, white bread as referent) and the average GL was 196 (SD 63).
Conclusion
Using this methodology for adding GI and GL values to nutrient databases, it is possible to assess associations between GI and/or GL and body weight and chronic disease outcomes (diabetes, cancer, heart disease). This method can be used in clinical and survey research settings where 24HRs are a practical means for assessing diet. The implications for using this methodology compel a broader evaluation of diet with disease outcomes.
Keywords: Glycemic index, 24-hour dietary recalls, Nutrition
The concept of the glycemic index (GI) has provided the scientific community with a new way to examine the quality of carbohydrate (CHO), which has been useful in research on the etiology and prevention of diabetes, coronary heart disease, obesity, prostate cancer, and other chronic diseases [16]. Historically, CHOs have been classified into two major forms: complex and simple. Monosaccharides, and disaccharides are simple CHOs. Polysaccharides are complex CHOs and include starch, cellulose, fiber, and related compounds. A defect of this classification scheme has been its inability to predict plasma glucose and insulin responses [7], which are critical factors in the genesis of many health outcomes.
The GI, developed by Jenkins et al. [8] in 1981, allows the comparison between CHO foods on the basis of their physiologic effects rather than on their chemical composition. The currently accepted physiologic method [9] for determining the GI value of a food is to feed at least 10 healthy subjects 50 g of digestible (available) CHO from the test food and then measure the effect on their capillary blood glucose levels over the next 2 h. The area under the 2-h blood glucose response curve (glucose AUC) for each person for this food is then derived. On a separate occasion, test subjects consume a 50-g portion of a reference food (white bread or glucose) and their 2-h capillary blood glucose response is again measured for the corresponding AUC. A GI value for the test food is then calculated for each person by dividing that person’s glucose AUC for the test food by the glucose or white bread AUC for the reference food. The final GI value for the test food is the average GI value for all subjects [8,10]. This ranking of CHO-containing foods according to the glycemic response has proved to be uniquely useful to our understanding of the effect of particular types of CHO on health [1,11,12].
Importantly, different foods containing equal amounts of CHO can produce a wide range of glycemic responses in comparison with the average blood glucose response of a referent food [9]. The effect of CHO on blood glucose and insulin concentrations is determined primarily by the amount of CHO consumed (quantity) and the rate of absorption (quality). Glycemic response and the subsequent determination of GI values are influenced by factors beyond type of CHO, such as fat, protein, fiber, and nutrient composition of the food. Cooking and other processing methods contribute to further variation in the response. In general, foods with a low degree of starch gelatinization or with a higher level of soluble fiber have slower rates of digestion and therefore lower GI values [11].
The glycemic load (GL) is determined by the combined effect of quality and quantity per serving of CHO ingested from a particular food. The GL is the GI of a food multiplied by its CHO content in grams. GI and GL are thus closely related, but they represent different ways to evaluate the response to a CHO-rich food. GL reflects a food’s physiologic effect better than the amount of CHO or the GI alone [13].
Although many researchers have applied GI estimation values to dietary assessment instruments [1417], the methodology used to obtain these values has only recently been documented in detail in a few studies using the food frequency questionnaire (FFQ) [1719]. However, although similar, description of methodology in assigning GI values in 24-hour dietary recalls (24HRs) is currently lacking. The objectives of the present study are to describe the methods used to establish the GI and GL from an extensive set of 24HRs and to calculate overall dietary GI and GL. The database of foods comes from the 24HRs collected from a healthy population in central Massachusetts, as part of the Seasonal Variation of Blood Cholesterol Levels (SEASONS) study, in which we also examined variations of GI and GL and then associations with body weight and blood lipids using 24HRs [2023]. This documentation will help to continue the effort to establish a consistent methodology to add GI values to dietary assessment tools and may result in greater use of the GI in research and clinical care.
