The spatial distribution of BUN reference values of Chinese healthy adults: a cross-section study.

The blood urea nitrogen (BUN) is generally regarded as a significant serum marker in estimating renal function. This study aims to explore the geographical distribution of BUN reference values of Chinese healthy adults, and provide a scientific basis for determining BUN reference values of Chinese healthy adults of different regions according to local conditions. A total of 25,568 BUN reference values of healthy adults from 241 Chinese cities were collected in this study, and 17 geographical indices were selected as explanatory variables. The correlation analysis was used to examine the significance between BUN reference value and geographical factors, then five significant indices were extracted to build two predictive models, including principal component analysis (PCA) and support vector regression (SVR) model, then the optimal model was selected by model test to predict BUN reference values of the whole China, finally the distribution map was produced. The results show that BUN reference value of Chinese healthy adult was characteristically associated with latitude, altitude, annual mean temperature, annual mean relative humidity, and annual precipitation. The model test shows, compared with SVR model, the PCA model possesses superior simulative and predictive ability. The distribution map shows that the BUN reference values of Chinese healthy adult are lower in the east and higher in the west. These results indicate that the BUN reference value is significantly affected by geographical environment, and the BUN reference values of different regions could be seen clearly on distribution map.

Keywords: Blood urea nitrogen (BUN); Chinese healthy adult; Geographical indices; Predictive model; Geographical distribution

Urea nitrogen is the main end product of protein metabolism in human body which is generally regarded as one of the mainly widely used serum makers in estimating renal function; it is also a seized item of health examination now. BUN plays a significant role in diagnosis of renal diseases, prediction of cardiovascular events caused by acute heart failure, and estimate of prognosis in patients with acute myocardial infarction (Ke 2013; Tao 2014; Tan and Yin 2015; Li et al. 2015; Zhu et al. 2017; Zhang 2017).

Current related studies (Ma et al. 2016; Zhao et al. 2016) mostly focus on the effect of non-environmental factors on BUN, such as gender and age, whereas there are few researches concerning the effect of geographical environment on BUN. For instance, Dr. Li of 18th Hospital of PLA (and others) found that BUN level could be affected by plateau condition; after being transferred to plains, the BUN level was going to recover (Li et al. 2017). Moreover, there is no uniform reference value of BUN, both at home and abroad, and its use is non-standard. For instance, the different medical institutions in the same region use different standards of BUN reference value, while they use the same detecting instrument and method. In order to provide a scientific basis for setting standard of BUN reference value in China, this study built two predictive models using principal component analysis (PCA) model and support vector regression (SVR) model, then selected the optimal model that possessed superior simulative and predictive ability by model test and constructed distribution map of BUN reference values of Chinese healthy adults (Ge et al.2009; Han et al. 2015).

In this study, the normal reference value is defined as the measured value within the current range of medical facility. We collected 25,568 BUN normal reference values (unit, mmol/L) from 324 medical facilities, including hospitals, research institutes, and universities. In this study, 75% of those data were used for simulating and building predictive model, and the rest was used as model testing data set. It was important to note that each subject was measured from healthy adults who were considered to be healthy after physical examination, participants were aged 21 to 86, their ratio of male and female was 1.2:1. The exclusion criteria were essential hypertension, dysfunction of main organs, severe (acutely or chronically) infections, traumas or operation within 6-month period, and so on.

All of the medical data were collected from 241 cities of 32 provinces, municipalities, and autonomous regions, while the material of Taiwan province, Hong Kong, and Macau Special Administrative Regions were lacking. The material of eastern plain region was much more than that of western plateau region (Fig. 1).Distribution map of observation points

PHOTO (COLOR)

This study selected 17 geographical indices including geographic location, terrain, climatic, and soil indices (Table 1). The data of first two types were from shared data provided by the State Bureau of Surveying and Mapping, the data of last two types were respectively from the China Meteorological Data Service Center and China Soil Database.

