Logistic LASSO Regression for Dietary Intakes and Obesity: NHANES (2007-2016)

doi:10.21203/rs.3.rs-2074608/v1

Download PDF

Research Article

Logistic LASSO Regression for Dietary Intakes and Obesity: NHANES (2007-2016)

https://doi.org/10.21203/rs.3.rs-2074608/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background: Many investigations have explored the relationship between dietary intake and obesity risk/incidence, but they have only assessed obesity-related dietary patterns and/or single nutrients, not taking into account the diversity of dietary variables or multicollinearity among multiple nutrients. Therefore, this study aimed to use logistic LASSO regression combined with logistic regression to overcome multicollinearity, and overall investigate the possible association between dietary factors and obesity by the National Health and Nutrition Examination Survey (NHANES) (2007-2016).

Method: Logistic LASSO regression was performed to examine the relationship between 59 dietary variables, and subsequently identify the most relevant variables associated with obesity. Then we used logistic regression to test the relevant variables, and finally used the receiver operating characteristic curve (ROC) to test the effect of logistic regression. Using the methods described above, we explored the relationship between dietary intake and obesity in 12135 NHANES participants (2007-2016).

Result: The set of factors screened by LASSO regression, obesity risk factors including cholesterol and PFUA 20:4 (β>0); protective factors including vitamin E, caffeine, folate, vitamin C and copper (β<0). Ultimately, after multivariate unadjusted and adjusted logistic regression tests as well as ROC tests, four in the factor set associated with obesity were selected. Statistically significant dietary factors only folate (OR=0.80), vitamin C (OR=0.95), copper (OR=0.60) and PFUA 20:4 (OR=2.57) in adjusted logistic regression.

Conclusion: Dietary intake of folate, vitamin C and copper negatively correlated with obesity, but PFUA 20:4 positively inversely. Necessary to assess the effective levels of folate, vitamin C and copper supplementation in obese subjects.

Obesity

NHANES

logistic LASSO regression

Dietary Intake

Multicollinearity

Obesity is currently a global pandemic and its prevalence increasing. Obesity is both an independent disease and a risk factor for multiple diseases including type 2 diabetes, cardiovascular disease, hypertension, etc.[1]. The Global Burden of Disease Study lists high body weight as one of the top ten risk factors for disease. According to the International Agency for Research on Cance, overweight, obesity and weight gain are associated with an increased risk of developing at least 13 types of cancer[2].

Numerous studies have explored the relationship between dietary intake and obesity risk/incidence. It is currently believed that obesity is mainly caused by an imbalance between caloric intake and energy expenditure[3]. There are also studies showing that food source and quality of nutrients are more important than quantity in the diet for obesity. However, due to the diversity of dietary variables, the macro includes the entire dietary pattern; food types include vegetables, fruits, fiber, fat, processed meat, sugary foods, lean meat and so on; and the micro includes macronutrients and micronutrients. Most studies assessed dietary patterns and/or single nutrients in association with disease outcomes[4]. In addition, multicollinearity among multiple nutrients may also pose statistical challenges, particularly limiting the number and types of dietary variables included in the model, potentially leading to conflicting findings and/or invalid results[5, 6]. Therefore, when studying the relationship between dietary factors and obesity, the combined effect of total dietary intake should be investigated. At the same time, factors such as demographic data (age, gender, ethnicity and pregnancy), socioeconomic status (family income and education), and personal behavior (smoking, alcohol use and sedentary time) must also be considered, as these all affect food intake, nutrient application[3, 7].

Since its inception in 1999, the National Health and Nutrition Examination Survey (NHANES) has collected dietary, lifestyle, laboratory, physical examination, and other health data on nearly 5,000 nationally representative participants covering a wide range of health conditions variables[8]. Thanks to the availability of large amounts of diet and health data, combined with recent statistical techniques, we have emerging opportunities to elucidate the associations between diet and obesity that were previously undetected due to small data volumes or traditional statistical methods[9]. This helps to better study the relationship between diet and obesity.

To date, traditional statistical methods have been utilized to analyze the majority of data on diet and obesity outcome, but they are limited in their ability to analyze and integrate a large number of factors and cannot deal with multicollinearity. LASSO regression can effectively overcome the multicollinearity of a large number of covariates[10]. It is a shrinkage method inspired by stepwise selection that actively selects from a vast and potentially multicollinear set of variables, resulting in a more pertinent and interpretable set of predictors[11].To prevent overfitting and achieve variable filtering, LASSO regression compresses some coefficients while making others zero, through a series of shrinkage operations. This automatically removes unnecessary or non-influencing covariates and selects non-zero variables to remain in the model[12]. Therefore, in this study, we aimed to overall investigate the relationship between dietary intake and obesity and overcome multicollinearity by using logistic LASSO regression and logistic regression, through the large database NHANES.

