NHANES_data_extraction.Rmd

---
title: "R Notebook"
output: html_notebook
---


How does using a categorical variable for continous variable? How does it affect the predictability and inference?


```{r, echo=FALSE, message=FALSE, warning=FALSE}
library(tidymodels)
library(tidyverse)
library(magrittr)
library(RNHANES)
library(skimr)
library(GGally)
```

Downloading the data 
```{r}
#Download demographic data
DEMO_I = nhanes_load_data("DEMO", "2007-2008", demographics = TRUE)

nhanes_load_data("EPH", "2007-2008")

write.csv(x = DEMO_I, file =  "DEMO_I.csv", row.names = FALSE)

DEMO_I %<>% 
  select(SEQN, SDDSRVYR, RIDSTATR, RIAGENDR, RIDAGEYR, RIDRETH3, DMDEDUC2, DMDMARTL, DMDHSEDU,  WTINT2YR, WTMEC2YR, SDMVPSU, SDMVSTRA, INDFMPIR)

skim(DEMO_I)

phenols <- nhanes_load_data("EPH", "2007-2008", demographics = TRUE)
read.csv(file = "https://www.cdc.gov/nchs/rands/files/RANDS4.csv") %>% 
  mutate(Source = '')
```



```{r}
# 1. Remove rows with at least one NA
DEMO_I %<>% 
  drop_na()

# 2. Convert age to categorical variable & convert FIPR to categorical variable & the rest

DEMO_I %<>% 
  mutate(age_cluster = (kmeans(x = RIDAGEYR, centers = 3))$cluster,
         FIPR_cluster = (kmeans(x = INDFMPIR, centers = 3))$cluster,
         RIDAGEYR = ifelse(RIDAGEYR <= 80, RIDAGEYR, 80),
         DMDEDUC2 = ifelse(DMDEDUC2 <= 5, DMDEDUC2, 5)) %>% 
  filter(RIAGENDR %in% c(1,2),
         RIDAGEYR >= 20)

skim(DEMO_I)

```




# Alcohol Data   

```{r}
ALQ_I = nhanes_load_data("ALQ_I", "2015-2016", demographics = FALSE)
write.csv(x = ALQ_I, file = "ALQ_I.csv",row.names = FALSE)
skim(ALQ_I)
```


```{r}

ALQ_data = DEMO_I %>% 
  left_join(ALQ_I %>% select(SEQN, ALQ130), by = "SEQN")

```


```{r}
ALQ_data %>% 
  select(age_cluster, FIPR_cluster, ALQ130) %>% 
  ggpairs()
```



```{r}
ALQ_data %>% 
  drop_na() %>% 
  filter(ALQ130 <= 100) %>% 
  select(age_cluster, FIPR_cluster, ALQ130) %>% 
  mutate(age_cluster = factor(age_cluster, levels = c(1,2,3),ordered = TRUE),
         FIPR_cluster = factor(FIPR_cluster, levels = c(1,2,3),ordered = TRUE)) %>% 
  ggpairs()
```

```{r}
ALQ_data %<>%
  filter(ALQ130 < 100) %>% 
  mutate(ALQ130 = ifelse(is.na(ALQ130)==TRUE,0,ALQ130))
```

```{r}
glimpse(ALQ_data)
```


```{r}
ALQ_data %<>% select(-SEQN, -WTINT2YR, -WTMEC2YR) 
set.seed(46)
# Put 3/4 of the data into the training set 
data_split <- initial_split(ALQ_data, prop = 3/4)

# Create data frames for the two sets:
train_data <- training(data_split)
test_data  <- testing(data_split)
```


```{r}
#Download albumin data
#ALB_CR_I = nhanes_load_data("ALB_CR", "2015-2016", demographics = FALSE)

#Download smoking (cotinine) data
#COT_I = nhanes_load_data("COT", "2015-2016", demographics = FALSE)

#Download Blood Pressure & Cholesterol (BPQ_I) data
#BPQ_I = nhanes_load_data("BPQ_I", "2015-2016", demographics = FALSE)

#Download Blood Pressure measurement (BPX_I) data
#BPX_I = nhanes_load_data("BPX_I", "2015-2016", demographics = FALSE)

#Download Body Measures (BMX_I) data
#BMX_I = nhanes_load_data("BMX", "2015-2016", demographics = FALSE)

#dowload alcohol data
#ALQ = nhanes_load_data("ALQ", "2015-2016", demographics = FALSE)

#dowload arsenic data
#UAS = nhanes_load_data("UAS", "2015-2016", demographics = FALSE)

#Download arsenic total data
#UTAS_I = nhanes_load_data("UTAS_I", "2015-2016", demographics = FALSE)

#dowload mercury data
#UHG = nhanes_load_data("UHG", "2015-2016", demographics = FALSE)

#Download urine heavy metal data
#UMS_I = nhanes_load_data("UMS_I", "2015-2016", demographics = FALSE)
```

```{r}
# SEQN - Respondent Sequence Number
# LBXCOT - Cotinine, Serum (ng/mL)
# LBDCOTLC - Cotinine, Serum Comment Code
# LBXHCT - Hydroxycotinine, Serum (ng/mL)
# LBDHCTLC - Hydroxycotinine, Serum Comment Code
# names(DEMO_I)[c(1,2,3,4,5,8,17,18,40,41,42,43,44,47)]

nhanes_an
```