Details of geographical indices

As the premise of spatial statistical analysis, spatial autocorrelation is the correlation of the same variable in different spatial locations; it is a measure of aggregation degree of spatial unit attribute values (Lichstein et al. 2002; Diniz-Filhot al. 2003; Xu 2006). Generally, the Moran index, Z-score, and P value are used to describe the spatial relationship, if the Moran I ≠ 0 and |Z| > 1.96, the spatial autocorrelation is significant. Otherwise, the data follow a random distribution and the spatial statistical analysis is meaningless. Based on the collected BUN reference values of sample points, the spatial autocorrelation was conducted in AcrGIS 11.0 software.

Correlation analysis is widely used to explore the dependencies between variables (Li et al. 2006; Wei et al. 2016; Long 2011). In this study, correlation analysis was used to determine the correlation between BUN reference value and 17 geographical indices (Table 2). The correlated geographical indices were extracted to build predictive model.

In order to select appropriate methods to build predictive models, the correlation among extracted geographical indices needs to be explored; collinearity diagnostics was used to determine whether the correlation was collinear in geographical indices. The variance inflation factor (VIF) is generally used to describe the degree of collinearity; the max (VIF) > 10 indicates there is collinearity among variables (Liu 2005; Marill 2004).

PCA applies dimensionality reduction approach in selection of a few numbers of important variables by linear transformation among multiple variables; actually, it assembles a set of synthetical and independent indices which derive from multiple original indices correlating with each other in order to replace previous indices (Liao et al. 2009; Liu et al. 2001).

In SPSS software, the PCA module is integrated into factor analysis module, thus the extraction of principal component has to be carried out in factor analysis. Firstly, the Kaiser-Meyer-Olkin (KMO) test and the Bartlett test are conducted with the aim of determining whether the factor analysis is feasible in this case. The KMO value > 0.5 and significance P value of Bartlett test < 0.05, which indicate that factor analysis is feasible. Then, the principal component could be extracted.

SVR is a type of support vector machine (SVM), with training sample set as data object; it analyzes the quantitative relation between input variable and numeric output variable and predicts the value of output variable from new samples which has the same distribution as training sample set (Foster et al. 2007; Li et al. 2012). In this study, SPSS Clementine software was used to build SVR model; as one of the efficient data mining software, SPSS Clementine is featured with straightforward operation and power data-processing ability by using graphical language programming.

With the extracted geographical indices from correlation analysis as input variables and BUN reference value as output variable, the SVR model was built using SPSS Clementine software. There were four kernel types, including radial basis function (RBF), polynomial, sigmoid, and linear, all of them were used to construct predictive models and calculate variable importance, respectively; then, we sorted each model's variable importance and chose the model whose variable importance was most similar to the result of correlation analysis.

In order to select the optimal predictive model, two predictive models were respectively used to predict the BUN reference value of observation points in the test data set with the aim to test the predictive ability of each model. The Theil inequality coefficient (TIC) and paired-sample t test were used to evaluate the predictive accuracy of each predictive model. The smaller the TIC, the higher the predictive precision. When the significance level P value of paired-sample t test exceeds 0.05, there is no significant difference between measured value and predicted value.

As shown in Fig. 2, the standardized statistic of the Z-score for BUN is 2.895, which exceeds the critical value of the significance level of 0.01. Therefore, there is strong spatial autocorrelation in BUN reference values; the distribution of BUN is not random.Moran's index chart

PHOTO (COLOR)

As shown in Table 2, from P value in result, there are five geographical indices, including X2, X3, X5, X7, and X8 correlated with BUN reference value. According to correlation coefficient r, latitude and altitude show a positive correlation with BUN reference value; however, annual mean temperature, annual mean relative humidity, and annual precipitation show a negative correlation with BUN reference value. These five geographical indices were extracted to build predictive models.