Sample

Data for the study were obtained from the 2007–2016 NHANES, which is an ongoing cross-sectional survey conducted by the National Center for Health Statistics at the Centers for Disease Control and Prevention (CDC) to assess the health and nutritional status of adults and children in the United States. The sample was selected by a complex, multistage, stratified, clustered probabilistic design, and the data included interviews and physical examinations.

The 2007–2016 cycle of NHANES included a nationally representative population of 50,588 persons, including 46,976 persons with physical activity data, 29,040 persons with alcohol consumption questionnaire data, 35,632 persons with smoking questionnaire data, and 48,710 persons examining and dietary data.

Weighted according to the method recommended by the CDC, which was detailed in the Statistical Analysis section. After weighing, adults aged 19–60 years were selected, Body mass index (BMI (kg/m²)) less than 18.5 kg/m², pregnancies, and individuals without complete data were excluded. Ultimately the sample was 12,135 people, including 3,670 normal people and 8,465 obese or overweight people, representing 115,827,669, 35,521,749 and 80,305,920 people. (Fig. 1)

BMI and Dietary Intake Assessment

BMI was derived from the body measurement data set and collected by trained health technicians, assisted by a recorder during the physical examination. Standard BMI categories defined by the CDC are normal (18.5–24.9 kg/m²), overweight (25.0-29.9 kg/m²), and obese (30.0 kg/m²). We used BMI 25 kg/m²as the dividing line, with BMI greater than or equal to 25 as overweight and obesity (hereinafter referred to as obesity), and BMI less than 25 kg/m² as normal, excluding people with BMI less than 18.5 kg/m².

Dietary intakes were derived from the first total nutrient intakes dataset, which was collected in person at the Mobile Examination Center, in order to estimate the types and amounts of food consumed 24 hours before the interview (from midnight to midnight), as well as the energy, nutrients, and other food components consumed from these foods and beverages. The United States Department of Agriculture (USDA) Food and Nutrition Dietary Research Database (FNDDS) was used to code interview data and convert them to total nutrient intakes for NHANES, and FNDDS nutritional values were based on the USDA National Standard Reference Nutrition Database.

Covariate Assessment

Age, gender, ethnicity, pregnancy, family income and education were obtained from the family and sample person demographics questionnaire. Smoking status was determined by the question from the smoking - cigarette use questionnaire, “These next questions are about cigarette smoking and other tobacco use. (Have you/Has SP) smoked at least 100 cigarettes in (your/his/her) entire life?”. Alcohol use was identified by the following question,” Had at least 12 alcohol drinks/1 year?” Sedentary time via the following question from the physical activity questionnaire “The following question is about sitting or reclining at work, at home, or school. Include time spent sitting at a desk, sitting with friends, traveling in a car, bus, or train, reading, playing cards, watching television, or using a computer. Do not include time spent sleeping. How much time (do you/does) usually spend sitting or reclining on a typical day?”.

Statistical Analysis

Statistical analysis was completed with R Statistical Software (version 4.2.0). The “survey” package (version 4.1-1) in R was utilized for the data weighting, univariate analyses and multivariate analysis accounting for the fact that the NHANES used a complex, multistage probability sampling design to select participants who were representative of the civilian, non-institutionalized U.S. population and to oversampling population subgroups to improve the reliability and precision of estimates of health-status measures for those specific subgroups. Specifically, we utilized the sampling weights provided by the NHANES for dietary analysis and calculated based on the samples we used that combined five survey cycles. We used the “survey” package combined with the “gtsummary” package (version 1.6.1) to carry out univariate analysis and multivariate analysis and made tables. All statistical tests were two-sided with a significance level of 0.05.

We used the "glmnet" package (version 4.1-4) to perform logistic LASSO regression. Referring to previous studies[6], our analysis was performed on a logarithmic scale of covariates that minimized the range of covariate values. Thus, in the model no covariate dominates, and all covariates have equal status. We included 59 dietary variables in the model as continuous variables. We utilized the ten-fold cross-validation to select the penalty term λ. The built-in function in R automatically generates two λ (λ. min and λ.1se), one λ corresponding to the smallest binomial deviation and the other λ corresponding to one standard error greater than the least binomial deviation.

Then, we use the “survey” package to complete the Logistic regression model for the association between dietary factor sets of LASSO screening and obesity. Model 1 included the set of dietary factors for obesity. Model 2 adjusted for age, gender, ethnicity, education, family income, smoking status, alcohol use and sedentary time. After that, we used the "pROC" (version 1.18.0) package to test the effect of the logistic regression model.

