Geographically Weighted Regression to Predict the Prevalence of Hypertension Based on the Risk Factors in South Kalimantan

Hypertension is one of the disease is not contagious diseases which is a public health problem. Uncontrolled Hypertension can trigger a degenerative disease such as congestive heart failure, renal failure and vascular disease. Hypertension is called the silent killer because his nature the condition is asymptomatic and can cause a fatal stroke. With the increasing prevalence of cases of degenerative diseases, one only hypertension, then the researchers want to predict the variables very big role as one of the risk factors of Genesis hypertension. With clearly know the risk factors that play against genesis hypertension is expected to be used as a reference for the prevention and control so that they can reduce the prevalence of hypertension and prevent deaths from degenerative diseases, especially hypertension. The results of the study showed that the results of the modeling the prevalence of hypertension in South Kalimantan Province using linier regression there is no factor that affect the genesis of hypertension. The prevalence of hypertension spread spatially because there are heterogenitas between the location of the observation that means that observations of a location depends on the observations in another location that the distance is near so do spatial regression modeling with adaptive gaussian kernel function, the result 5 groups. Group I consists of the districts Tanah Laut and Tanah Bumbu; group II, Kota Baru; Group III consists of Banjar, Kota Banjar Baru, Kota Banjarmasin; Group IV on the Barito Kuala Regency and the Group V consists of Tapin, H S Selatan, H S Tengah, H S Utara, Tabalong, Balangan.


INTRODUCTION
A research influenced by aspects of territorial characteristic (spatial) then need to be considered spatial data on the model.Spatial data is data that contains the location information.On spatial data, often observations in a location dependent on observation in other locations near (neighboring).
The first law of geography advanced by Tobler in 1979, stated that all things are related to each one with the other but something close more had the effect of something far [1].The law is the basis of the examination of the problems based on the effect of the location or spatial method.In the modeling language, when the classic regression model used as a tool of analysis on spatial data, then can cause the conclusion that less accurate

Geographically Weighted Regression to Predict the Prevalence of Hypertension Based on the Risk Factors in South Kalimantan
Suroto 141 because of the assumption of the error free to each other and the assumption homogenitas not met [2].Hypertension is one of the disease is not contagious diseases which is a public health problem.Uncontrolled Hypertension can trigger a degenerative diseases such as congestive heart failure, renal failure and vascular disease.Hypertension is called the silent killer because his nature the condition is asymptomatic and can cause a fatal stroke [3].Although not treated, prevention and managements can decrease the occurrence of hypertensi and the accompanying disease.Hypertension is the cause of the death of number 3 after stroke and tuberculosis, namely 6.7 % from the population of death in all age in Indonesia.The problem of hypertension that occurs in South Kalimantan would not escape from the factors causing the hypertension essential where genetic factors, environment, behavior, health services also contribute to cause a high case of hypertension in South Kalimantan Province [4].To identify the multitude of causative factors that affect it may need to be an analysis or the development of the model is spatial [5].
Spatial effects testing done with test heterogenitas and spatial dependencies.If there is a settlement of securities heterogenity is by using point approach.Spatial regression points between the other Geographically Weighted Regression (GWR) with the scale of the measurement of the response variable is the interval and ratio,Geographically Weighted Poisson Regression (GWPR) with the variable data is response count, Geographically Weighted Logistic Regression (GWLR) with the scale of the measurement of the response variable is the nominal [6].
Based on the [7], the prevalence of hypertension in Indonesia that acquired through the measurement at the age > 18 years of 25.8%, the highest in Bangka Belitung securities have totaled 30.9%, followed South Kalimantan 30,8 % , East Kalimantan 29.6% and West Java 29.4%.The prevalence of hypertension in Indonesia that is obtained through the questionnaire diagnosed health workers of 9.4%, diagnosed health workers or is drinking drugs of 9.5%.So there are 0.1% that medication itself.Respondents who have a normal blood but is drinking Hypertension medications 0.7%.So the prevalence of hypertension in Indonesia by 26.5%.
The problem of hypertension that occurs in South Kalimantan would not escape from the factors causing the hypertension essential where genetic factors, environment, behavior, health services also contribute to cause a high case of hypertension in South Kalimantan Province.To identify the multitude of causative factors that affect it may need to be an analysis or development models.With the increasing prevalence of hypertension in a region, then the researchers want to predict the variables very big role as one of the risk factors of Genesis hypertension.With clearly know the risk factors that play against genesis hypertension with GWR approach is expected to be used as a reference for the prevention and control so that they can reduce the prevalence of hypertension.