From the result, the VIF values of X2, X3, X5, X7, X8 were VIF1 = 23.215, VIF2 = 3.990, VIF3 = 15.770, VIF4 = 3.724, and VIF5 = 6.239, respectively. There were two items exceeding 10, which showed strong collinearity existed in these geographical indices, therefore, linear model cannot be used in this study; otherwise, the predicted result would not be consistent with real situation. In this case, the principal component analysis (PCA) and support vector regression (SVR) were chosen to build predictive models since these two methods could solve collinearity problem (Latifoğlu et al. 2008; El-Dereny and Rashwan 2011; Adnan et al. 2006).

From the results, the KMO value of five extracted geographical indices was 0.609 and the significance P value of the Bartlett test was 0.000 < 0.05, which indicated that the factor analysis was suitable for this study. Then, the factor analysis was performed; the principal components were extracted via total variance. From Table 3, the cumulative contribution of the first two components is 91.010%. The first two components were consequently extracted to represent original five indices; they were named Z1 and Z2, respectively. According to the eigenvectors of principal components, the formula of Z1 and Z2 could be obtained as follows:Z1=−0.480X2−0.238X3+0.496X5+0.472X7+0.496X8;Z2=−0.381X2+0.920X3+0.049X5-0.044X7+0.065X8.

PHOTO (COLOR)

With Z1 and Z2 as independent variables and BUN reference value as dependent variable, the regression analysis was performed in SPSS 22.0 software to estimate parameters. The regression equation was obtained as follows:Y1=5.0929−0.000248Z1+0.0000396Z2±0.532.

PHOTO (COLOR)

Therein, Y1 is BUN reference value of Chinese healthy adults, and 0.532 is the value of residential standard deviation. After further calculations, the relationship between Y1 and five extracted indices was converted and the regression equation was obtained; it was the PCA model:Y1=5.0929+0.0001039X2+0.0000953X4−0.0001210X5−0.0001188X7−0.0001203X8±0.532.

PHOTO (COLOR)

Table 4 shows the variable importance in each model with different kernel. From the result of correlation analysis, the two most significant variables were X3 and X7; therefore, the linear kernel was selected as the kernel function to build SVR model in this study.

After comparing measured value and predicted value, the TIC of PCA and SVR models were respectively 0.052 and 0.053. According to the results of paired-sample t test in Table 5, under the confidence of 95%, the significance level P value of PCA model and SVR model were respectively 0.997 and 0.908. From those results, PCA model was better than SVR model, and it was considered as the optimal predictive model for this study.

According to geographical data of 2333 observation points of China, the optimal predictive model was used to predict BUN reference value of healthy adults in 2322 Chinese cities. Then the Kolmogorov-Smirnov (K-S) method was used to test the normality of predicted data. The results showed that the significance level P value was 0.000 < 0.05, which indicated that the predicted data did not follow normal distribution; therefore, the disjunctive kriging method was used for interpolation (Xu 2006). The trend analysis was performed, as shown in Fig. 3; the change of predicted data at X axis and Y axis were second-order, which indicated that predicted data needed to be processed by second-order transformation before interpolation.Spatial trend of predicted data

PHOTO (COLOR)

With the help of geostatistics module of ArcGIS software, the spatial distribution map of BUN reference value of Chinese healthy adult was built (Fig. 4). Different colors in Fig. 4 show different BUN reference values. If the color of two regions is similar, the BUN reference value of them will be close.Spatial distribution map of BUN reference values of Chinese healthy adults

PHOTO (COLOR)

In spatial autocorrelation analysis, the significance level P value was 0.0037. This result suggests that BUN reference value of Chinese healthy adult is significantly correlated with geographical environment; accordingly, the BUN reference values vary with spatial attributes, it is feasible to explore the relationship between them and build predictive models.