Sample characteristics

Demographic data and well-established obesity risk factor data (numbers (%) or mean (SD), respectively) are shown in Table 1. Significant differences (P < 0.005) between obese adults and non-obese adults were observed for age 38.6 vs. 34.0 years, gender, ethnicity, education, family income and sedentary time (380.2 min vs. 355.4 min, respectively). Obese adults were more likely to be older; male (68.4% vs. 56.7%, respectively); Mexican American (12.1% vs. 7.0%, respectively), or non-Hispanic black (12.4% vs. 10.0%, respectively); high school graduate (22.9% vs. 19.2%, respectively) or less than high school (15.3% vs. 13.5%, respectively); family income $20000-$75000 (48.3% vs. 45.4%, respectively) or >$75000 (33.9% vs. 33.5%, respectively) and sedentary time longer. However, there is no significant difference between smoking and alcohol use in the obese population and the normal population (p > 0.05).

Table 1

Baseline characteristics in participants with and without obesity.
Characteristic	Overall, N = 115,827,669¹	Normal, N = 35,521,749¹	Obesity, N = 80,305,920¹	p-value²
Age	37.2 (11.28)	34.0 (11.06)	38.6 (11.09)	< 0.001
Gender				< 0.001
Female	4,430 (35.2%)	1,518 (43.3%)	2,912 (31.6%)
Male	7,705 (64.8%)	2,152 (56.7%)	5,553 (68.4%)
Ethnicity				< 0.001
Mexican American	2,054 (10.5%)	407 (7.0%)	1,647 (12.1%)
Non-Hispanic Black	2,517 (11.7%)	652 (10.0%)	1,865 (12.4%)
Non-Hispanic White	4,946 (63.9%)	1,601 (66.2%)	3,345 (62.9%)
Other race	2,618 (13.9%)	1,010 (16.7%)	1,608 (12.6%)
Education				< 0.001
College education	6,827 (63.5%)	2,221 (67.3%)	4,606 (61.8%)
High school graduate	2,757 (21.8%)	789 (19.2%)	1,968 (22.9%)
Less than high school	2,551 (14.8%)	660 (13.5%)	1,891 (15.3%)
Family income				0.048
<$20000	2,829 (18.9%)	929 (21.1%)	1,900 (17.8%)
$20000-$75000	6,155 (47.4%)	1,748 (45.4%)	4,407 (48.3%)
>$75000	3,151 (33.8%)	993 (33.5%)	2,158 (33.9%)
Smoking	5,190 (43.6%)	1,548 (42.1%)	3,642 (44.2%)	0.226
Alcohol use	9,618 (83.2%)	2,913 (84.0%)	6,705 (82.9%)	0.206
Sedentary time (min)	372.6 (205.18)	355.4 (200.18)	380.2 (206.91)	< 0.001
¹Mean (SD) (weighted); n (unweighted) (%(weighted)).
²T-test for continuous variables; chi-squared test with Rao & Scott's second-order correction for discrete variables

Dietary intake

Univariate results showed that 29 dietary factors in dietary intake were associated with BMI (p < 0.05), 21 of dietary intake factors were positively associated with obesity, and 8 conversely. Furthermore, the percentage of energy supplied by protein (protein, % energy), the percentage of energy supplied by fat (fat, % energy), cholesterol, caffeine, total choline, phosphorus, zinc, sodium, selenium, moisture, SFA, MUFA, SFA14:0, SFA16:0, SFA18:0, MFUA16:1, MFUA18:1, MFUA20:1, PFUA18:4, PFUA20:4 and PFUA22:5 among obese adults were higher than the normal population. The intake levels of energy supplied by carbohydrate (carbohydrate, % energy), fiber, vitamin E, beta-carotene, folate, vitamin C, copper and caffeine in obese adults were lower than those of the normal population (Table 2).