METHODS
The data used is the Basic Health Research data [7].The data will be analyzed in this research is data genesis hypertension in regency/city of South Kalimantan Province.Based on the results of the analysis of the previous library, hypertension served on the following conceptual framework [8].

Geographically Weighted Regression to Predict the Prevalence of Hypertension Based on the Risk
Factors in South Kalimantan Suroto 142 The response variable Y that is used in this research is the percentage of patients with hypertension are diagnosed per sub-districts in South Kalimantan Province.While the variables predictors Xi used there in Table 1.
Table 1.The variables Predictors [9] Variables Description X1 The percentage of the inhabitants of gender male X2 The percentage of population with education Completed SD/MI X3 The percentage of population with smoking habit every day X4 The percentage of the population of physical activity X5 Percentage of the population who consume the fruits of 7 times in 1 weeks X6 Percentage of the population who consume vegetables 7 times in 1 weeks

X7
Percentage of the population who consume salty food more than 1 times per day

X8
The percentage of the population consuming fatty food consumption/ order/ fried more than 1 times per day X9 The percentage of the population with ownership of health insurance The variables X1 and X2 used to describe the spread of environmental aspects according to their geographic region.Aspects of the behavior described by the variables X3, X4, X5, X6, X7 and X8.While aspects of the Ministry of Health is portrayed through the variable X9.

Description of the Prevalence of Hypertension
Description of this research includes the mean and standard deviation from each of the research variables.Now in detail is presented in the following table.Table 2. shows that the percentage of the highest Hypertension Prevalence on the sub-district Pulau Laut Tanjung Selayar of 29.508 percent and lowest on sub-district Tatah Makmur of 1.471 percent while the average percentage of the prevalence of hypertension by 14.978 percent with standard deviation of 5.764 percent.The percentage of the inhabitants of gender male (X1) highest on the sub-district Danau Panggang (55.556 percent), and the lowest on the sub-district Pulau Laut Tanjung Selayar (32.787).The percentage of population with education Completed SD/MI (X2) highest on the sub-district Pamukan Barat of 69.230 percent and lowest on sub-district Tatah Makmur of 7.320 per cent while the average percentage of the prevalence of hypertension by 33.660 percent with standard deviation of 12.340 percent.The percentage of population with smoking habit every day (X3) has an average of 23.750 percent with standard deviation 7.507 percent.The percentage of the population physical activity (X4) has an average of 86.855 percent with standard deviation 9.988 percent.Percentage of the population who consume the fruits of 7 times in 1 weeks (X5) has an average of 20.289 percent with standard deviation 7.070 percent.Percentage of the population who consume vegetables 7 times in 1 weeks (X6) has an average of 55.470 percent with standard deviation 15.2 percent.Percentage of the population who consume salty food more than 1 times per day (X7) has an average of 45.08 percent with standard deviation 13.170 per cent.The percentage of the population consuming fatty food consumption/ order/ fried more than 1 times per day (X8) has an average of 31.670percent with standard deviation 15.62 percent and the percentage of the population with ownership of health insurance (X9) has an average of 15.002 percent with standard deviation 11.946 percent.

Geographically Weighted Regression to Predict the Prevalence of Hypertension Based on the Risk Factors in South Kalimantan
Suroto 144

Regression modeling the prevalence of hypertension
Before performing the analysis using spatial regression method, done regression modeling double linier first.Regression modeling linear to the prevalence of hypertension and factors suspected to influence using parameter assessment method Ordinary Least Square (OLS) [10] which aims to know the variables significant on the prevalence of hypertension globally.The first step is to detect multicolinearity to know whether or not the relationship between the free variable (predictors), then continued with double linier regression modeling (global) covers the review of the significance of the parameters simultaneously or partially, and residual assumptions IIDN test [11].

Multicolinearity Detection
One of the conditions in the multiple regression analysis with some variables predictors is no cases multikolinieritas or not there is a variable predictors that have a correlation with other predictors variable.Tracing muticolinearity done based on the value of Variance Inflation Factor (VIF).The following is the value of VIF in each of the variables predictors.Based on the Table 3, obtained the information that all variables predictors have a VIF value less 10.This detects that there are not cases multikolinieritas or not there is a variable predictors that have a correlation with other predictors variable.