Dietary assessment
The 24HRs were collected by trained dietitians using the Nutritional Data System (NDS), which was developed and is maintained by the University of Minnesota’s Nutrition Coordinating Center (NDS DOS versions 2.6, 2.7, and 2.8) [24]. CHO-containing foods (individual and mixed dishes) were then re-entered into a more current version of NDS (NDS 5.0_35 for Windows), which provides greater depth of analysis, specifically on types of CHO (i.e., glucose, fructose, sucrose), types of fiber (soluble or insoluble), and fat content. NDS provides 136 nutrients and food components (such as phytoestrogens and carotenoids) for analysis on more than 18 000 foods in the database. All foods from the physiologically tested and published International Table of Glycemic Index and Glycemic Load Values (International Table) [9] and University of Sydney online databases (http://www.glycemicindex.com) were subsequently entered into the same NDS system to facilitate nutrient correlations with study foods. Detailed food preparation methods and portion sizes were systematically collected as part of our study. This allowed for better estimation of GI and enabled GL calculation on an individual level.
Estimation of GI and GL
Available data were derived from 9067 24HRs providing information on a total of 1482 different foods. Because the GI is based on foods that contain CHO, we used the NDS nutrient analysis program to exclude food items without CHO, leaving a total of 1261 CHO-containing food items for analysis in the SEASONS study.
The next step involved matching our foods directly to those in the International Table and/or the updated University of Sydney online database (the same investigators established both). The international Table provides GI averages from studies that tested foods for physiologic response and geographical location, whereas the University of Sydney online database provides additional detail as to the population in which the food was tested (normal, diabetic, or metabolic syndrome), the reference for who performed the testing, average GI values from each study, and the duration of the testing (2 or 3h). All foods from the International Table or the online database that were matched to our foods were analyzed through NDS to provide a standard platform of comparison. It should be noted that NDS provides a breakdown of total CHO including fiber, whereas the International Table is based on available CHO (with total fiber content subtracted). This is one of the reasons that all foods from the International Table used in comparison with those in the present study were first analyzed through NDS.
It also was deemed important to match foods to those in the International Table by selecting the food with best geographic and botanical matches, when that information was available, and to consider food preparation methods (addition of other ingredients, cooking time, etc.) that are described as part of the collection of the 24HR. If more than one study performed physiologic testing and was included in the International Table, GI values of the pertinent botanical and geographic studies were averaged to arrive at the established GI.
The International Table and the University of Sydney databases are excellent sources of information but not an exhaustive listing of all foods and corresponding GI values. In particular, these databases do not contain many of the foods from our 24HRs of American foods. Thus, each food not found in the International Table [9] had to be matched to a similar food. In a simple example that highlights only types of sugars (one of the factors in determination of the GI), consider Pepsi, with ~27 g of total CHO per 250-mL serving, containing ~41% fructose (GI 19), ~37% glucose (GI 100), ~20% sucrose (GI 68), and ~0.9% maltose (GI 105). Pepsi is a “best match” to beverages in the International Table and then to Coca Cola, which also has ~27 g of the same percentages of these types of CHO per 250-mL serving. Coca Cola has a GI value of 63; this then becomes the assigned GI value for Pepsi. This is a much better match than to a similar beverage, Fanta, which has ~32 g of total CHO per 250-mL serving, with ~18% fructose, ~17% glucose, ~62% sucrose, and 0.8% maltose and a GI value of 68.
Similar foods were determined by ingredient breakdown. First, the nutritionist (B.C.O.) estimated which foods from the International Table or online database should be compared with our dataset. Second, portion size was standardized between foods, which were then matched as follows: total CHO was broken down into sugars as described by the example above, and fiber content was characterized and compared (soluble versus insoluble, amylose versus amylopectin). Further consideration for matches assessed fat content, acidity, particle size, protein, and cooking and processing methods. NDS provides in-depth analysis of most nutritional factors such as fat and fiber that allow for close links to similar foods in the International Table or online database. In general, foods with a low degree of starch gelatinization or with a higher level of soluble fiber have slower rates of digestion and absorption and therefore lower GI values [11]. Some factor information (acidity, particle size, amylose:amylopectin ratio) was obtained through food science references [2528]. Our study’s foods that were not matched directly to foods in the International Table and University of Sydney online database were matched to the available foods listed there with similar characteristics, as described above, to achieve estimates of GI value. The nutritionist made final determinations based on all available information and from subjective experience and knowledge of foods.
Mixed dishes represented an interesting challenge because the evaluative process involved breaking down ingredients of the mixed dish to consider direct or similar matches of individual ingredients. NDS was employed for nutrient calculation and composition, with attention to type of CHO such as sucrose, fructose, glucose, food preparation methods, and other GI factors (i.e., fiber, fat, acid). Using the same analytical techniques described above, the ingredients were then matched to tested foods in the International Table or the University of Sydney online database. We assigned the GI for each ingredient, by calculating the overall GI according to a weighted average of the GI values of each food, based on the proportion of the total CHO contributed by each food. [29].
