DATA 606 Data Project Proposal
Eric Lehmphul
10/31/2021
Data Preparation
Retrieved dataset from Kaggle: https://www.kaggle.com/sulianova/cardiovascular-disease-dataset. It is a dataset relating to cardiovascular disease and relative variables of interest.
library(tidyverse)
library(GGally)
# load data
url <- "https://raw.githubusercontent.com/SaneSky109/DATA606/main/Data_Project/Data/cardio_train.csv"
cardio.data <- read.csv(url, sep = ";")# remove unecessary column: id
cardio.data <- cardio.data[,-1]# create factors
cardio.data$cardio <- factor(cardio.data$cardio)
cardio.data$gender <- factor(cardio.data$gender)
cardio.data$cholesterol <- factor(cardio.data$cholesterol)
cardio.data$gluc <- factor(cardio.data$gluc)
cardio.data$smoke <- factor(cardio.data$smoke)
cardio.data$alco <- factor(cardio.data$alco)
cardio.data$active <- factor(cardio.data$active)# rename factor levels
levels(cardio.data$cardio) <- c("No", "Yes")
levels(cardio.data$gender) <- c("Female", "Male")
levels(cardio.data$cholesterol) <- c("Norm", "Higher", "Highest")
levels(cardio.data$gluc) <- c("Norm", "Higher", "Highest")
levels(cardio.data$smoke) <- c("No", "Yes")
levels(cardio.data$alco) <- c("No", "Yes")
levels(cardio.data$active) <- c("No", "Yes")# transform age since it is in days
cardio.data$age <- cardio.data$age/365
# remove outliers of ap_hi
# I am assuming the that these measures are errors and
# I am just dropping them due to problems it will cause with modeling
# Highest pressure recorded in an individual was 370/360.(https://pubmed.ncbi.nlm.nih.gov/7741618/)
summary(cardio.data$ap_hi)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## -150.0 120.0 120.0 128.8 140.0 16020.0cardio.data <- cardio.data[cardio.data$ap_hi <= 370,]
cardio.data <- cardio.data[cardio.data$ap_hi > 0,]
summary(cardio.data$ap_hi)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 1.0 120.0 120.0 126.7 140.0 309.0# remove outliers of ap_lo
summary(cardio.data$ap_lo)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## -70.00 80.00 80.00 96.65 90.00 11000.00cardio.data <- cardio.data[cardio.data$ap_lo <= 360,]
cardio.data <- cardio.data[cardio.data$ap_lo > 0,]
summary(cardio.data$ap_lo)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 1.00 80.00 80.00 81.35 90.00 190.00glimpse(cardio.data)## Rows: 68,985
## Columns: 12
## $ age <dbl> 50.39178, 55.41918, 51.66301, 48.28219, 47.87397, 60.03836~
## $ gender <fct> Male, Female, Female, Male, Female, Female, Female, Male, ~
## $ height <int> 168, 156, 165, 169, 156, 151, 157, 178, 158, 164, 169, 173~
## $ weight <dbl> 62, 85, 64, 82, 56, 67, 93, 95, 71, 68, 80, 60, 60, 78, 95~
## $ ap_hi <int> 110, 140, 130, 150, 100, 120, 130, 130, 110, 110, 120, 120~
## $ ap_lo <int> 80, 90, 70, 100, 60, 80, 80, 90, 70, 60, 80, 80, 80, 70, 9~
## $ cholesterol <fct> Norm, Highest, Highest, Norm, Norm, Higher, Highest, Highe~
## $ gluc <fct> Norm, Norm, Norm, Norm, Norm, Higher, Norm, Highest, Norm,~
## $ smoke <fct> No, No, No, No, No, No, No, No, No, No, No, No, No, No, Ye~
## $ alco <fct> No, No, No, No, No, No, No, No, No, No, No, No, No, No, Ye~
## $ active <fct> Yes, Yes, No, Yes, No, No, Yes, Yes, Yes, No, Yes, Yes, No~
## $ cardio <fct> No, Yes, Yes, Yes, No, No, No, Yes, No, No, No, No, No, No~Research question
My research question is: Are gender, age, body weight, body height, blood pressure, cholesterol, glucose levels, smoking, drinking alcohol and activity level of an individual associated with the likelihood of contracting cardiovascular disease?
I plan to use a logistic regression model to answer this research question since the target variable is a binary outcome.
Cases
The cases are the number of people who participate in the medical examination. There were a total of 70,000 cases in the original data file. After data pre-processing, the number of cases is 68,985. This change is due to the removal of rows that seemed to be errors such as extremely high and low blood pressure (-1,000 or 15,000).
nrow(cardio.data)## [1] 68985Data collection
The data was collected from medical information given by patient and examination results. “All of the dataset values were collected at the moment of medical examination.” (https://www.kaggle.com/sulianova/cardiovascular-disease-dataset)
The data was downloaded from Kaggle (https://www.kaggle.com/sulianova/cardiovascular-disease-dataset) and then I uploaded it to Github to be used to import the data into R.
