*3.2. Binary Logistic Regression Analysis*

3.2.1. The Association between RGCS and Daily Total Dietary Energy, Macronutrients, Vitamins, and Minerals

For the crude model and adjusted model I, we found that energy, protein, total fat, dietary fiber, vitamin B1, vitamin B6, vitamin E, iron, and zinc intake were associated with RGCS, including good RGCS and poor RGCS (Figure 2A,B). However, after controlling of all covariates and time fixed effect (Figure 2C,D), opposite results were obtained from the former two models, for instance, three kinds of macronutrients, minerals, and dietary fiber actually had no statistical association with RGCS. Eventually, the statistical results in the robust check model (Figure 2D and Table 2), after controlling for the potential confounders and years fixed effect, suggested a significantly negative correlation between daily total dietary vitamin B6 intake and RGCS (adjusted OR = 0.848; 95% CI: 0.738, 0.973; *p*-value = 0.019).

**Figure 2.** Forest plot for odds ratio (OR) and 95% confidence interval (CI) of daily total dietary nutrient and energy intake ((**A**) without covariates; (**B**) gender, age, and race were controlled; (**C**) all potential confounders in the study were controlled; (**D**) all potential confounders and the years fixed effect in the study were controlled).




**Table 2.** *Cont.*

<sup>a</sup> A total of 15 dietary variables were entered in the crude model: protein, carbohydrate, total fat, dietary fiber, vitamin b1, vitamin b2, vitamin b6, total folate, vitamin b12, vitamin e, calcium, magnesium, iron, zinc, copper. <sup>b</sup> Three variables were adjusted in model I: gender, age, race. <sup>c</sup> A total of 17 variables were adjusted in model II: gender, age, race, education level, BMI, moderate or severe physical activity, hypertension, the doctor informing them that they had diabetes, having at least 12 cups of alcoholic drink per year, consuming over 100 cigarettes in their lifetime, food security, PIR, energy, insulin, glucose, hemoglobin. <sup>d</sup> Robust check model: Based on model II, years fixed effect was adjusted. \* *p*-value < 0.05; \*\* *p*-value < 0.01; \*\*\* *p*-value < 0.001. † BMI, body mass index; PIR, poverty income ratio; RGCS, recent glycemic control states; SE, standard error.

#### 3.2.2. The Association between Adjusted Covariates and RGCS

The statistical results of the robust check model, all covariates, and time fixed effect being adjusted demonstrated that age (taking "<40 years old" as a reference, OR[40–59 years old] = 2.659, *p*-value < 0.001; OR[≥60 years old] = 2.186, *p*-value < 0.001), gender (taking "female" as a reference, OR[Male] = 1.365, *p*-value = 0.008), race (taking "other races" as a reference, OR[Mexican American] = 0.796, *p*-value = 0.131; OR[Non-Hispanic Black] = 1.486, *p*-value = 0.016; OR[Non-Hispanic White] = 1.044, *p*-value = 0.804), education level (taking ">high school" as a reference, OR[≤High School] = 1.205, *p*-value = 0.083), BMI (taking "<25.0" as a reference, OR[25.0–29.9] = 2.042, *p*-value < 0.001; OR[≥30.0] = 1.079, *p*-value = 0.626), hypertension (taking "no" as a reference, OR[Yes] = 1.381, *p*-value = 0.001), the doctor informing them that they had diabetes (taking "no" as a reference, OR[Borderline] = 12.072, *p*-value < 0.001; OR[Yes] = 3.130, *p*-value < 0.001), insulin (OR = 1.003, *p*-value = 0.132), glucose (OR = 1.063, *p*-value < 0.001), and hemoglobin (OR = 0.918, *p*-value = 0.017) were significantly associated with RGCS in Table 2.

#### *3.3. Model Fitting, Linear Discriminant Analysis, and ROC Analysis*

3.3.1. Model Fitting and Linear Discriminant Analysis of Daily Total Dietary Vitamin B6 Intake, Glycohemoglobin, and RGCS

After smooth curve fitting of daily total dietary vitamin B6 intake and glycohemoglobin being conducted in Figure 3, the robust check model, a linear discriminant analysis of daily total dietary vitamin B6 intake and RGCS, was also fitted in Figure 4. The statistical analysis graphs showed that the established robust check model could not only distinguish American adults with different RGCS well, but pointed out that the negative correlation between daily total dietary vitamin B6 intake and RGCS did exist. It was indicated that daily total dietary vitamin B6 intake might have a potential predictive value for RGCS of American adults.

**Figure 3.** The model fitting processes of non-linear regression curve ((**A**,**B**), scatter plots; (**C**), the optimal smooth curve).

**Figure 4.** The linear discriminant analysis of daily total dietary vitamin B6 intake, glycohemoglobin (**A**), and RGCS (**B**).

#### 3.3.2. ROC Analysis of Daily Total Dietary Vitamin B6 Intake

ROC analysis of daily total dietary vitamin B6 intake was performed to calculate the area under the curve (AUC), which was used to evaluate the discrimination accuracy among people with good and poor RGCS. As shown in Figure 5, after controlling for all potential confounders and the years fixed effect, the predictive potential or accuracy of the multivariate logistic regression robust check model (AUC = 0.977; 95% CI: 0.974, 0.980; *p*-value < 0.001) was higher than those of the crude model (AUC = 0.535; 95% CI: 0.519, 0.550; *p*-value < 0.001), adjusted model I (AUC = 0.710; 95% CI: 0.697, 0.723; *p*-value < 0.001), and adjusted model II (AUC = 0.975; 95% CI: 0.974, 0.979; *p*-value < 0.001). The test results of DeLong between robust check model and crude model, and adjusted model I and adjusted model II showed that there were two statistically significant results (robust check model vs. crude model, *Z* = 53.54, *p*-value < 0.001; robust check model vs. adjusted model I, *Z* = 39.542, *p*-value < 0.001; robust check model vs. adjusted model II, *Z* = 0.751, *p*-value = 0.452).