There were three foods in the database unknown to us, and for which there was not enough information to establish a GI value, because the early version of NDS that was used for the original collection did not supply enough nutrient information. These foods are no longer commercially available and are not in the current version of NDS. These three foods, listed in the early version of NDS as “Second Nature,” “Figurines,” and “Sego Lite,” contributed little to the CHO consumption in our study and were omitted from the final analysis.
There was a small number of different commercial sugar cookies (n = 13) for which an average GI value was assigned. These 13 foods were grouped together with an average GI for the group, rather than assessed individually as we did with the rest of the foods. This was done because the early version of NDS was unable to provide sufficient data to allow analysis through the current NDS database. We had to select the GI value based on macronutrients and fiber rather than on specific types of CHO and food preparation techniques. All 13 of these foods were processed foods, likely to contain refined CHO, and were assigned values commensurate with similar cookies or digestives in the International Table and University of Sydney online database.
Confidence in estimations
Two nutritionists with a master’s degree were involved with assignment of GI values in this study. Each nutritionist doing estimations assigned a subjective quality control confidence level, ranging from 1 to 5, to each food’s GI resolution. Blinded to previous estimations, this estimate was redone by each nutritionist before running results from the analysis, and approximately 10% of the foods were adjusted after further analysis. Because we did not do direct testing of the foods themselves, a level of 5 (100% confident of GI value) was not achieved. A value of 4 was assigned if the food had a direct match (same food) to the International Table, a value of 3 if the food was closely related to a food in the International Table or online database (matched on GI factors and nutrient information), a value of 2 if the food was lacking a close match on one of the GI factors, and a value of 1 if the food was assigned an average GI value for the group of foods, rather than matching on individual GI factors and nutrient information (such as the 13 types of cookie). Of 1261 foods, 1160 (91%) achieved a level of ≥3.
Calculation of daily GI and GL
The recommended formulas [30] for the calculation of daily GL and GI are as follows:
equation M1
equation M2
where GIi is the GI for food i, CHOi is the CHO content in food i (grams per day), and n is the number of foods eaten per day. The NDS program provided the CHO content of foods necessary for GL calculation. The GI value for the food was multiplied by 1.43 to convert from glucose to white bread as the referent GI value [9].
The average caloric intake calculated from the 9067 24HRs was 1966 kcal/day (SD 569), the percentage of calories from CHO was 51 (SD 7.5), and the daily intake of CHO was 247 g (SD 74). Study participants were mostly European American, middle-aged, and overweight; men and women were equally represented, and subjects were predominantly non-smokers. Table 1 lists subjects’ general characteristics.
Table 1
Table 1
Demographic information of SEASONS study participants (n = 641)*
In total, 1261 CHO-containing foods were identified from the 9067 24HRs from the SEASONS study. GI values of 602 foods were obtained from direct matches (47.7%), which accounted for 22.36% of dietary CHO. GI values from 656 foods (52.1%) were estimated, contributing to the remaining 77.64% of estimable dietary CHO. The GI values from three foods (0.2%) could not be assigned; these three foods contributed little CHO. A confidence level of 4 (direct match with food in table) was assigned to 602 foods (48% of total), a confidence level of 3 (close match) was assigned to 558 foods (44% of total), a confidence level of 2 was assigned to 85 foods (6.7% of total), 13 foods were assigned a confidence level of 1 (1%), and 3 foods (0.2%) were unable to be assigned (and therefore dropped from analyses). The average daily GI was 84 (SD 5.1, white bread 100) and the GL was 196 (SD 63). The distribution of GI for the SEASONS study population is shown in Figure 1.
Fig. 1
Fig. 1
Distribution of glycemic index from the Seasonal Variation of Blood Cholesterol Levels study population.
