Show Code
knitr::opts_chunk$set(
dev = "png",
dpi = 150,
fig.path = "figures1/"
)
dir.create("figures1", showWarnings = FALSE)Clinical Data Exploratory Analytics Pipeline
Data Cleaning · Hypothesis Testing · Visualization — UCI Heart Disease Dataset
Pipeline Overview
1 Stage 1 — Data Cleaning & Preparation
1.1 Data Ingestion
Show Code
# Step 0 — define column names (used by recoding + type audit chunks below)
col_names <- c("age", "sex", "cp", "trestbps", "chol", "fbs", "restecg",
"thalach", "exang", "oldpeak", "slope", "ca", "thal", "target")
# Step 1 — load real dataset
raw <- read.csv("C:\\Users\\PC\\Downloads\\archive (9)\\heart_cleveland_upload.csv",
header = TRUE,
na.strings = "?")
# Step 2 — rename 'condition' column to 'target'
names(raw)[names(raw) == "condition"] <- "target"
# Step 3 — confirm it loaded correctly
cat(sprintf("Dataset loaded: %d rows × %d columns\n", nrow(raw), ncol(raw)))Dataset loaded: 297 rows × 14 columns1.2 Raw Data Snapshot
Show Code
raw |>
head(10) |>
kable(caption = "Table 1.1 — First 10 rows of raw clinical data") |>
kable_styling(bootstrap_options = c("striped","hover","condensed"),
font_size = 12, full_width = TRUE) |>
column_spec(14, bold = TRUE,
background = ifelse(head(raw,10)$target == 1, "#FDEDEC", "#D5F5E3"))| age | sex | cp | trestbps | chol | fbs | restecg | thalach | exang | oldpeak | slope | ca | thal | target |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 69 | 1 | 0 | 160 | 234 | 1 | 2 | 131 | 0 | 0.1 | 1 | 1 | 0 | 0 |
| 69 | 0 | 0 | 140 | 239 | 0 | 0 | 151 | 0 | 1.8 | 0 | 2 | 0 | 0 |
| 66 | 0 | 0 | 150 | 226 | 0 | 0 | 114 | 0 | 2.6 | 2 | 0 | 0 | 0 |
| 65 | 1 | 0 | 138 | 282 | 1 | 2 | 174 | 0 | 1.4 | 1 | 1 | 0 | 1 |
| 64 | 1 | 0 | 110 | 211 | 0 | 2 | 144 | 1 | 1.8 | 1 | 0 | 0 | 0 |
| 64 | 1 | 0 | 170 | 227 | 0 | 2 | 155 | 0 | 0.6 | 1 | 0 | 2 | 0 |
| 63 | 1 | 0 | 145 | 233 | 1 | 2 | 150 | 0 | 2.3 | 2 | 0 | 1 | 0 |
| 61 | 1 | 0 | 134 | 234 | 0 | 0 | 145 | 0 | 2.6 | 1 | 2 | 0 | 1 |
| 60 | 0 | 0 | 150 | 240 | 0 | 0 | 171 | 0 | 0.9 | 0 | 0 | 0 | 0 |
| 59 | 1 | 0 | 178 | 270 | 0 | 2 | 145 | 0 | 4.2 | 2 | 0 | 2 | 0 |
1.3 Missing Value Audit
Show Code
# Tabular summary
miss_summary <- raw |>
summarise(across(everything(),
list(N_Missing = ~sum(is.na(.)),
Pct_Missing = ~round(mean(is.na(.))*100, 2)))) |>
pivot_longer(everything(),
names_to = c("Variable","Stat"),
names_sep = "_(?=N_|Pct_)") |>
pivot_wider(names_from = Stat, values_from = value) |>
filter(N_Missing > 0)
miss_summary |>
rename(`Variable` = Variable,
`Missing (n)` = N_Missing,
`Missing (%)` = Pct_Missing) |>
kable(caption = "Table 1.2 — Variables with missing data") |>
kable_styling(bootstrap_options = c("striped","hover"),
full_width = FALSE) |>
row_spec(0, bold = TRUE) |>
column_spec(3, color = "white",
background = "#E74C3C")| Variable | Missing (n) | Missing (%) |
|---|---|---|
| NA | NA | NA |
| :——– | ———–: | ———–: |
Figure 1.1 — Missing value heatmap
Show Code
# Visual heatmap
miss_matrix <- is.na(raw) |> as.data.frame() |>
mutate(row_id = row_number()) |>
pivot_longer(-row_id, names_to = "Variable", values_to = "Missing")
ggplot(miss_matrix |> filter(Missing),
