RPubs link information

• Rpubs link comes here: Link

Introduction

• Heart failure happens when heart cannot circulate enough blood to supply the body’s requirements. It is one life-threatening condition where even with advances in healthcare, it continues to be a leading driver of hospitalisation and death worldwide (AHA 2025).
• Patients with heart failure often present with a range of clinical characteristics, including age, platelets, anaemia, and blood pressure.
  
• Ejection fraction is also one of a key measure of heart function and indicates the count of blood heart ejected during each heartbeat. Another common risk factor is high blood pressure that may contribute to worsening heart health.
• Heart failure cannot be cure but understanding which patient characteristics are associated with mortality may help healthcare professionals identify high-risk patients on time and improve clinical decision-making.

Problem Statement

• Is it true that low ejection fraction can lead to death? Is there a relationship between high blood pressure mortality among patients with heart failure?
• An open dataset of Heart Failure Clinical Records will be analysed and used to address these questions.
• RStudio will be used to do the statistical analysis such as:
  • Descriptive statistics and visualisations to explore the data.
  • Compare mean ejection fraction between groups of recorded patients by using a two-sample t-test.
    
  • Examine the relationship between high blood pressure and death event by using a Chi-square test of association.

Dataset Descriptions

• The data used in this study were obtained from the UC Irvine Machine Learning Repository’s Heart Failure Clinical Records Dataset (UCI 2020). It is an observational data collected from patients receiving treatment for heart failure. This includes demographic, laboratory, and medical information.
• It consists of 299 observations and 13 variables including:
  • Age: age of the patient measured in years.
  • Anaemia: whether the patient was diagnosed with anaemia (1 = Yes, 0 = No).
  • creatinine_phosphokinase (CPK): CPK enzyme amount in the blood, measured in mcg/L
  • diabetes: diabetes or not (1 = Yes, 0 = No)
  • ejection_fraction: proportion of blood ejected from the heart with beat (%)
  • serum_sodium: Amount of sodium present in the blood, measured in mEq/L
  • high_blood_pressure: whether patients has high blood pressure (1 = Yes, 0 = No)
  • platelets: Platelet count in the bloodstream, kiloplatelets per millilitre
  • sex: 0 = Female, 1 = Male
  • smoking: smoke or not (1 = Yes, 0 = No)
  • time: day of patients checking in
  • DEATH_EVENT: Patient outcome (0 = Survived, 1 = Died)
  • serum_creatinine: serum creatinine levels within the bloodstream measured mg/dL
• This analysis will only used 3 variables including ejection_fraction, high_blood_pressure, and death_event.

Data Preprocessing

# Import the dataset
heart <- read_csv("heart.csv")

# Check the dataset
head(heart, n=5)

# Convert binary variables use in this analysis to factor and label 
heart$DEATH_EVENT <- heart$DEATH_EVENT %>% 
      factor(levels=c(0,1), labels=c("Survived","Died"))
heart$high_blood_pressure <- heart$high_blood_pressure %>% 
      factor(levels = c(0,1), labels = c("No","Yes"))
levels(heart$DEATH_EVENT)

## [1] "Survived" "Died"

levels(heart$high_blood_pressure)

## [1] "No"  "Yes"

Data Preprocessing CONT

# Check missing values
colSums(is.na(heart)) -> table1

knitr::kable(table1)

• No missing values in the dataset

Descriptive Statistics and Visualizations

# Ejection_fraction summary statistic 
heart %>% group_by (DEATH_EVENT) %>%  summarise(Min = min(ejection_fraction, na.rm = TRUE),
                                                Q1 = quantile(ejection_fraction,probs = .25, na.rm = TRUE),
                                                Median = median(ejection_fraction, na.rm = TRUE),
                                                Mean = mean(ejection_fraction, na.rm = TRUE),
                                                Q3 = quantile(ejection_fraction,probs = .75, na.rm = TRUE),
                                                Max = max(ejection_fraction, na.rm = TRUE),
                                                SD = sd(ejection_fraction, na.rm = TRUE),
                                                n = n(),
                                                Missing = sum(is.na(ejection_fraction))) -> table2
knitr::kable(table2)

• Survivors had a higher mean ejection fraction of 40.27% compared with 33.47% among patients who died.
• The median ejection fraction was also higher for survivors at 38% compared with 30% for deceased patients.
• The deceased group showed greater variability in ejection fraction values, with a standard deviation of 12.53 compared with 10.86 for survivors.
• Among the recorded patients, 203 survived and 96 died during the study period.

Descriptive Statistics and Visualizations CONT

# Frequency table for high_blood_pressure by Death_Event
heart %>% xtabs(~ high_blood_pressure + DEATH_EVENT, data = .) %>% prop.table(2) %>% addmargins() -> table3
knitr::kable(table3)

• Among patients who survived, 67.49% did not have high blood pressure and 32.51% had high blood pressure.
• Among patients who died, 59.38% did not have high blood pressure and 40.63% had high blood pressure.
• The proportion of patients with high blood pressure was higher in the deceased group, suggesting a possible association between high blood pressure and mortality. However, a chi-square test is required to determine whether this relationship is statistically significant.

Descriptive Statistics and Visualizations CONT

# Box plot of Ejection_Fractions
boxplot(ejection_fraction ~ DEATH_EVENT, data = heart,
        ylab = "Ejection Fraction (%)", xlab = "Death Event",
        main = "Ejection Fraction by Death Event")

• Survivors had a higher median ejection fraction (approximately 38%) than patients who died ( approximately 30%). The deceased group showed slightly greater variability in ejection fraction values than the survived group.
• Several high-value outliers were identified but retained because they were valid clinical observations.

