Dietary glycemic index, glycemic load, and head and neck cancer risk: a pooled analysis in an international consortium

High dietary glycaemic index (GI) and glycaemic load (GL) may increase cancer risk. However, limited information was available on GI and/or GL and head and neck cancer (HNC) risk. We conducted a pooled analysis on 8 case-control studies (4081 HNC cases; 7407 controls) from the International Head and Neck Cancer Epidemiology (INHANCE) consortium. We estimated the odds ratios (ORs) and 95% confidence intervals (CIs) of HNC, and its subsites, from fixed- or mixed-effects logistic models including centre-specific quartiles of GI or GL. GI, but not GL, had a weak positive association with HNC (ORQ4 vs. Q1 = 1.16; 95% CI = 1.02–1.31). In subsites, we found a positive association between GI and laryngeal cancer (ORQ4 vs. Q1 = 1.60; 95% CI = 1.30–1.96) and an inverse association between GL and oropharyngeal cancer (ORQ4 vs. Q1 = 0.78; 95% CI = 0.63–0.97). This pooled analysis indicates a modest positive association between GI and HNC, mainly driven by laryngeal cancer.


BACKGROUND
Most head and neck cancers (HNCs) are attributed to tobacco smoking and/or alcohol drinking. 1 Diet has been suggested to play a role in HNC etiology, with non-starchy vegetables and selected healthy dietary patterns being inversely related with HNC risk. 2 2 2 Average daily glycemic index (GI) ranks carbohydrate foods based on the postprandial blood glucose response; average glycemic load (GL) estimates the impact of carbohydrate consumption using the GI, while taking into account the amount of carbohydrates that are consumed. 3 Higher GI and GL are moderately associated with risk of several cancers 4 , likely because of stimulation of insulin release and bioactivity of insulin-like growth factor-1, which has proliferative, angiogenic, anti-apoptotic, and estrogen stimulating properties. 5 Only two studies 6,7 have investigated the effect of GI and GL on HNC risk, with inconsistent findings; one of these studies 6 reported results by sub-site, based, however, on a limited number of cases.
The objective of this paper is to assess the association of GI or GL with HNC and its subsites (i.e., oral cavity, oropharynx, hypopharynx, and larynx) using pooled dietary data from eight case-control studies participating in the International Head and Neck Cancer Epidemiology (INHANCE) consortium. 8

METHODS
Within data version 1.5 of the INHANCE dataset, information on GI and GL was available from 3 case-control studies. In addition, we calculated GI and/or GL intakes from studyspecific food items and food composition databases for another 5 studies, giving a total of 8 studies included in the analysis. Details on individual studies and data pooling methods have been previously described 8 and are summarized in Supplementary Table   S1. Informed consents and institutional review board approvals were obtained within the framework of the original studies.

Selection of subjects
Cases were included if their cancer had been originally classified as invasive cancer of the oral cavity, pharynx, larynx, or unspecified oral cavity/pharynx. Corresponding controls from the original studies were included in the analysis. We excluded subjects with missing information on the site of origin of cancer, or GI or GL value, and those with missing or implausible (<500 or >5,500 kcal/day) non-alcohol energy intake. Thus, our analysis included 11,488 subjects, with 4,081 HNC cases and 7,407 controls (4,264 hospital-based and 3,143 population-based controls). There were 810 oral cavity, 1,172 oropharynx, 343 hypopharynx, 1,338 larynx, and 418 unspecified oral cavity/pharynx cancer cases.

Specification of variables
Study-specific food-frequency questionnaires (FFQs) and food composition tables allowed us to calculate individual values of GI and GL for the 4 studies lacking information on both the exposures [Los Angeles, Boston, Seattle (1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995) and Memorial Sloan Kettering Cancer Center (MSKCC) studies]. In detail, as described previously 9 , the GI of a food was expressed as a percentage of the glycemic response elicited by white bread as a standard food with a GI of 100. The average daily GI for each subject was computed by summing the products of the GI value of each food times the amount of available carbohydrates in that food consumed per day, divided by the total amount of available carbohydrates (g) consumed per day. The average daily GL (g) was calculated by summing the products of the GI value of each food times the amount of available carbohydrates in that food consumed per day, divided by 100. Each GL unit represents the equivalent of 1 g of carbohydrate from white bread. Therefore, we preliminary converted frequencies of consumption into servings/day and servings/day into grams/day; then, we assigned the corresponding GI to each food item and applied the previous  -text and   Table S2 for GI/GL calculation and study-specific GI values).