According to correlation analysis, there were five geographical indices correlated with BUN reference value. By collinearity diagnostics, there were collinearity problems among these indices. In this case, the PCA and SVR, which could effectively resolve collinearity problem, were used to construct predictive model, respectively. After constructing predictive models, selecting optimal model, prediction, and interpolation, the spatial distribution map was obtained. As shown in Fig. 4, the BUN reference value is overall high in the west and low in the east, it varies regularly with altitude, which is basically in accordance with the three gradient terrains of China. With the increase of the altitude, the BUN reference value becomes higher. From the perspective of regional distribution, BUN reference value is relatively low in southern coastal areas and northern China plain, and high in Qinghai and Tibet regions. These results show that the altitude is the most significant factor, which is consistent with the result of correlation analysis.

As an excretory organ of human body, kidney plays a vital role in maintaining homeostasis and normal metabolism. BUN is mainly excreted by kidney, so if the functional disorder occurs in kidney, the excretion of BUN will be affected and the concentration of BUN in blood will be influenced with it (Kirtane et al. 2005; Manoeuvrier et al. 2017).

The most significant factor affecting BUN reference value of Chinese healthy adult is altitude. The kidney has a huge demand for oxygen since it has complicated structure and function. Along with higher altitude, the air becomes thinner, which could aggravate hypoxia and lead to kidney damage (Wang et al. 2011; Mazzali et al. 2003). More specifically, the hypoxemia strengthens the activity of sympatheticoadreno-medullary and enhances the vascular tone, then the renal arteriole will constrict and the renal blood flow and renal plasma flow will reduce with it. This process ultimately lead to the reduction of glomerular filtration rate and discharge of BUN will decrease with it; therefore, the concentration of BUN in blood increases (Yuan and Ma 1999; Gonzales 2011). However, renal dysfunction is not found in normal subjects living in high-altitude areas, the main reason for this is that, under the circumstance with hypobaric hypoxia, the physiology adaptability can help body to overcome it and relieve renal hypoxia; accordingly, the kidney damage caused by hypoxia does not exceed renal compensation (Gao et al. 2006).

As for climatic factors, annual mean relative humidity and annual precipitation were negatively associated with BUN reference value of Chinese healthy adults. From costal to inland regions of China, as the moisture condition gradually becomes poor, the annual precipitation and annual mean relative humidity decrease correspondingly. Compared with coastal areas, the humidity of inland area is relatively low and the evaporation of body water is fast, which contributes to relatively high blood viscosity and low renal blood flow as well as renal plasma flow, so BUN level of people living in inland areas of China is relatively high. Moreover, latitude shows a positive correlation with BUN reference value; on the contrary, annual mean temperature shows a negative correlation with BUN reference value. These two opposite correlations result from that annual mean temperature decrease with the increase of latitude from the south to the north of China. In other words, latitude affects BUN reference value by affecting temperature conditions. In southern China, its climate belongs to tropical monsoon climate and subtropical monsoon climate, compared with northern China, the annual mean temperature, annual relative humidity, and annual precipitation are higher; accordingly, BUN level of people living in southern China is lower than that of people living in northern China.

In addition, the impacts of soil factor on BUN reference value were not reflected in correlation analysis; soil texture and constituent vary in different regions, which affects regional air quality, microclimate, and so on, thus human health could be affected indirectly. Certainly, the relationships between BUN reference value and each geographical index are not independent of each other; the spatial disparity of reference value is the integrated result of multitude of geographical factors.

In this study, we find that the latitude (X2), altitude (X3), annual mean temperature (X5), annual mean relative humidity (X7), and annual precipitation (X8) are significantly correlated with BUN reference value of Chinese healthy adults; the altitude is the most significant factor. The blood viscosity could be affected by geographical factors in various ways, which leads to different BUN reference values in different regions. The distribution law of BUN reference value in China is that it is high in the west and low in the east, which is consistent with altitude trend and climate features.

Putting geographical factors into medical analysis will be beneficial to determine medical reference values of different regions according to local conditions. Nevertheless, the influence mechanism of each impact factor on BUN remains to be studied further.

This work was supported by grant 40971060 from the Nature Science Foundation of China, and grants 2016CSY012, 2016CSZ005, and 2016TS055 from the fundamental Research Funds for the Central University.

The authors declare that they have no competing interests.