aes(x = Variable, y = row_id)) +
geom_tile(fill = CLINICAL_RED, alpha = 0.8) +
scale_y_reverse() +
labs(title = "Missing Value Locations",
subtitle = sprintf("Total missing cells: %d (%.1f%% of data)",
sum(is.na(raw)),
mean(is.na(raw))*100),
x = "Variable", y = "Row Index") +
theme(axis.text.x = element_text(angle = 45, hjust = 1))
1.4 Data Type Audit & Recoding
Show Code
# Record raw column types
type_audit <- data.frame(
Variable = col_names,
Raw_Type = sapply(raw, class),
Description = c("Age (years)","Sex (0=F, 1=M)",
"Chest pain type (1-4)",
"Resting blood pressure (mmHg)",
"Serum cholesterol (mg/dL)",
"Fasting blood sugar > 120 mg/dL (0/1)",
"Resting ECG results (0-2)",
"Max heart rate achieved (bpm)",
"Exercise-induced angina (0/1)",
"ST depression (oldpeak)",
"Slope of peak ST segment (1-3)",
"Number of major vessels (0-3)",
"Thalassemia type (3/6/7)",
"Target: heart disease (0/1)"),
Recoded_To = c("numeric","factor","factor","numeric","numeric",
"factor","factor","numeric","factor","numeric",
"factor","numeric","factor","factor"),
row.names = NULL
)
type_audit |>
kable(caption = "Table 1.3 — Variable type audit and recoding plan") |>
kable_styling(bootstrap_options = c("striped","hover","condensed")) |>
row_spec(which(type_audit$Raw_Type != type_audit$Recoded_To),
background = "#FEF9E7")| Variable | Raw_Type | Description | Recoded_To |
|---|---|---|---|
| age | integer | Age (years) | numeric |
| sex | integer | Sex (0=F, 1=M) | factor |
| cp | integer | Chest pain type (1-4) | factor |
| trestbps | integer | Resting blood pressure (mmHg) | numeric |
| chol | integer | Serum cholesterol (mg/dL) | numeric |
| fbs | integer | Fasting blood sugar > 120 mg/dL (0/1) | factor |
| restecg | integer | Resting ECG results (0-2) | factor |
| thalach | integer | Max heart rate achieved (bpm) | numeric |
| exang | integer | Exercise-induced angina (0/1) | factor |
| oldpeak | numeric | ST depression (oldpeak) | numeric |
| slope | integer | Slope of peak ST segment (1-3) | factor |
| ca | integer | Number of major vessels (0-3) | numeric |
| thal | integer | Thalassemia type (3/6/7) | factor |
| target | integer | Target: heart disease (0/1) | factor |
1.5 Cleaned Dataset
Show Code
df <- raw |>
# Handle missing: median imputation for numeric, mode for factor
mutate(
ca = ifelse(is.na(ca), median(ca, na.rm = TRUE), ca),
thal = ifelse(is.na(thal), as.numeric(names(sort(table(thal),
decreasing=TRUE))[1]), thal)
) |>
# Recode categoricals with labels
mutate(
sex = factor(sex, levels = 0:1, labels = c("Female","Male")),
fbs = factor(fbs, levels = 0:1, labels = c("≤120 mg/dL",">120 mg/dL")),
exang = factor(exang, levels = 0:1, labels = c("No","Yes")),
restecg = factor(restecg, levels = 0:2,
labels = c("Normal","ST-T Abnormality","LV Hypertrophy")),
slope = factor(slope, levels = 1:3,
labels = c("Upsloping","Flat","Downsloping")),
cp = factor(cp, levels = 0:3, labels = c("Typical Angina","Atypical Angina",
"Non-Anginal Pain","Asymptomatic")),
thal = factor(thal, levels = c(3,6,7),
labels = c("Normal","Fixed Defect","Reversable Defect")),
target = factor(target, levels = 0:1,