Table 2

Dietary intake of adults with and without obesity
Characteristic	Normal, N = 35,521,749¹	Obesity, N = 80,305,920¹	p-value²
Energy (kcal)	2,357.02 (1,081.11)	2,378.52 (1,042.99)	0.172
Protein, % energy	15% (5%)	16% (6%)	0.001
Carbohydrate, % energy	49% (11%)	48% (12%)	< 0.001
Fat, % energy	33% (9%)	34% (9%)	< 0.001
Sugar (g)	126.03 (87.82)	123.95 (83.78)	0.683
Fiber (g)	18.28 (11.88)	17.45 (10.98)	0.019
Cholesterol (mg)	293.24 (247.67)	330.32 (262.98)	< 0.001
Vitamin E (mg)	9.67 (8.11)	9.06 (7.17)	0.068
Retinol (µg)	446.43 (405.61)	441.68 (449.51)	0.453
Vitamin A (RE)	658.02 (729.48)	630.31 (628.18)	0.062
Alpha-carotene (µg)	440.23 (1,888.89)	384.79 (1,414.66)	0.074
Beta-carotene (µg)	2,279.51 (6,387.08)	2,038.86 (4,410.91)	0.005
Beta-cryptoxanthin (µg)	87.30 (214.00)	73.98 (144.64)	0.378
Lycopene (µg)	6,129.88 (10,579.55)	5,690.52 (9,606.70)	0.182
Lutein-zeaxanthin (µg)	1,709.66 (4,263.43)	1,484.19 (3,281.70)	0.173
Vitamin B1 (mg)	1.80 (1.18)	1.73 (0.95)	0.307
Vitamin B2 (mg)	2.31 (1.45)	2.34 (1.46)	0.503
Niacin (mg)	28.88 (17.09)	29.31 (17.46)	0.308
Vitamin B6 (mg)	2.39 (1.81)	2.35 (1.89)	0.675
Caffeine (mg)	160.65 (220.14)	184.23 (229.35)	< 0.001
Folate (µg)	460.35 (307.86)	425.98 (260.95)	< 0.001
Total choline (mg)	351.36 (209.98)	374.27 (225.80)	< 0.001
Vitamin B12 (µg)	6.88 (8.22)	6.87 (7.91)	0.345
Vitamin C (mg)	90.84 (111.62)	80.49 (94.33)	< 0.001
Vitamin D (µg)	4.77 (5.50)	4.95 (6.17)	0.279
Vitamin K (µg)	128.74 (431.24)	109.78 (159.38)	0.106
Calcium (mg)	1,050.67 (671.20)	1,049.24 (649.45)	0.572
Phosphorus (mg)	1,489.34 (748.82)	1,540.93 (759.00)	0.002
329.52 (173.95)	323.86 (164.12)	0.384
Iron (mg)	16.16 (9.74)	15.73 (9.01)	0.370
Zinc (mg)	12.26 (7.48)	12.88 (9.30)	0.008
Copper (mg)	1.40 (0.85)	1.35 (0.82)	0.031
Sodium (mg)	3,876.33 (2,005.38)	3,957.41 (1,939.44)	0.021
Potassium (mg)	2,817.02 (1,425.25)	2,837.61 (1,357.67)	0.316
Selenium (mg)	123.04 (69.89)	127.70 (70.57)	0.004
Theobromine (mg)	40.94 (85.23)	38.24 (77.72)	0.050
Moisture (g)	3,175.99 (1,630.15)	3,380.78 (1,710.51)	< 0.001
Water (g)	1,201.60 (1,251.10)	1,283.15 (1,351.39)	0.115
SFA (g)	28.59 (18.22)	29.88 (18.08)	< 0.001
MUFA (g)	31.22 (18.78)	32.66 (18.98)	0.001
PUFA (g)	19.90 (12.85)	20.45 (13.35)	0.161
SFA 4:0 (g)	0.59 (0.59)	0.59 (0.58)	0.554
SFA 6:0 (g)	0.33 (0.34)	0.34 (0.33)	0.470
SFA 8:0 (g)	0.28 (0.31)	0.28 (0.29)	0.384
SFA 10:0 (g)	0.53 (0.50)	0.53 (0.48)	0.291
SFA 12:0 (g)	0.89 (1.44)	0.87 (1.30)	0.636
SFA 14:0 (g)	2.45 (2.07)	2.51 (2.06)	0.049
SFA 16:0 (g)	15.47 (9.44)	16.22 (9.45)	< 0.001
SFA 18:0 (g)	7.01 (4.58)	7.46 (4.56)	< 0.001
MFUA 16:1 (g)	1.25 (0.99)	1.35 (0.94)	< 0.001
MFUA 18:1 (g)	28.93 (17.57)	30.20 (17.80)	0.003
MFUA 20:1 (g)	0.31 (0.27)	0.33 (0.30)	0.001
MFUA 22:1 (g)	0.03 (0.10)	0.03 (0.09)	0.076
PFUA 18:2 (g)	17.60 (11.55)	18.11 (12.03)	0.153
PFUA 18:3 (g)	1.80 (1.32)	1.81 (1.35)	0.542
PFUA 18:4 (g)	0.01 (0.04)	0.02 (0.04)	0.003
PFUA 20:4 (g)	0.15 (0.14)	0.17 (0.15)	< 0.001
PFUA 20:5 (g)	0.03 (0.12)	0.03 (0.11)	0.087
PFUA 22:5 (g)	0.02 (0.04)	0.03 (0.04)	< 0.001
¹Mean (SD) (weighted).²T-test for continuous variables.
Statistical descriptions were performed on raw data, and statistical analyses were performed on a logarithmic scale. SFA, saturated fatty acids. MUFA, monounsaturated fatty acids. PUFA, polyunsaturated fatty acids.

Logistic LASSO regression to screen dietary obesity factor set

The logistic LASSO model, a shrinkage method, actively selected from a large and potentially multicollinear set of 59 dietary macro and micronutrient intake variables in the regression, resulting in a more relevant and interpretable set of predictors. As λ increases, the penalty is more stringent, and there are constantly covariate coefficients of 0 that are screened out, so we choose the larger λ, and λ. min was 0.001805283 and λ.1se was 0.01534046 (Fig. 2A) in our study. Figure 2B shows the results for variables with non-zero coefficients in LASSO regression at λ. min (0.001805283) and their corresponding coefficients for different values of the penalty parameter. When λ gradually increased and equal to λ.1se, only cholesterol (β = 0.022), vitamin E (β =-0.001), caffeine (β =-0.056), folate (β =-0.007), vitamin C (β =-0.016), copper (β =-0.013) and PFUA 20:4 (β = 0.306)remain in the model (Table 3).