**Figure 5.** Receiver operator characteristic (ROC) curve indicator of poor RGCS (crude model: without covariates; adjusted model I: gender, age, and race were controlled; adjusted model II: all potential confounders in the study were controlled; robust check model: all potential confounders in the study were controlled, and the years fixed effect in the study was included).

#### **4. Discussion**

Although the reliable data of our study were from the national representative sample published by the Centers for Disease Control and Prevention of the United States, there were still some missing values in our collected data, which might have a subtle influence on our results of the statistical analysis. However, given the sufficient sample size of this study, the deviation caused by missing values could be reduced. Therefore, the reliability and authenticity of our findings were within acceptable limits. In addition, we did not ignore the interactions among nutrients when fitting the saturation model, but these interactions that might have biological significance (such as the interaction between vitamin B6 and vitamin B12) were not statistically significant when they were included in the robust check model.

The statistical model constructed in our research combined the specific macronutrients, minerals, vitamins, dietary fiber, and energy of the daily total diet, demographic, and medical indicators of American adults, because some statistical models mentioned in other studies might only be relevant for a small group of people with diabetes and not be suitable for American adults in terms of predicting their RGCS. Thus, our robust check model was closer to the real-world results than the models established by those research institutes [27,29].

Eventually, we found only daily total dietary vitamin B6 intake negatively correlated with RGCS, that is, the higher the daily total intake of dietary vitamin B6 was accompanied with better RGCS. Similarly, Mascolo et al. also concluded that the vitamin B6 level was significantly negatively associated with diabetes mellitus in diabetic people and suggested that vitamin B6 had a significantly protective effect on diabetic complications [30]. In addition, although covariates adjusted in our robust check model could not answer our research issues, they could still provide the theoretical foundation and scientific guidance for our health education related to glycemic control for American adults, which had a valuable public health significance.

The establishment of the robust check model of nutrients and RGCS was only the preliminary step of this study, which was mainly used to qualitatively find the associated factors affecting the RGCS. After screening the statistically significant daily total dietary vitamin B6 intake with this model, we further performed linear discriminant model and ROC analysis between RGCS and daily total dietary vitamin B6 intake as well as HbA1c, respectively (Figures 4 and 5), which could provide a quantitative reference and prediction accuracy of daily total dietary vitamin B6 intake for American adults who need to control blood glucose.

Vitamin B6 was an intriguing molecule that was involved in a wide range of metabolic, physiological, and developmental processes. Its active form, 5'-pyridoxal phosphate (PLP), was a co-factor for approximately 150 metabolic responses to glucose, lipid, amino acids, DNA, and neurotransmitters [30–34]. These studies showed that vitamin B6 had a potential to regulate body metabolism (including blood glucose). Although the United States, South Korea, and Japan published the recommended total dietary intake of vitamin B6 (fluctuating around 1.1~1.6 mg/d) for specific populations [35–37], it might not be suitable for American adults who need to control blood glucose. Therefore, it was necessary for the relevant health management agencies in the United States to formulate the recommended value of the total daily dietary vitamin B6 intake of RGC for American adults, so as to provide an effective way for them to obtain better RGCS. However, the formulation of total daily dietary vitamin B6 recommended intake remains to be further explored.

#### **5. Conclusions**

In summary, our results indicated that only daily total dietary vitamin B6 intake was significant negatively associated with RGCS among all dietary nutrients we studied. Although this study provided a ROC prediction result of daily total dietary vitamin B6 intake for RGCS, we might require further validation of whether it would have a positive and effective preventive effect and biological implications on RGCS of American adults.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/1 0.3390/nu13114168/s1, Figure S1: Participant flow chart.

**Author Contributions:** The authors' responsibilities were as follows—Y.B., L.P.: contributed to data collection, analysis, and manuscript writing; H.Z., J.Y.: participated in the research design; L.P.: devoted to research design and manuscript writing. All authors have read and agreed to the published version of the manuscript.

**Funding:** The research received no sources of support such as commercial or non-commercial funding. The findings and conclusions expressed in this article were the authors' and do not necessarily represent the official position of the CDC or the U.S. Department of Health and Human Services. There was no private sponsor who played any role in the decision to design the study, collect data, analyze or interpret data, write reports, or submit manuscripts.

**Institutional Review Board Statement:** Ethical review and approval were waived for this study, since all the data from National Health and Nutrition Examination Survey is publicly accessible.

**Informed Consent Statement:** Informed consent from all subjects was obtained by National Health and Nutrition Examination Survey.

**Data Availability Statement:** Data described in the manuscript, codebook, and analytic code will not be made available because the data used in this study were from the NHANES database, which is a free and open database for all researchers around the world.

**Conflicts of Interest:** The authors stated that no conflict of interest would be deemed to prejudice the impartiality of the reported study.

#### **Abbreviations**

AUC: area under the curve; BMI, body mass index; CI, confidence interval; HbA1c, glycosylated hemoglobin; NCHS, National Center for Health Statistics; NHANES, National Health and Nutrition Examination Survey; OR, odds ratio; PIR, poverty income ratio; RGCS, recent glycemic control status; ROC, receiver operator characteristic; SE, standard error; USDA, United States Department of Agriculture; VIF, variance inflation factor.

### **References**