Table 2 lists the top 20 CHO sources of the SEASONS participants. For ease of comparison with other studies, the categories of white rice (3.71% of CHO) and potatoes (3.50% of CHO), which have a range of GI values depending on the particular variety of rice or potato, were collapsed into the average GI value for rice or potatoes, respectively. These top 20 foods accounted for 48% of CHOs consumed from all 1261 foods. Table 3 lists the top 20 highest GI foods and the frequency of consumption by this population. Interestingly, the foods with the highest GI values were not necessarily frequently consumed (cornstarch 0.44%, an ingredient added as a thickening agent with a GI of 159, and dehydrated potatoes 0.35%, with a GI of 134). The top 20 foods with the highest GLs are listed in Table 4. The ranking of these foods is quite similar to the top 20 CHO foods, indicating that CHO, and not GI ranking, had the greatest influence on GL in this population.
Table 2
Table 2
Primary sources of carbohydrate from top 20 contributors
Table 3
Table 3
Top 20 highest GI foods and frequency of consumption
Table 4
Table 4
Top 20 highest GL foods
This report describes our methods of estimating the GI, which determined our calculations of the overall GI and GL in our study. The clinical utility of the GI as a method of blood glucose control remains in question, because the American Diabetes Association (ADA) primarily recommends CHO counting for blood glucose control, although recognizing that several GI factors influence the glycemic response to foods [31,32] and that GI and GL may have additional benefit. It is noted in the ADA position paper that integration of the GI as a primary method of glycemic control is quite complex when it is applied to mixed foods and mixed meals. The GI is not currently labeled on American food products and only recently has the GI begun to appear on European labels. Although physiologic testing is the standard for establishing GI values, this testing is performed reputably in only a few places, and it is costly and time consuming. Therefore, it may not be within the range of many researchers and clinicians to consider.
Our methods are consistent with those described in three recent publications. Flood et al. [18] described a similar methodology in applying matching and estimation to the National Cancer Institute Diet History Questionnaire Database, a food frequency dietary assessment instrument. Neuhouser et al. [19] described their methods used for adding the GI to the FFQ used in the Women’s Health Initiative, recognizing the importance of methodology in GI estimation. Using a semiquantitative 114-item FFQ, Schulz et al. [17] also briefly described their methods in assigning GI. There are many studies that have examined the association of GI and GL with disease, but few studies (using FFQ or 24HR) have mentioned their methods of GI estimation; those that explained methods did so only in a cursory fashion as part of a summary description, and not as the main focus of the paper [17]. The studies by Flood et al. and Neuhouser et al. are among the first to describe in detail the steps they used to determine GI values for their FFQ database. It is important that researchers are consistent in the methodology of assigning GI values to dietary assessment instruments. This will improve the evaluation of GI and any association with chronic disease, and it will assist with utilization of GI and GL in clinical and research settings.
We chose the 24HR for estimation of the GI for several reasons. The FFQ includes a short list of predefined food items and food groups (both with standardized portions and totaling about 60–150 food items), and queries about average diet over a specified period in the recent past (usually 3 mo to 1 y). Nutrient scores are derived from comparison with a standard portion size multiplied by the frequency of food consumed. The groupings of foods inherent in the design of an FFQ may, and probably do, represent a wide range of GI values. The actual consumption of the study population of particular foods within categories is not known, and because of its semiquantitative nature, the precise portion size eaten is unknown. It is inherent in the nature of FFQs that average GI values must be assigned to a grouping of foods that may actually have wide ranges of GI values and would be expected to vary considerably from person to person. However, both dietary assessment methods have their place; the FFQ is much less expensive to administer than 24HRs, and the methodology for determining the GI in FFQ can be accomplished as described by Flood et al. and Neuhouser et al. Because of the practical utility of FFQs in large cohort studies, few 24HRs have been tested for this purpose [14,33]. The 24HR can be used as validation of the FFQ in this dimension, in a manner similar to that of nutrient comparison [3436].
The GI is developed to measure CHOs consumed with respect to glycemic effect; calculation of GL is based on knowledge of specific food items, in combination with the portion size and CHO content of each food item consumed. Therefore, the GL based on multiple 24HRs may be more accurate with regard to the actual food consumption of individuals and populations than that based on the FFQ. Using this methodology in the context of 24HR is important per se (because it is a practicable assessment method in clinical research). It also has added relevance because the 24HR is often the referent method in validation studies of structured questionnaires including the FFQ [3739], and it is reasonable to consider GI or GL as a parameter for which to test in validation.