Table 3

The estimated coefficients for logistic Lasso regression, and multivariate logistic regression models for the association between dietary factor sets of LASSO screening and obesity
Variables	Coefficients¹	Unadjusted OR² (95%CI)	p-value	Adjusted OR³ (95%CI)	p-value
Cholesterol (mg)	0.022	1.11 (1.00-1.22)	0.040	1.09 (0.99–1.2)	0.644
Vitamin E (mg)	-0.001	0.92 (0.83–1.04)	0.175	0.93 (0.82–1.05)	0.241
Caffeine (mg)	-0.056	1.03 (1.00-1.06)	0.037	1.00 (0.97–1.04)	0.796
Folate (µg)	-0.007	0.84 (0.76, 0.92)	0.001	0.80 (0.72–0.9)	< 0.001
Vitamin C	-0.016	0.96 (0.92–1.01)	0.086	0.95 (0.9–0.99)	0.031
Copper (mg)	-0.013	0.69 (0.52–0.92)	0.012	0.60 (0.44–0.8)	0.001
PFUA 20:4 (g)	0.306	3.46 (1.65–7.24)	< 0.001	2.57 (1.19–5.54)	0.017
OR = Odds Ratio; CI = Confidence Interval; PFUA 20:4, octadecatetraenoic.
¹When λ = λ.1se (0.01534046), the estimated coefficients for logistic Lasso regression, variables with coefficients of zero are not shown.
²Modle1.Unadjusted survey-weighted logistic regression model for the association between dietary factor sets of LASSO screening and obesity.
³Model2. Adjusted for age, gender, ethnicity, education, family income, smoking status, alcohol use status and sedentary time.

Logistic regression and ROC tests uesd to validate the obesity factor set

Model 1 was unadjusted survey-weighted logistic regression model for the association between dietary factor sets of LASSO screening and obesity. And model 2 was logistic regression adjusted for age, gender, ethnicity, education, family income, smoking status, alcohol use status and sedentary time based on Model 1. We compared the two models with the ROC test and found that the area under the curve (AUC) of modle1 was smaller compared with modle 2 (p < 0.05). That is to say, adjustment of demographic information and individual behavioral information is necessary to analyse the differences of between dietary intake and obesity in our study (Fig. 3). Through the above analysis, we finally found that folate (OR = 0.80), vitamin C (OR = 0.95) and copper (OR = 0.60) are protective factors for obesity, PFUA 20:4 (OR = 2.75) risk factor inversely. In addition, cholesterol, vitamin E and caffein initially statistically significant in the LASSO regression, were excluded because of no statistical significance in adjusted survey-weighted logistic regression (P > 0.05) (Table 3).

The powerful shrinkage technique, LASSO regression has been applied to analyze large databases with large amounts of dietary data like NHANES[6, 13]. But to our knowledge, this is the first study to use LASSO regression to assess the relationship between numerous dietary variables and obesity. Initial univariate results indicated that 29 out of 59 dietary factors were associated with obesity; then robust LASSO regression shrinkage to minimize multicollinearity between variables, 59 dietary factors were investigated as a whole, and 7 dietary factors were selected as the most associated with obesity. Afterward, by a series of logistic regression and ROC tests, folate, vitamin C, copper and PFUA 20:4 retained, where folate, vitamin C and copper were protective factors for obesity, and PFUA 20:4 was the opposite. The pathogenesis of obesity is more complex than overeating and a sedentary lifestyle, including a complex interplay between behavioral, environmental, genetic, physiological, social and economic factors. Finally, we retained the 4 adiposity dietary factors that were significant after adjustment for age, gender, ethnicity, family income, education, smoking, alcohol use and sedentary time.

Compared with normal-weight individuals, overweight and obese individuals are more vulnerable to vitamin deficiency due to differences in dietary intake of fruits, vegetables and energy[14, 15].Preliminary population studies have demonstrated that people with obesity have insufficient amounts of folate, which may be related to the volume dilution of blood in obese subjects and/or low folate intake in obese people[16]. Our research once again found that folate intake was lower in obese adults than normal adults. To make matters worse, vitamin deficiency, especially folate, may aggravate the pathological state of obesity[17]. Lower serum folate levels were linked to higher body fat percentages. In a population study, BMI higher than 50 kg/m2 was mainly associated with folate deficiency[18]. The development of obesity is often accompanied by insulin resistance, chronic inflammation, endothelial dysfunction and oxidative stress[19, 20]. Folate deficiency may exacerbate obesity in different ways. In vitro tests have shown that the main circulating metabolite of folate (5-methyltetrahydrofolate) can increase the production of nitric oxide, the main endogenous relaxation factor of the endothelium, which can scavenge superoxide free radicals and help in the management of progressive endothelial damage caused by obesity[21]. Studies have shown that folate supplementation can improve insulin resistance associated with obesity, the possible mechanism may be inhibition of β-subunit insulin receptor and its substrates by the active homocysteine, and reduce insulin stimulates glycogen synthesis[22].