labels = c("No Disease","Disease"))
)
cat(sprintf("✔ Clean dataset: %d rows × %d columns | Missing: %d\n",
nrow(df), ncol(df), sum(is.na(df))))✔ Clean dataset: 297 rows × 14 columns | Missing: 436Show Code
# Before / After comparison
data.frame(
Metric = c("Rows","Columns","Missing Values",
"Numeric Variables","Factor Variables"),
Before_Cleaning = c(nrow(raw), ncol(raw), sum(is.na(raw)),
sum(sapply(raw, is.numeric)),
sum(sapply(raw, is.factor))),
After_Cleaning = c(nrow(df), ncol(df), sum(is.na(df)),
sum(sapply(df, is.numeric)),
sum(sapply(df, is.factor)))
) |>
kable(caption = "Table 1.4 — Before vs After cleaning comparison") |>
kable_styling(bootstrap_options = c("striped","hover"),
full_width = FALSE) |>
column_spec(3, bold = TRUE, color = "white", background = CLINICAL_BLUE)| Metric | Before_Cleaning | After_Cleaning |
|---|---|---|
| Rows | 297 | 297 |
| Columns | 14 | 14 |
| Missing Values | 0 | 436 |
| Numeric Variables | 14 | 6 |
| Factor Variables | 0 | 8 |
1.6 Descriptive Statistics
Show Code
# Continuous variables
num_vars <- c("age","trestbps","chol","thalach","oldpeak")
num_labels <- c("Age (years)","Resting BP (mmHg)","Cholesterol (mg/dL)",
"Max HR (bpm)","ST Depression")
df |>
dplyr::select(all_of(num_vars)) |>
describe() |>
dplyr::select(n, mean, sd, median, min, max, skew, kurtosis) |>
mutate(across(where(is.numeric), ~round(., 2))) |>
`rownames<-`(num_labels) |>
kable(caption = "Table 1.5 — Descriptive statistics: continuous variables") |>
kable_styling(bootstrap_options = c("striped","hover","condensed"))| n | mean | sd | median | min | max | skew | kurtosis | |
|---|---|---|---|---|---|---|---|---|
| Age (years) | 297 | 54.54 | 9.05 | 56.0 | 29 | 77.0 | -0.22 | -0.55 |
| Resting BP (mmHg) | 297 | 131.69 | 17.76 | 130.0 | 94 | 200.0 | 0.69 | 0.76 |
| Cholesterol (mg/dL) | 297 | 247.35 | 52.00 | 243.0 | 126 | 564.0 | 1.11 | 4.30 |
| Max HR (bpm) | 297 | 149.60 | 22.94 | 153.0 | 71 | 202.0 | -0.53 | -0.09 |
| ST Depression | 297 | 1.06 | 1.17 | 0.8 | 0 | 6.2 | 1.23 | 1.44 |
Show Code
# Categorical variables
cat_vars <- c("sex","cp","fbs","restecg","exang","slope","thal","target")
cat_summary <- map_dfr(cat_vars, function(v) {
df |>
count(.data[[v]]) |>
mutate(Variable = v,
Level = as.character(.data[[v]]),
Pct = round(n / sum(n) * 100, 1)) |>
dplyr::select(Variable, Level, n, Pct)
})
cat_summary |>
kable(caption = "Table 1.6 — Frequency table: categorical variables",
col.names = c("Variable","Level","Count","Percent (%)")) |>
kable_styling(bootstrap_options = c("striped","hover","condensed"),
font_size = 11) |>
collapse_rows(columns = 1, valign = "middle")| Variable | Level | Count | Percent (%) |
|---|---|---|---|
| sex | Female | 96 | 32.3 |
| Male | 201 | 67.7 | |
| cp | Typical Angina | 23 | 7.7 |
| Atypical Angina | 49 | 16.5 | |
| Non-Anginal Pain | 83 | 27.9 | |
| Asymptomatic | 142 | 47.8 | |
| fbs | ≤120 mg/dL | 254 | 85.5 |
| >120 mg/dL | 43 | 14.5 | |
| restecg | Normal | 147 | 49.5 |
| ST-T Abnormality | 4 | 1.3 | |
| LV Hypertrophy | 146 | 49.2 | |
| exang | No | 200 | 67.3 |
| Yes | 97 | 32.7 | |
| slope | Upsloping | 137 | 46.1 |
| Flat | 21 | 7.1 | |
| NA | 139 | 46.8 | |