Statistical analysis
Multiple logistic regression models were used to estimate the odds ratios (ORs) of HNC and the corresponding 95% confidence intervals (CIs) according to center-specific quartiles of GI or GL among controls (Supplementary Tables S3 for descriptive statistics of GI and GL distributions). In the presence of heterogeneity of GI or GL intakes across centers, we used a random-slope logistic regression model, whereas a fixed-effects model was used otherwise. 10 The models included the following potential confounders: age, sex, race/ethnicity, study center, education, cigarette smoking intensity, cigarette smoking duration, cigar smoking status, pipe smoking status, alcohol drinking intensity, and the product term of cigarette smoking and alcohol drinking intensities. For GI, models were further adjusted for energy intake without alcohol; for GL, models were further adjusted for energy intake without alcohol and carbohydrates. For both GI and GL models, we used center-specific control-based quartiles of energy intake. Separate analyses were carried out by HNC sub-sites and in strata of selected covariates. In sensitivity analyses, we further adjusted for history of diabetes or excluded subjects with diabetes (information available for 6 studies). Analyses were performed using the SAS software (version 9.4, SAS Institute, Cary, NC). Supplementary Table S4

DISCUSSION
In this large dataset, we observed a positive association between GI and HNC risk, essentially driven by laryngeal cancer. GL was not associated with the risk of overall HNC or its sub-sites, except for a possible inverse association with oropharyngeal cancer.
Inconsistent associations of GI and GL with HNC risk may be partly due to differences in the underlying dietary patterns. Indeed, higher dietary GL is strongly associated with higher carbohydrate intakes, while a higher GI is also associated with lower intakes of dairy products, legumes, fruit and vegetables. 11 In line with this hypothesis, an overlapping INHANCE-based analysis including 7 of the 8 current studies showed a positive association of laryngeal cancer with an "Animal products and cereals" dietary pattern, which was simultaneously based on high-GI (e.g. cereals) and low-GL (e.g. meat) foods. 10 Only two previous studies 6,7 have examined the association between GI or GL and HNC risk, with one of them partially overlapping with the current dataset. 6 An analysis 6 of three Italian case-control studies on upper aero-digestive tract cancers reported a positive association with higher GI (ORQ5 vs. Q1=1.5; 95% CI=1.1-2.0) and GL (ORQ5 vs. Q1=1.8; 95% CI=1.1-2.9) in quintiles. Although in the same direction, the association was weaker with oral and pharyngeal cancers combined or laryngeal cancer. 6  Limitations of the current analyses included possible recall bias and non-differential misclassification of GI/GL quartiles. In addition, food items contributing to GI differed in part across regions (Supplementary Table S2). However, all our FFQs were either reproducible and valid or were modifications of existing FFQs, already tested for reproducibility and validity. We were able to adjust for major potential confounders and our large sample size provided the necessary statistical power to examine the association in HNC sub-sites and strata. 8 In conclusion, findings from this large-scale pooled analysis support a positive effect of average daily GI on the risk of HNC, and in particular of laryngeal cancer.

ADDITIONAL INFORMATION
Ethical approval and consent to participate: The Informed consent and institutional review board approval were obtained within the framework of the original studies, according to the rules existing at the time of data collection. In addition, a central Institutional Review Board approval was obtained from the University of Utah, #42912.

Consent to publish: Not applicable.
Data availability: The dataset used and analyzed during the current study is available from the corresponding author on reasonable request.

Conflicts of interest:
The authors declare no competing interests.  Models adjusted for age, sex, race/ethnicity, study center, education level, center-specific control-based quartiles of energy intake (without alcohol for glycemic index; without alcohol and carbohydrate for glycemic load), cigarette smoking intensity (number of cigarettes per day), cigarette smoking duration, cigar smoking status, pipe smoking status, alcohol drinking intensity (number of drinks per day), and the product (interaction) term for cigarette smoking intensity and alcohol drinking intensity. b. The number of controls differed across sub-sites because a few studies considered cancers of the oral cavity, oropharynx, and hypopharynx only; therefore, they contributed to the analysis with fewer controls than those studies with all cancer sub-sites included (see Supplementary Table 4). c. P-value for heterogeneity between study centers. d. Based on the likelihood ratio test of heterogeneity between study centers, we reported the fixed-effects estimates when Pheterogeneity > 0.1 and the mixed-effects estimates when Pstudies < 0.1.  0.770 0.432 a. Adjusted for age, sex, race/ethnicity, study center, education levels, energy intake (without alcohol for glycemic index; without alcohol and carbohydrate for glycemic load), cigarette smoking intensity (number of cigarettes per day), cigarette smoking duration, cigar smoking status, pipe smoking status, alcohol drinking intensity (number of drinks per day), and the product (interaction) term for cigarette smoking and alcohol drinking, when appropriate. b. P for heterogeneity across strata. c. P for