Like previous studies, we found that obese people had lower vitamin C intake than normal people[15]. Furthermore, we found that vitamin C is a protective factor for obesity. And there are many studies reported the role of vitamin C in anti-obesity. Vitamin C is necessary for the post-translational modification of lipoprotein peptides to create collagen molecules that are cross-linked and stretchy. Defective collagen synthesis decreases exercise motivation resulting Obesity[23]. Additionally, vitamin C is necessary for the manufacture of carnitine, which is in charge of moving long-chain fatty acids across the mitochondrial membrane for -oxidation and subsequent fat oxidation[24]. Furthermore, carnitine transport in muscle depends on γ-butyrobetaine, which is synthesized by a vitamin C-dependent process. Thus, β-oxidation is reduced when vitamin C is insufficient. Johnston demonstrated this by showing that individuals with moderate vitamin C levels had elevated (30%) oxidized lipids compared with overweight or obese adults[25]. Leptin expression in both humans and animals has been demonstrated to be reduced or inhibited by vitamin C[20]. In addition, vitamin C modifies the expression of several important obesity-related proteins. At the same time, Vitamin C supplementation reduced C-reactive protein levels in 75% of obese subjects and could ameliorate endothelial dysfunction caused by oxidative stress[26].

Copper is a component of copper metalloenzymes, which are necessary for the healthy operation of the brain system, blood vessels, bones, and hematological[27]. This finding was consistent with previous findings of inadequate intake of minerals such as copper, in obese people, which were lower than in normal people. In addition, previous studies have shown that dietary copper intake is associated with a higher risk of obesity, and higher serum copper levels are associated with obesity risk in children and adults[28]. Our results, on the contrary, suggested that low levels of copper intake may be associated with obesity, and the reason may be that copper plays a crucial role in the regulation of enzymes, thereby catalyzing antioxidant metabolism, glucose metabolism, pancreatic β-cell function and insulin signaling[29]. Copper may be functionally related to fat accumulation [30].

Our study found that the intake of octadecatetraenoic (PFUA 20:4) in the obese population was higher than that in the normal population, which may be a risk factor for obesity. Animal studies have found that PFUA 20:4 can exacerbate diet-induced obesity. PFUA 20:4 metabolites are important factors in the initiation and resolution of inflammation and have been implicated in the pathophysiology of obesity[31]. Previous studies have found that BMI is positively correlated with pro-inflammatory 5-and 11-hydroxyeicosaenoic acid (HETE), PFUA 20:4 derivatives. 5-HETE can increase TNF-α-induced hepatocyte apoptosis, and inhibition of 5-HETE can reduce hepatic macrophage infiltration in HFD obese mice. Increased proinflammatory 5-HETE May contribute to pathological changes in obese tissues, thereby promoting oxidative stress and increasing ROS, thereby affecting PUFA peroxidation. 11-HETE is a non-enzymatic PFUA 20:4-derived oxygen lipid, which is a marker of lipid peroxidation, and 11-HETE may also be a pathological biomarker associated with obesity. However, anti-inflammatory PFUA 20:4 metabolites, such as lipoxinA4 (LXA4), attenuate proinflammatory M1 macrophages and boost anti-inflammatory M2 macrophages, which diminish obesity-induced adipose inflammation[32, 33]. PFUA 20:4 metabolism is one of the most complex regulatory systems in humans due to the enormous variety of PFUA 20:4 derivatives and their diverse ramifications, as well as the varied expression and activity of PFUA 20:4 metabolites in different cells and organs and their cross-cellular biogenesis[31]. Therefore, more research is needed to better understand the complexity of eicosanoid metabolism, how PFUA 20:4 influences obesity and may help improve current prevention and treatment strategies for obesity.

Limitations: We found that the dietary intake of folate, vitamin C and copper in obese people are less than that in the normal population, and they are protective factors for obesity, therefore necessary to assess the effective levels them supplementation in obese subjects, inversely, PFUA 20:4 is a risk factor for obesity. But further prospective studies and intervention experiments are needed to identify the underlying mechanisms of the association of dietary factors with obesity and to identify effective supplemental doses, as NHANES is a cross-sectional survey. In addition, the ROC test found that the AUC (0.579, 0.656) of the model was less than 0.75, so the use of dietary factors of intake for predicting obesity was limited.