Although the GI has been determined for several FFQs, its use in the 24HR is only rarely documented. Recently, two studies have established the GI from 24HRs [14,33]. Davis et al. [33] reported that 44% of foods were best-matched to the International Table on CHO profiles, 33% were matched to similar foods based on CHO profiles, and 23% were matched to a range average of similar foods. Davis et al. summarized their methods of GI estimation (they ran their foods through the NDS as we did), but the main focus of their report was not to describe the methods of estimation. Our study results showing a mean daily GI of 84 were close to the estimation of Davis et al. who reported an average daily GI of 89 in an elderly population. We have listed estimated GI and GL values in comparison with other populations and studies in Table 5.
Table 5
Table 5
A summary of GI and GL values from the literature
There were several limitations in our methodology. First, we acknowledge that our source of estimation of the GI, the International Table and online databases, may be subject to error in GI values, including within-subject and between-laboratory variations. However, Wolever et al. [40,41] demonstrated that, if the recommended physiologic methods were used, the results of GI agreed reasonably well (i.e., ±5%). In addition, because each subject’s glycemic response to a given food is standardized to that subject’s glycemic response to the reference food, this effectively takes between-subject variation out of the equation. Second, data collected using this early version of NDS were incomplete for some foods in terms of nutrient data available, preparation methods, and brands. Subsequent versions of NDS have improved tremendously, but still fall short of capturing all the factors influencing the determination of the GI of any particular food. If the GI is to become clinically useful, this additional information will have to be made widely available (eventually appearing on food labels and in the NDS database) and methods of determination must be standardized. Third, although the NDS has an excellent database of individual items, there are still many items that may be missed (although we encountered few such foods in this study).
In addition, the International Table (and in Europe and Canada) total CHOs contain only digestible CHOs, whereas the NDS is a United States–based nutritional data system, and subsumes soluble and insoluble fiber under total CHO. This is an important reason to standardize the analysis platform (running all foods compared through the same analytical software) to arrive at our estimations. Other studies that have described methodology have also emphasized the importance of this consideration [1719].
Our study population consisted largely of white, middle-class people who were mostly members of a health maintenance organization. The study protocol involved a lengthy series of clinic visits and diet assessments, and participants were likely to have been relatively motivated. For these reasons, although our methods of GI determination may be used in a variety of studies, our dataset of foods may not be fully generalizable to other socioeconomic strata and to other cultures and ethnic groups.
Conclusion
This is the first study that describes a methodology for adding GI and GL into a 24HR-derived database, the most widely used dietary assessment method used in large nutritional surveys, such as the National Health and Nutrition Examination Survey. We believe that describing and understanding methods to estimate the GI will increase the research and clinical utility of the GI and GL and help to standardize the GI estimation for those studies relying on dietary assessment tools.
Acknowledgments
The authors thank Philip Merriam, Laura Robidoux, and Priscilla Cirillo for assistance with study recruitment and data collection; Kelly Scribner for coordination of 24-h recalls; and SEASONS dieticians who conducted the 24-h recalls: Susan Nelson, Christine Singelton, Pat Jeans, Karen Lafayette, Deborah Lamb, Stephanie Olson, and Eileen Capstraw.
Footnotes
The project described was supported by grants R01-HL52745 to Dr. Ira S. Ockene and 1 R21 HL074895-01 to Dr. Yunsheng Ma from the National Heart, Lung, and Blood Institute (NHLBI). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NHLBI.