| thal | NA | 297 | 100.0 |
| target | No Disease | 160 | 53.9 |
| Disease | 137 | 46.1 |
2 Stage 2 — Hypothesis Testing
2.1 Normality Testing
Show Code
# Anderson-Darling test for each continuous variable
norm_results <- map_dfr(num_vars, function(v) {
x <- df[[v]]
ad <- ad.test(x)
shp <- shapiro.test(x[1:min(length(x), 5000)])
tibble(
Variable = v,
N = length(x),
Mean = round(mean(x), 2),
SD = round(sd(x), 2),
AD_stat = round(ad$statistic, 3),
AD_pval = round(ad$p.value, 4),
SW_pval = round(shp$p.value, 4),
Normal = ifelse(ad$p.value > 0.05, "✔ Yes", "✘ No")
)
})
norm_results |>
rename(`Variable` = Variable, `n` = N,
`Mean` = Mean, `SD` = SD,
`A-D Stat` = AD_stat,
`A-D p-value` = AD_pval,
`S-W p-value` = SW_pval,
`Normal?` = Normal) |>
kable(caption = "Table 2.1 — Normality tests: Anderson-Darling & Shapiro-Wilk") |>
kable_styling(bootstrap_options = c("striped","hover","condensed")) |>
column_spec(8, bold = TRUE,
color = ifelse(norm_results$Normal == "✔ Yes",
CLINICAL_GREEN, CLINICAL_RED))| Variable | n | Mean | SD | A-D Stat | A-D p-value | S-W p-value | Normal? |
|---|---|---|---|---|---|---|---|
| age | 297 | 54.54 | 9.05 | 1.547 | 5e-04 | 0.0054 | ✘ No |
| trestbps | 297 | 131.69 | 17.76 | 2.505 | 0e+00 | 0.0000 | ✘ No |
| chol | 297 | 247.35 | 52.00 | 1.582 | 4e-04 | 0.0000 | ✘ No |
| thalach | 297 | 149.60 | 22.94 | 2.254 | 0e+00 | 0.0001 | ✘ No |
| oldpeak | 297 | 1.06 | 1.17 | 13.420 | 0e+00 | 0.0000 | ✘ No |
2.2 Q-Q Plots
Show Code
qq_plots <- map(seq_along(num_vars), function(i) {
v <- num_vars[i]
lab <- num_labels[i]
x <- df[[v]]
qq <- qqnorm(x, plot.it = FALSE)
qdf <- data.frame(theoretical = qq$x, sample = qq$y)
q_ref <- quantile(qdf$sample, c(0.25, 0.75))
q_thr <- quantile(qdf$theoretical, c(0.25, 0.75))
sl <- diff(q_ref) / diff(q_thr)
ic <- q_ref[1] - sl * q_thr[1]
ggplot(qdf, aes(theoretical, sample)) +
geom_point(colour = CLINICAL_BLUE, alpha = 0.45, size = 1.2) +
geom_abline(slope = sl, intercept = ic, colour = CLINICAL_RED, size = 0.9, linetype = "dashed") +
labs(title = lab, x = "Theoretical Quantiles", y = "Sample Quantiles")
})
grid.arrange(grobs = qq_plots, ncol = 2,
top = grid::textGrob("Q-Q Plots — Continuous Clinical Variables",
gp = grid::gpar(fontface="bold",
col=CLINICAL_BLUE, cex=1.1)))
2.3 Test 1 — Independent Samples t-test: Age by Diagnosis
Show Code
age_disease <- df$age[df$target == "Disease"]
age_nodisease <- df$age[df$target == "No Disease"]
lev_test <- leveneTest(age ~ target, data = df)
t_result <- t.test(age ~ target, data = df, var.equal = FALSE)
# Results table
tibble(
Group = c("No Disease","Disease"),
N = c(length(age_nodisease), length(age_disease)),
Mean_Age = round(c(mean(age_nodisease), mean(age_disease)), 2),
SD_Age = round(c(sd(age_nodisease), sd(age_disease)), 2),
Median_Age = c(median(age_nodisease), median(age_disease))
) |>
kable(caption = "Table 2.2 — Age by diagnosis group",
col.names = c("Group","n","Mean Age","SD","Median")) |>
kable_styling(bootstrap_options = c("striped","hover"),
full_width = FALSE)| Group | n | Mean Age | SD | Median |
|---|---|---|---|---|
| No Disease | 160 | 52.64 | 9.55 | 52 |