Decsriptive Statistics and Visualizations CONT

# Histogram of ejection_fraction to check skewness
hist(heart$ejection_fraction,
     main = "Distribution of Ejection Fraction", xlab = "Ejection Fraction (%)")
abline(v = mean(heart$ejection_fraction), col = "red", lwd = 2)
abline(v = median(heart$ejection_fraction), col = "blue", lwd = 2)

• Most patients had ejection fraction values between 20% and 50%, with the highest concentration around 35% to 40%. The distribution is positively skewed. The dataset has 299 observations which meet Central Limit Theorem. If not, a transformation might needed to remove outliers and make the distribution normal.

Hypothesis Testing

Two Sample test on the mean

• To compare average ejection fraction levels between survivors and deceased patients, a two-sample t-test was performed.
• mu_1 = mean ejection fraction is the same for patients who survived.
• mu_2 = mean ejection fraction is the same for patients who died.
• State the hypothesis: \[H_0: \mu_1 = \mu_2 \] \[H_A: \mu_1 \ne \mu_2\]

Two Sample test on the mean CONT

# Checking assumptions
## Normality
ggqqplot(heart, x="ejection_fraction", facet.by = "DEATH_EVENT")

• Although the histogram indicated that ejection fraction was not normally distributed, both groups contained more than 30 observations. Therefore, the Central Limit Theorem supports the assumption that the sampling distribution of the mean is approximately normal.

Two Sample test on the mean CONT

# Checking assumptions
leveneTest(ejection_fraction ~ DEATH_EVENT, data = heart)

• According to Levene’s Test, P-value = 0.0553> 0.05. There is insufficient evidence to conclude that the variances differ between the survived and deceased groups.Therefore, the equal variance can be used.

Two Sample test on the mean CONT

# Conduct t-test with equal variance with 95% CI
t.test(ejection_fraction ~ DEATH_EVENT, data = heart, 
                var.equal = TRUE, 
                alternative = "two.sided")

## 
##  Two Sample t-test
## 
## data:  ejection_fraction by DEATH_EVENT
## t = 4.8056, df = 297, p-value = 2.453e-06
## alternative hypothesis: true difference in means between group Survived and group Died is not equal to 0
## 95 percent confidence interval:
##  4.013671 9.580849
## sample estimates:
## mean in group Survived     mean in group Died 
##               40.26601               33.46875

• Decisions: reject the H_0: mu1 = mu2 as the p-value = 2.453e-06 < 0.05 and 95% confidence interval is [4.013671, 9.580849] does not contains 0.
• Conclusion: There is sufficient evidence to conclude that the mean ejection fraction is significantly different between patients who survived and patients who died.

Hypthesis Testing Cont.

Chi-square Test of Association

• Chi-square test of association was use to identify the relationship between high blood pressure status and death event among heart failure patients.
• H_0: there is no association between high blood pressure status and death event
• H_1: there is an association between high blood pressure status and death event

# Checking assumption: No more than 25% of the cells in the contingency table should have expected frequencies less than 5
assume <- table(heart$high_blood_pressure,
             heart$DEATH_EVENT)

chi_assume <- chisq.test(assume)
chi_assume$observed

##      
##       Survived Died
##   No       137   57
##   Yes       66   39

chi_assume$expected

##      
##        Survived     Died
##   No  131.71237 62.28763
##   Yes  71.28763 33.71237

• All cells in the contingency table has expected count above 5. Therefore, the assumption is checked.

Chi-square Test of Association CONT

• find degree of freedom \[df=(row-1)(column-1)\] \[df=(2-1)(2-1)\] \[df=1\]

# P-value
chi_assume$p.value

## [1] 0.2141034

• Decision: As p-value = 0.214 greater than 0.05 level of significant, Failed to reject H_0.
• Conclusion: There is not enough evidence to conclude that high blood pressure and death event are associated among heart failure patients.

Discussion

• The findings revealed that survivors generally exhibited better heart function, with a mean ejection fraction of 40.26 compared with 33.46 for patients who died. This difference was statistically significant (p < 0.05), highlighting the potential importance of ejection fraction as an indicator of patient outcomes.
• The Chi-square test on the other hand found no significant association between high blood pressure and death event (p = 0.214). Although high blood pressure is a known risk factor for heart disease, the results suggest that hypertension alone may not be a strong predictor of mortality within this sample of heart failure patients.
• Overall, this study found that low ejection fraction was significantly associated with mortality, while high blood pressure has no relationship with the death outcome.
• A key strength of this analysis is its practical relevance to healthcare decision-making. The results highlight the importance of ejection fraction as an indicator of mortality risk, which may assist clinicians in identifying vulnerable patients and supporting treatment planning.
• However, the dataset is small with only 299 patients and the analysis only considered ejection fraction and high blood pressure. Other clinical factors that may influence mortality were not included and should be considered with a larger dataset in future studies.

References

AHA (American Heart Association) (2025) What is Heart Failure?, American Heart Association website, accessed 26 May 2026. https://www.heart.org/en/health-topics/heart-failure/what-is-heart-failure

UCI Machine Learning Repository (2020) Heart Failure Clinical Records, UCI Machine Learning Repository website, accessed 26 May 2026. https://archive.ics.uci.edu/dataset/519/heart-failure-clinical-records