1. Gilbertson HR, Thorburn AW, Brand-Miller JC, Chondros P, Werther GA. Effect of low-glycemic-index dietary advice on dietary quality and food choice in children with type 1 diabetes. Am J Clin Nutr. 2003;6:83–90. [PubMed]
2. Liu S, Willett WC, Stampfer MJ, Hu FB, Franz M, Sampson L, et al. A prospective study of dietary glycemic load, carbohydrate intake, and risk of coronary heart disease in US women. Am J Clin Nutr. 2000;71:1455–61. [PubMed]
3. Norris SL, Zhang X, Avenell A, Gregg E, Bowman B, Schmid CH, et al. Long-term effectiveness of weight-loss interventions in adults with pre-diabetes: a review. Am J Prev Med. 2005;28:126–39. [PubMed]
4. Augustin LS, Galeone C, Dal Maso L, Pelucchi C, Ramazzotti V, Jenkins DJ, et al. Glycemic index, glycemic load and risk of prostate cancer. Int J Cancer. 2004;112:446–50. [PubMed]
5. Salmeron J, Ascherio A, Rimm EB, Colditz GA, Spiegelman D, Jenkins DJ, et al. Dietary fiber, glycemic load, and risk of NIDDM in men. Diabetes Care. 1997;20:545–50. [PubMed]
6. Ma Y, Olendzki BC, Chiriboga D, Rosal MC, Sinagra E, Crawford S, et al. PDA-assisted low glycemic index dietary intervention for type 2 diabetes: a pilot study. Eur J Clin Nutr. 2006 May 17; [Epub ahead of print] [PubMed]
7. Crapo PA, Reaven G, Olefsky J. Plasma glucose and insulin responses to orally administered simple and complex carbohydrates. Diabetes. 1976;25:741–7. [PubMed]
8. Jenkins DJ, Wolever TM, Taylor RH, Barker H, Fielden H, Baldwin JM, et al. Glycemic index of foods: a physiological basis for carbohydrate exchange. Am J Clin Nutr. 1981;34:362–6. [PubMed]
9. Foster-Powell K, Holt SH, Brand-Miller JC. International table of glycemic index and glycemic load values: 2002. Am J Clin Nutr. 2002;76:5–56. [PubMed]
10. Brand-Miller JC. Glycemic load and chronic disease. Nutr Rev. 2003;61(5 pt 2):S49–55. [PubMed]
11. Wolever TM, Jenkins DJ, Jenkins AL, Josse RG. The glycemic index: methodology and clinical implications. Am J Clin Nutr. 1991;54:846–54. [PubMed]
12. Michaud DS, Fuchs CS, Liu S, Willett WC, Colditz GA, Giovannucci E. Dietary glycemic load, carbohydrate, sugar, and colorectal cancer risk in men and women. Cancer Epidemiol Biomarkers Prev. 2005;14:138–47. [PubMed]
13. Willett WC. Eat, drink, and be healthy. New York: Simon and Schuster; 2001.
14. Scholl TO, Chen X, Khoo CS, Lenders C. The dietary glycemic index during pregnancy: influence on infant birth weight, fetal growth, and biomarkers of carbohydrate metabolism. Am J Epidemiol. 2004;159:467–74. [PubMed]
15. Lajous M, Willett W, Lazcano-Ponce E, Sanchez-Zamorano LM, Hernandez-Avila M, Romieu I. Glycemic load, glycemic index, and the risk of breast cancer among Mexican women. Cancer Causes Control. 2005;16:1165–9. [PubMed]
16. Liese AD, Schulz M, Fang F, Wolever TM, D’Agostino RB, Jr, Sparks KC, et al. Dietary glycemic index and glycemic load, carbohydrate and fiber intake, and measures of insulin sensitivity, secretion, and adiposity in the Insulin Resistance Atherosclerosis Study. Diabetes Care. 2005;28:2832–8. [PubMed]
17. Schulz M, Liese AD, Mayer-Davis EJ, D’Agostino RB, Jr, Fang F, Sparks KC, et al. Nutritional correlates of dietary glycaemic index: new aspects from a population perspective. Br J Nutr. 2005;94:397–406. [PubMed]
18. Flood A, Subar AF, Hull SG, Zimmerman TP, Jenkins DJ, Schatzkin A. Methodology for adding glycemic load values to the National Cancer Institute Diet History Questionnaire database. J Am Diet Assoc. 2006;106:393–402. [PubMed]
19. Neuhouser ML, Tinker LF, Thomson C, Caan B, Horn LV, Snetselaar L, et al. Development of a glycemic index database for food frequency questionnaires used in epidemiologic studies. J Nutr. 2006;136:1604–9. [PubMed]
20. Ma Y, Olendzki B, Chiriboga D, Hebert JR, Li Y, Li W, et al. Association between dietary carbohydrates and body weight. Am J Epidemiol. 2005;161:359–67. [PMC free article] [PubMed]
21. Ma Y, Olendzki BC, Li W, Hafner AR, Chiriboga D, Hebert JR, et al. Seasonal variation in food intake, physical activity, and body weight in a predominantly overweight population. Eur J Clin Nutr. 2006;60:519–28. [PMC free article] [PubMed]
22. Ma Y, Li Y, Chiriboga DE, Olendzki BC, Hebert JR, Li W, et al. Association between carbohydrate intake and serum lipids. J Am Coll Nutr. 2006;25:155–63. [PMC free article] [PubMed]
23. Ma Y, Griffith JA, Chasan-Taber L, Olendzki BC, Jackson E, Stanek EJ, III, et al. Association between dietary fiber and serum C-reactive protein. Am J Clin Nutr. 2006;83:760–6. [PMC free article] [PubMed]
24. NDS. Nutrition Data System. Version 2.9. Minneapolis: University of Minnesota, The Nutrition Coordinating Center; 1996.