AUC, The area under the curve

BMI, Body mass index

CDC, The National Center for Health Statistics at the Centers for Disease Control and Prevention

CI, Confidence interval

FNDDS, Food and Nutrition Dietary Research Database

MUFA, Monounsaturated fatty acids

NHANES， The National Health and Nutrition Examination Survey

OR, Odds ratio

PUFA, Polyunsaturated fatty acids

ROC， Receiver operating characteristic curve

SFA, Saturated fatty acids

USDA, The United States Department of Agriculture

Ethics approval and consent to participate

This research analyzed de-identifed information downloaded from the National Health and Nutrition Examination Survey public database, The National Center for Health Statistics Ethics Review Committee granted ethics approval. All participants provided written informed consent before participating in the study.

Consent for publication

Not applicable

Availability of data and materials

The datasets generated and/or analysed during the current study are available in the National Health and Nutrition Examination Survey public database (http://www.cdc.gov/nchs/nhanes.htm)

Competing interests

The authors declare they have no actual or potential competing conflict of interest.

Funding

This study was supported by the Applied Technology Research and Development Plan of Heilongjiang Province (GA20C012) and the National Natural Science Foundation of China (81973035).

Authors' contributions

QN designed the study. QN and YD wrote the manuscript. QN, RW, XTand XL collected, analyzed and interpreted the data. YN critically reviewed the manuscript. All authors read and approved the final manuscript.

Acknowledgements

Not applicable.