The Significance of Parameters Linear Regression tests the prevalence of hypertension
The following is the significance test good linier regression parameters simultaneously or partial to know the influence of predictors variables used.The hypothesis to test the significance of the parameters simultaneously on the linier regression is as follows [6].H0 : β1 = β2 = ...= β9 = 0 (the parameters do not affect the significant impact on model) H1 : at least one βk ≠ 0 ; k = 1.2, ... ,9 (at least one parameters that affect the significant impact on model) Table 4., produce F-Statistic value of 2.14 and p-value of 0.030.Based on the level of the significance () of 5 percent and F(0.05;9;142) of 1,.46, obtained the decision Reject H0 because the value of F-Statistic > F(0.05;9;142) or p-value < 0.05.This can be interpreted that there is at least one parameters that affect the significant impact on the prevalence of hypertension.
Next to know the variables predictors anywhere that provides significantly influence, then the test is done the significance of parameters partially presented in Table 5.The following is the hypothesis test the significance of the parameters spatially against linier regression model (global) [6].

Testing the Assumptions of a Residual IIDN
After testing the significance of the parameters simultaneously and partially, then the next step is to test the assumptions residual of identical, independent, and normal distribution (IIDN).
-Test the assumption of identical Residual One of the assumption of the test in the OLS regression is a residual must be homoscedasticitys variance (is identical) or in cases of heteroscedasticity.How to identify the existence of the case of heteroskedastisitas is to create a regression model between a residual and predictor variables.When there are variables predictors that affect the model significantly, it can be said that the residual is not identical or in cases of heteroscedasticity.Testing the assumption of identical residual provides information that there are not cases heteroscedasticity or identic residual with significant () of 0. -Test the assumption of Independent Residual [11] Test the assumption of independent residual used to know whether or not the relationship between the residual periods.The test statistics used is Durbin-Watson.Based on the attachment 4 obtained the value .So that the decision can be taken is to fail to reject H0 because .It shows that there is no relationship between the residual, so that the assumption of independent residual have been fulfilled.

-Test the assumption Normal distribution
The assumption of the normal distribution test is done with the following Kolmogorov-Smirnov test.
H0: Data normal distribution H1: Data not normal distribution Based on the Figure 2 obtained the information that the points of red spread near linear line (normal) which means that the data has been normal distribution.In addition, also can be seen from the value of P-value that is more > 0.15.So that the decision can be taken is to fail to reject H0 on equal significant () of 5%, because the value of the Pvalue is greater than .This means that the data has been meet the assumption normal distribution.Based on the results of the test of assumption, it can be concluded that a residual on the linear regression model (global) meets the independent assumption, and data has been normal distribution, but the assumption is identical not fulfilled.Factors in South Kalimantan Suroto 147

Spatial Regression Modeling the Prevalence of Hypertension
The analysis using the GWR method aims to know the variables that affect the prevalence of Genesis hypertension on each observation location regency/city in South Kalimantan province.The first step is done to get the GWR model is to determine the coordinates of the point latitude and longitude on each location to count the distance euclidean and determine the optimum bandwidth values based on the criteria of Cross validation (CV).The next step is to determine the matrix pembobot with kernel functions: Fixed Gaussian, fixed bi-square, Adaptive Gaussian, Adaptive Bi-Square and assess the GWR model parameters.The matrix pembobot obtained for each location and then used to form a model, so that obtained the model vary in each location of observation [12].
GWR model hypothesis testing consists of two test, test the suitability of the GWR model and test the significance of the parameters GWR model.The following is the results of the hypothesis testing GWR model [6].H0 : ; (There is no significant difference between the linear regression model (global) and GWR model) H1 : at least one  6 shows the comparison of the model with pembobot GWR estimates that vary.Testing the suitability of the GWR model is done by using the difference in the number of residual square GWR model and global regression model.GWR Model will vary significantly with global regression model if can reduce the amount of residual square significantly.Table 6., shows that the value of the smallest AICc is the GWR model with adaptive Gaussian kernel pembobot amounting 956.153.So using the significance level    5 percent so it can be concluded that the GWR model differ significantly with global regression model.This means that the kernel pembobot GWR model with adaptive Gaussian more worthy to illustrate the percentage of the prevalence of hypertension in South Kalimantan Province.
Next is a test of the significance of the parameters GWR model with adaptive Gaussian kernel pembobot partially to know any parameters that affect the prevalence of hypertension in each location of observation.Sub Division that have common

Figure 1 .
Figure 1.GWR Modeling flow in the case of hypertension

Figure 2 .
Figure 2. Probability plots of Residual of the Hypertension prevalence

Table 3 .
The value of their respective VIF Predictors Variables

Table 4 .
ANOVA table the prevalence of hypertension in South Kalimantan Province

Table 5 .
The results of the Test Regression Model Parameters Linear Partial

Table 6 .
2, ....,9 (No difference between significant linear regression model (global) and GWR model) The estimation of GWR on the weight of the Kernel Function