25. McGee H. On food and cooking. The science and lore of the kitchen. New York: Charles Scribner’s Sons; 1984.
26. Groff J, Gropper SS. Advanced nutrition and human metabolism. 3rd ed. Belmont, CA: Wadsworth; 2000.
27. Mahant K, Escott-Stump S, editors. Krause’s food, nutrition, & diet therapy. 10th ed. Philadelphia: WB Saunders; 2000.
28. Hallfrisch J, Facn, Behall KM. Mechanisms of the effects of grains on insulin and glucose responses. J Am Coll Nutr. 2000;19(suppl):320S–5. [PubMed]
29. Wolever TM, Jenkins DJ. The use of the glycemic index in predicting the blood glucose response to mixed meals. Am J Clin Nutr. 1986;43:167–72. [PubMed]
30. Pi-Sunyer FX. Glycemic index and disease. Am J Clin Nutr. 2002;76:290S–8. [PubMed]
31. American Diabetes Association. Standards of medical care in diabetes—2006. Diabetes Care. 2006;29(suppl 1):S4–42. [PubMed]
32. Sheard NF, Clark NG, Brand-Miller JC, Franz MJ, Pi-Sunyer FX, Mayer-Davis E, et al. Dietary carbohydrate (amount and type) in the prevention and management of diabetes: a statement by the American Diabetes Association. Diabetes Care. 2004;27:2266–71. [PubMed]
33. Davis MS, Miller CK, Mitchell DC. More favorable dietary patterns are associated with lower glycemic load in older adults. J Am Diet Assoc. 2004;104:1828–35. [PubMed]
34. Dennis LK, Snetselaar LG, Nothwehr FK, Stewart RE. Developing a scoring method for evaluating dietary methodology in reviews of epidemiologic studies. J Am Diet Assoc. 2003;103:483–7. [PubMed]
35. Mayer-Davis EJ, Vitolins MZ, Carmichael SL, Hemphill S, Tsaroucha G, Rushing J, et al. Validity and reproducibility of a food frequency interview in a multi-cultural epidemiology study. Ann Epidemiol. 1999;9:314–24. [PubMed]
36. Hu FB, Rimm E, Smith-Warner SA, Feskanich D, Stampfer MJ, Ascherio A, et al. Reproducibility and validity of dietary patterns assessed with a food-frequency questionnaire. Am J Clin Nutr. 1999;69:243–9. [PubMed]
37. Hebert JR, Ockene IS, Hurley TG, Luippold R, Well AD, Harmatz MG. Development and testing of a seven-day dietary recall. Dietary Assessment Working Group of the Worcester Area Trial for Counseling in Hyperlipidemia (WATCH) J Clin Epidemiol. 1997;50:925–37. [PubMed]
38. Hebert JR, Gupta PC, Bhonsle RB, Murti PR, Mehta H, Verghese F, et al. Development and testing of a quantitative food frequency questionnaire for use in Kerala, India. Public Health Nutr. 1998;1:123–30. [PubMed]
39. Hebert JR, Gupta PC, Bhonsle RB, Sinor PN, Mehta H, Mehta FS. Development and testing of a quantitative food frequency questionnaire for use in Gujarat, India. Public Health Nutr. 1999;2:39–50. [PubMed]
40. Wolever TM. Effect of blood sampling schedule and method of calculating the area under the curve on validity and precision of glycaemic index values. Br J Nutr. 2004;91:295–301. [PubMed]
41. Wolever TM, Jenkins DJ, Vuksan V, Josse RG, Wong GS, Jenkins AL. Glycemic index of foods in individual subjects. Diabetes Care. 1990;13:126–32. [PubMed]