Ward ZJ, Bleich SN, Cradock AL, Barrett JL, Giles CM, Flax C, Long MW, Gortmaker SL, Projected US. State-Level Prevalence of Adult Obesity and Severe Obesity. N Engl J Med. 2019;381:2440–50.
Steele CB, Thomas CC, Henley SJ, Massetti GM, Galuska DA, Agurs-Collins T, Puckett M, Richardson LC. Vital Signs: Trends in Incidence of Cancers Associated with Overweight and Obesity - United States, 2005–2014, MMWR. Morbidity and mortality weekly report, 66 (2017) 1052–1058.
González-Muniesa P, Mártinez-González M-A, Hu FB, Després J-P, Matsuzawa Y, Loos RJF, Moreno LA, Bray GA, Martinez JA, Obesity. Nat Reviews Disease Primers. 2017;3:17034.
Estruch R, Ros E. The role of the Mediterranean diet on weight loss and obesity-related diseases. Reviews in endocrine & metabolic disorders. 2020;21:315–27.
Ardalan ZS, Sparrow MP, Muir JG, Gibson PR. Dietary fat and the faecal microbiome: where collinearity may lead to incorrect attribution of effects to fat. Gut. 2020;69:1.
McEligot AJ, Poynor V, Sharma R, Panangadan A, Logistic LASSO Regression for Dietary Intakes and Breast Cancer, Nutrients, 12 (2020) E2652.
Lin X, Li H. Obesity: Epidemiology, Pathophysiology, and Therapeutics. Front Endocrinol. 2021;12:706978.
Wang L, Martínez Steele E, Du M, Pomeranz JL, O'Connor LE, Herrick KA, Luo H, Zhang X, Mozaffarian D, Zhang FF. Trends in Consumption of Ultraprocessed Foods Among US Youths Aged 2–19 Years, 1999–2018. 326: Jama; 2021. pp. 519–30.
Dolley S. Big Data's Role in Precision Public Health. Front public health. 2018;6:68.
Lin Q, Zhao Z, Liu JS. Sparse Sliced Inverse Regression Via Lasso. J Am Stat Assoc. 2019;114:1726–39.
Li Z, Sillanpää MJ. Overview of LASSO-related penalized regression methods for quantitative trait mapping and genomic selection, TAG. Theoretical and applied genetics. Theoretische und angewandte Genetik. 2012;125:419–35.
Sun C, Zhu B, Zhu S, Zhang L, Du X, Tan X. Risk Factors Analysis of Bone Mineral Density Based on Lasso and Quantile Regression in America during 2015–2018. Int J Environ Res Public Health. 2021;19:355.
Qu Y, Pan C, Guo S, Wu H. Dietary Intake and Asthma in Preschoolers: A Logistic Lasso Regression Analysis. Front Pead. 2022;10:870529.
Astrup A, Bügel S, Overfed but undernourished: recognizing nutritional inadequacies/deficiencies in patients with overweight or obesity, International Journal of Obesity (2005), 43 (2019) 219–232.
Thomas-Valdés S, Tostes M, Anunciação PC, da Silva BP. H.M.P. Sant'Ana, Association between vitamin deficiency and metabolic disorders related to obesity. Crit Rev Food Sci Nutr. 2017;57:3332–43.
Thomas-Valdés S, Tostes MdGV, Anunciação PC, da Silva BP. H.M.P. Sant'Ana, Association between vitamin deficiency and metabolic disorders related to obesity. Crit Rev Food Sci Nutr. 2017;57:3332–43.
Association between vitamin. deficiency and metabolic disorders related to obesity - PubMed.
Pellegrini M, Rahimi F, Boschetti S, Devecchi A, De Francesco A, Mancino MV, Toppino M, Morino M, Fanni G, Ponzo V, Marzola E, Abbate Daga G, Broglio F, Ghigo E, Bo S. Pre-operative micronutrient deficiencies in patients with severe obesity candidates for bariatric surgery. J Endocrinol Investig. 2021;44:1413–23.
Perticone F, Ceravolo R, Candigliota M, Ventura G, Iacopino S, Sinopoli F, Mattioli PL, Obesity and body fat distribution induce endothelial dysfunction by oxidative stress: protective effect of vitamin C, Diabetes, 50 (2001) 159–165.
Ellulu MS. Obesity, cardiovascular disease, and role of vitamin C on inflammation: a review of facts and underlying mechanisms. Inflammopharmacology. 2017;25:313–28.
Gao L, Siu KL, Chalupsky K, Nguyen A, Chen P, Weintraub NL, Galis Z, Cai H, Role of uncoupled endothelial nitric oxide synthase in abdominal aortic aneurysm formation: treatment with folic acid, Hypertension (Dallas, Tex.: 1979), 59 (2012) 158–166.
Gargari BP, Aghamohammadi V, Aliasgharzadeh A. Effect of folic acid supplementation on biochemical indices in overweight and obese men with type 2 diabetes. Diabetes Res Clin Pract. 2011;94:33–8.
Evans E, Piccio L, Cross AH. Use of Vitamins and Dietary Supplements by Patients With Multiple Sclerosis: A Review. JAMA Neurol. 2018;75:1013–21.
Belsky JB, Wira CR, Jacob V, Sather JE, Lee PJ. A review of micronutrients in sepsis: the role of thiamine, l-carnitine, vitamin C, selenium and vitamin D. Nutr Res Rev. 2018;31:281–90.
Johnston CS. Strategies for healthy weight loss: from vitamin C to the glycemic response. J Am Coll Nutr. 2005;24:158–65.
Block G, Jensen CD, Dalvi TB, Norkus EP, Hudes M, Crawford PB, Holland N, Fung EB, Schumacher L, Harmatz P. Vitamin C treatment reduces elevated C-reactive protein. Free Radic Biol Med. 2009;46:70–7.
Stehle P, Stoffel-Wagner B, Kuhn KS. Parenteral trace element provision: recent clinical research and practical conclusions. Eur J Clin Nutr. 2016;70:886–93.
Gu K, Li X, Xiang W, Jiang X. The Relationship Between Serum Copper and Overweight/Obesity: a Meta-analysis, Biological trace element research, 194 (2020) 336–347.
Sánchez C, López-Jurado M, Aranda P, Llopis J. Plasma levels of copper, manganese and selenium in an adult population in southern Spain: influence of age, obesity and lifestyle factors. Sci Total Environ. 2010;408:1014–20.
Yang H, Liu CN, Wolf RM, Ralle M, Dev S, Pierson H, Askin F, Steele KE, Magnuson TH, Schweitzer MA, Wong GW, Lutsenko S, Obesity is associated with copper elevation in serum and tissues, Metallomics: integrated biometal science, 11 (2019) 1363–1371.
Mak IL, Lavery P, Agellon S, Rauch F, Murshed M, Weiler HA. Arachidonic acid exacerbates diet-induced obesity and reduces bone mineral content without impacting bone strength in growing male rats. J Nutr Biochem. 2019;73:108226.
Pickens CA, Sordillo LM, Zhang C, Fenton JI, Obesity is positively associated with arachidonic acid-derived 5- and 11-hydroxyeicosatetraenoic acid (HETE), Metabolism, 70 (2017) 177–191.
Börgeson E, Johnson AM, Lee YS, Till A, Syed GH, Ali-Shah ST, Guiry PJ, Dalli J, Colas RA, Serhan CN, Sharma K, Godson C. Lipoxin A4 Attenuates Obesity-Induced Adipose Inflammation and Associated Liver and Kidney Disease. Cell Metab. 2015;22:125–37.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Logistic LASSO Regression for Dietary Intakes and Obesity: NHANES (2007-2016)

Status:

Version 1

Abstract

Figures

Introduction

Material And Methods

Sample

BMI and Dietary Intake Assessment

Covariate Assessment

Statistical Analysis

Results

Sample characteristics

Dietary intake

Logistic LASSO regression to screen dietary obesity factor set

Logistic regression and ROC tests uesd to validate the obesity factor set

Discussion

Abbreviations

Declarations

Ethics approval and consent to participate

Consent for publication

Availability of data and materials

Competing interests

Funding

Authors' contributions

Acknowledgements

References

Additional Declarations

Status:

Version 1