To determine the correlation between obesity and food type intake, we first assessed the suitability of the BRFSS dataset for our analysis. Use of the NHANES obesity prevalence data was hampered by the availability of only five data points, NHANES (1999-2000), (2001-2002), (2003-2004), (2005-2006), and (2007-2008), in contrast to the 14 data points in the BRFSS dataset between 1995 and 2008, which is statistically more favorable for our study. However, the BRFSS dataset, which is based on self-reported weights and heights, is generally considered inferior to the NHANES obesity prevalence data [20
]. Despite the qualitative and quantitative differences between the NHANES and BRFSS data, our results show that the obesity trends between 1995 and 2008 derived from the two datasets were remarkably similar by regression analysis (). Therefore, we used the BRFSS data for our correlation study with the food trends data since it is statistically more robust than the NHANES dataset.
Fig. 1 Comparison of rising obesity trends of NHANES and BRFSS datasets. Alignment of the NHANES and BRFSS obesity trend datasets was performed to find the optimal correspondence. The left ordinate indicates the obesity prevalence by median %; and the right (more ...)
We analyzed the USDA ERS Loss-Adjusted Food Availability Data, which include seven major aggregated food groups including 1, meat, eggs, and nuts; 2, dairy; 3, fruit; 4, vegetables; 5, flour and cereal products; 6, added fats and oils, and dairy fats; and 7, caloric sweeteners. These groups are further comprised of more than 100 individual or specific food types (commodities). Analysis of these food types revealed that a large number of them including fresh vegetables, fresh fruit, beverage milk, fish and shellfish, fruit juice, nuts, and others, showed either negative trends or no change in trends of consumption and did not coincide with rising trends in obesity ().
Fig. 2 Food types showing no correlation with rising obesity trends. The left ordinate indicates the obesity prevalence by median %; and the right ordinate shows the average daily per capita calories consumed for each indicated food type. Rising obesity did (more ...)
Since energy imbalance resulting from excess calorie intake is thought to contribute to obesity, we first analyzed the trends in calorie intake between 1995 and 2008. The food availability data indicated that the average daily per capita total calorie intake has plateaued since year 2000, whereas obesity exhibited a rising trend (), and Pearson's analysis showed a correlation coefficient of 0.79 (). In contrast, strong positive correlations with obesity were unexpectedly found for chicken and corn products (), with Pearson's correlation coefficients of 0.96 and 0.99, respectively ().
Fig. 3 Regression analysis of the relationship between food type consumption and obesity trends by Pearson's correlation. Regression analysis by Pearson's correlation was performed to determine the relationship between obesity trends and the average daily per (more ...)
Correlation between trends in food type consumption and obesity
We also observed a positive correlation between total cheese intake and obesity (). However, further analysis revealed that, with the exception of cheddar and mozzarella cheese, most other cheeses, such as provolone, parmesan, Swiss cheese, blue cheese, and others, showed little or no changes in consumption trends between 1995 and 2008, and Pearson's analysis of either cheddar () or mozzarella () did not show correlation with rising obesity.
Even though correlation with obesity was not found for "Added Fats and Oils, and Dairy Fats" (), with a correlation coefficient of 0.86 (), analysis of Salad and Cooking Oils () and Dairy Fats () revealed correlation with obesity, each with a correlation coefficient of 0.97 (). These correlations subsequently did not cross-validate upon further analysis by multiple linear regression (see below).
Additionally, either poor or negative correlations were found for foods such as flour and cereal products, shortening, red meat, caloric sweeteners, and HFCS, with correlation coefficients of -0.03, -0.18, -0.40, -0.74, and -0.38, respectively (, and ). The consumption of refined cane and beet sugar () as well as sweet corn as a fresh vegetable () also did not correlate with obesity. The consumption of corn as a fresh vegetable constituted only a small percentage (averaging 0.01%) of the total calorie intake between 1995 and 2008.
To further test these positive correlations with obesity trends, we performed a fitting by multiple linear regression analysis with food types that showed correlation coefficients > 0.95, which included chicken, corn products, dairy fats, salad and cooking oils, and total cheese, in a full model function. This analysis showed that only corn products had p-values smaller than 0.05 (), suggesting that consumption of corn products had a significant effect on rising obesity trends. In the reduced model, we analyzed corn products and total cheese, which have p-values closest to 0.05 from the full model analysis, and our results confirmed a correlation between corn products, but not total cheese, and obesity trends ().
Multiple linear regression analysis of food types and obesity trends
The observed correlation between consumption of corn products and rising obesity is surprising. It is noteworthy that HFCS is classified separately as a caloric sweetener and not aggregated with other corn products. Moreover, HFCS showed a negative correlation with rising obesity (). We were not able to fully analyze whether or not corn product consumption correlated with obesity trends between 1970 and 1994 because the National Health and Nutrition Examination Survey (NHANES) datasets are only available in four cross-sectional, nationally representative surveys prior to 1995, including NHANES I (1971-1975), II (1976-1980), and III (1988-1994) [22
], thus yielding only three data points for Pearson's correlation analysis of corn-rich products. Nevertheless, we showed that the trends in obesity prevalence and corn product consumption between 1970 and 1994 did not align ().
Fig. 4 Trends in corn consumption and rising obesity. (A) Relationship of corn product consumption and obesity prevalence between 1970 and 1994. Obesity prevalence from NHANES I (1971-1975), II (1976-1980), and III (1988-1994) were plotted against corn product (more ...)
We were also aware that genetically modified (GM) corn has been planted in the U.S. since 1996 [23
]. To further investigate the relationship between bioengineered corn and rising obesity, we obtained data on the adoption of GM corn from the USDA, which covered the period between 2000 and 2008, for comparison with rising obesity. These data did not take into account the use of GM corn for other purposes besides as a food or animal feed. Despite this limitation, our result shows that the trends of obesity and adoption of GM corn were similar ().
We further asked whether or not the consumption of corn products might be associated with the demographic distribution of the population. Using the NHANES stratified obesity prevalence data between NHANES III (1988-1994), NHANES (1999-2000), (2001-2002), (2003-2004), (2005-2006), and (2007-2008), we examined the relationship between corn product consumption and race/ethnicity of men and women between 1995 and 2008. Our results show that the trends of obesity and corn product consumption rose in parallel irrespective of gender among non-Hispanic white men and women (), non-Hispanic black men and women (), and Mexican-American men and women (), thus suggesting that the association of rising obesity trends with increased corn product consumption is independent of race/ethnicity and gender.
Fig. 5 Correlation of corn products intake with NHANES obesity prevalence data stratified by race/ethnicity and gender. Obesity prevalence data stratified by race/ethnicity and gender from NHANES III (1988-1994), NHANES (1999-2000), (2001-2002), (2003-2004), (more ...)