Type of study
This is an observational study since the analysis is on events that have already occurred.
Data Source
The link to where I retrieved the data is: https://www.kaggle.com/sulianova/cardiovascular-disease-dataset
Dependent Variable
The response variable is cardio. This is a qualitative variable since it is a categorical binary variable. cardio is an indicator variable that indicates whether or not someone has cardiovascular disease.
Independent Variable
There are multiple variables that I am considering for analysis. The list contains a group of both quantitative and qualitative variables:
age(quantitative): Age of patient in yearsgender(qualitative): Gender of patientheight(quantitative): Height of patient in cmweight(quantitative): Weight of patient in kgap_hi(quantitative): Systolic blood pressureap_lo(quantitative): Diastolic blood pressurecholesterol(qualitative): Cholesterol level of patientgluc(qualitative): Glucose level of patientsmoke(qualitative): Binary variable to determine if a patient smokesalco(qualitative): Binary variable to determine if a patient drinks alcoholactive(qualitative): Yes/No if patient is physically active
Relevant Summary Statistics
Summary Statistics
summary(cardio.data)## age gender height weight
## Min. :29.58 Female:44932 Min. : 55.0 Min. : 11.00
## 1st Qu.:48.37 Male :24053 1st Qu.:159.0 1st Qu.: 65.00
## Median :53.98 Median :165.0 Median : 72.00
## Mean :53.33 Mean :164.4 Mean : 74.12
## 3rd Qu.:58.42 3rd Qu.:170.0 3rd Qu.: 82.00
## Max. :64.97 Max. :250.0 Max. :200.00
## ap_hi ap_lo cholesterol gluc smoke
## Min. : 7.0 Min. : 1.00 Norm :51747 Norm :58650 No :62924
## 1st Qu.:120.0 1st Qu.: 80.00 Higher : 9339 Higher : 5088 Yes: 6061
## Median :120.0 Median : 80.00 Highest: 7899 Highest: 5247
## Mean :126.3 Mean : 81.35
## 3rd Qu.:140.0 3rd Qu.: 90.00
## Max. :240.0 Max. :190.00
## alco active cardio
## No :65288 No :13571 No :34844
## Yes: 3697 Yes:55414 Yes:34141
##
##
##
## Visualizations
This ggpairs plot of all the variables present in the dataset I wish to analyze is extremely difficult to read due to the number of variables. I will break the graph into smaller pieces below to further explore the data.
entire.plot <- ggpairs(cardio.data)
entire.plot
Cardiovascular Disease Outcome vs All Variables
Boxplots
- Age: Age may be strongly associated with determining cardiovascular disease due to the boxplots (A) for the two categories differing from one another.
- Height: Height may not be strongly associated with determining cardiovascular disease due to the boxplots (B) for the two categories appear to be almost identical. There are a large number of outliers for Height.
- Weight: Weight may be strongly associated with determining cardiovascular disease due to the boxplots (C) for the two categories differing from one another. There are a large number of outliers for Weight.
- Systolic blood pressure: Systolic blood pressure may be strongly associated with determining cardiovascular disease due to the boxplots (D) for the two categories differing from one another.
- Diastolic blood pressure: Diastolic blood pressure may be strongly associated with determining cardiovascular disease due to the boxplots (E) for the two categories differing from one another.
library(ggpubr)
# Age
age1 <- ggplot(cardio.data, aes(x = cardio, y = age)) +
geom_boxplot() +
ggtitle("Boxplot: Age")
# Height
height1 <- ggplot(cardio.data, aes(x = cardio, y = height)) +
geom_boxplot() +
ggtitle("Boxplot: Height")
# Weight
weight1 <- ggplot(cardio.data, aes(x = cardio, y = weight)) +
geom_boxplot() +
ggtitle("Boxplot: Weight")
# Systolic blood pressure
sbp <- ggplot(cardio.data, aes(x = cardio, y = ap_hi)) +
geom_boxplot() +
ggtitle("Boxplot: Systolic blood pressure")
# Diastolic blood pressure
dbp <- ggplot(cardio.data, aes(x = cardio, y = ap_lo)) +
geom_boxplot() +
ggtitle("Boxplot: Diastolic blood pressure")
# Put plots together
ggarrange(age1, height1, weight1, sbp, dbp,
labels = c("A","B","C","D","E"),
ncol = 3, nrow = 2)
Histograms and Bar Plots
Age, Gender, Height, Weight
- Age: A larger portion of people who have cardiovascular disease are older than people who do not have cardiovascular disease, according to the histograms below.
- Gender; There appears to be a substantially larger number of females than males present in this dataset. I wonder if this implies that women are more prone to heart conditions. The number of Male and Female partipants seems to be fairly equal in both factor levels of
cardio - Height: For both factor levels of
cardio, theheightappear to follow a bell curve - Weight: For both factor levels of
cardio, most of the weights fall between 50 and 100 kg.
primary_var <- "cardio"
pairs <- ggpairs(cardio.data, columns = c(1:4,12))
pvar_pos <- match(primary_var, pairs$xAxisLabels)
plots <- lapply(1:pairs$ncol, function(j) getPlot(pairs, i = pvar_pos, j = j))
matrix1 <- ggmatrix(
plots,
nrow = 1,
ncol = pairs$ncol-1,
xAxisLabels = pairs$xAxisLabels[1:4],
title = "Cardiovascular Disease vs Age, Gender, Height, and Weight"
)
matrix1
Systolic blood pressure, Diastolic blood pressure, Cholesterol, Smoke
- Systolic blood pressure: A large portion of the data 100 and 150. For people who don’t have cardiovascular disease, there is a unimodal peak that is about 2 to 2.5 times the length of any other histogram bin.
- Diastolic blood pressure: A large portion of the data 50 and 100. For people who don’t have cardiovascular disease, there is a unimodal peak that is about 2 to 2.5 times the length of any other histogram bin.
- Cholesterol: There is an unequal distribution of factor levels of cholesterol. Most patients have a normal cholesterol level.
- Glucose: There is an unequal distribution of factor levels of glucose. Most patients have a normal glucose level.
primary_var <- "cardio"
pairs <- ggpairs(cardio.data, columns = c(5:8,12))
pvar_pos <- match(primary_var, pairs$xAxisLabels)
plots <- lapply(1:pairs$ncol, function(j) getPlot(pairs, i = pvar_pos, j = j))
matrix1 <- ggmatrix(
plots,
nrow = 1,
ncol = pairs$ncol-1,
xAxisLabels = pairs$xAxisLabels[1:4],
title = "Cardiovascular Disease vs Systolic blood pressure,
Diastolic blood pressure, Cholesterol, Glucose"
)
matrix1
Smoke, Alcohol, and Activity level:
For all 3 variables, there is an uneven distribution of factor levels. There is a much larger proportion of non smokers in this dataset. The dataset contains mostly non alcoholics. The dataset has a larger number of active people.
primary_var <- "cardio"
pairs <- ggpairs(cardio.data, columns = c(9:11,12))
pvar_pos <- match(primary_var, pairs$xAxisLabels)
plots <- lapply(1:pairs$ncol, function(j) getPlot(pairs, i = pvar_pos, j = j))
matrix1 <- ggmatrix(
plots,
nrow = 1,
ncol = pairs$ncol-1,
xAxisLabels = pairs$xAxisLabels[1:3],
title = "Cardiovascular Disease vs Smoke,
Alcohol, Activity Level"
)
matrix1
Numerical Variables: Scatterplots, Density Plots, Correlation Matrix
There does not appear to be any strong correlation between any of the numeric variables I plan to use as independent variables. The strongest correlation exists between the Systolic blood pressure and Diastolic blood pressure with a correlation of 0.65.
Since the explanatory variables are not strongly correlated with each other, there is not an issue with multicollinearity in the data/model.
ggpairs(cardio.data, columns = c(1,3:6))
Mosaic Plots
Cardiovascular Disease Outcome by Age
plot(cardio.data$cardio ~ cardio.data$age,
xlab = "Age",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Age")
Cardiovascular Disease Outcome by Gender
plot(cardio.data$cardio ~ cardio.data$gender,
xlab = "Gender",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Gender")
Cardiovascular Disease Outcome by Height
plot(cardio.data$cardio ~ cardio.data$height,
xlab = "Height",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Height")
Cardiovascular Disease Outcome by Weight
plot(cardio.data$cardio ~ cardio.data$weight,
xlab = "Weight",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Weight")
Cardiovascular Disease Outcome by Systolic blood pressure
plot(cardio.data$cardio ~ cardio.data$ap_hi,
xlab = "Systolic Blood Pressure",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Systolic Blood Pressure")
Cardiovascular Disease Outcome by Diastolic blood pressure
plot(cardio.data$cardio ~ cardio.data$ap_hi,
xlab = "Diastolic Blood Pressure",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Diastolic Blood Pressure")
Cardiovascular Disease Outcome by Cholesterol Level
plot(cardio.data$cardio ~ cardio.data$cholesterol,
xlab = "Cholesterol Level",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Cholesterol Level")
Cardiovascular Disease Outcome by Glucose Level
plot(cardio.data$cardio ~ cardio.data$gluc,
xlab = "Glucose Level",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Glucose Level")
Cardiovascular Disease Outcome by Smoking Level
plot(cardio.data$cardio ~ cardio.data$smoke,
xlab = "Smoking",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Smoking")
Cardiovascular Disease Outcome by Alcohol Level
plot(cardio.data$cardio ~ cardio.data$alco,
xlab = "Alcohol",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Alcohol Level")
Cardiovascular Disease Outcome by Activity Level
plot(cardio.data$cardio ~ cardio.data$active,
xlab = "Active",ylab = "Cardio",
main = "Cardiovascular Outcome vs. Activity Level")