| Disease | 137 | 56.76 | 7.90 | 58 |
Show Code
tibble(
Test = "Welch Two-Sample t-test",
Variable = "Age",
`t statistic` = round(t_result$statistic, 3),
df = round(t_result$parameter, 1),
`p-value` = round(t_result$p.value, 4),
`95% CI` = sprintf("[%.2f, %.2f]",
t_result$conf.int[1],
t_result$conf.int[2]),
Significant = ifelse(t_result$p.value < 0.05, "Yes ***", "No"),
Interpretation = "Patients with heart disease are significantly older"
) |>
kable(caption = "Table 2.3 — Welch t-test results: Age by diagnosis") |>
kable_styling(bootstrap_options = c("striped","hover")) |>
column_spec(7, bold = TRUE, color = "white",
background = ifelse(t_result$p.value < 0.05,
CLINICAL_RED, CLINICAL_GREY))| Test | Variable | t statistic | df | p-value | 95% CI | Significant | Interpretation |
|---|---|---|---|---|---|---|---|
| Welch Two-Sample t-test | Age | -4.064 | 294.7 | 1e-04 | [-6.11, -2.12] | Yes *** | Patients with heart disease are significantly older |
Show Code
ggplot(df, aes(target, age, fill = target, colour = target)) +
geom_violin(alpha = 0.25, trim = FALSE, size = 0.7) +
geom_boxplot(width = 0.18, outlier.shape = 21,
outlier.alpha = 0.5, colour = "grey30", fill = "white") +
stat_summary(fun = mean, geom = "point",
shape = 23, size = 4, fill = "white") +
scale_fill_manual(values = DIAG_COLOURS, guide = "none") +
scale_colour_manual(values = DIAG_COLOURS, guide = "none") +
annotate("text", x = 1.5, y = 76,
label = sprintf("t(%.0f) = %.2f\np = %.4f",
t_result$parameter,
t_result$statistic,
t_result$p.value),
size = 3.8, colour = CLINICAL_GREY, fontface = "italic") +
labs(title = "Age Distribution by Diagnosis",
subtitle = "Violin + box plot | diamond = group mean",
x = "Diagnosis", y = "Age (years)")
2.4 Test 2 — Chi-Square: Sex × Diagnosis
Show Code
sex_tab <- table(df$sex, df$target)
chi_sex <- chisq.test(sex_tab)
# Frequency & expected counts
as.data.frame.matrix(sex_tab) |>
mutate(Total = `No Disease` + Disease,
`Disease Rate` = percent(Disease / Total, 0.1)) |>
`rownames<-`(c("Female","Male")) |>
kable(caption = "Table 2.4 — Sex × diagnosis contingency table") |>
kable_styling(bootstrap_options = c("striped","hover"),
full_width = FALSE)| No Disease | Disease | Total | Disease Rate | |
|---|---|---|---|---|
| Female | 71 | 25 | 96 | 26.0% |
| Male | 89 | 112 | 201 | 55.7% |
Show Code
tibble(
Test = "Pearson Chi-Square",
Variables = "Sex × Diagnosis",
`χ² statistic` = round(chi_sex$statistic, 3),
df = chi_sex$parameter,
`p-value` = round(chi_sex$p.value, 4),
`Cramér's V` = round(sqrt(chi_sex$statistic /
(nrow(df) * (min(dim(sex_tab))-1))), 3),
Significant = ifelse(chi_sex$p.value < 0.05, "Yes ***", "No"),
Effect = "Small-Medium"
) |>
kable(caption = "Table 2.5 — Chi-square test: sex and diagnosis") |>
kable_styling(bootstrap_options = c("striped","hover")) |>
column_spec(7, bold = TRUE, color = "white", background = CLINICAL_RED)| Test | Variables | χ² statistic | df | p-value | Cramér's V | Significant | Effect |
|---|---|---|---|---|---|---|---|
| NA | NA | NA | NA | NA | NA | NA | NA |
| :---- | :--------- | ------------: | --: | -------: | ----------: | :----------- | :------ |
Show Code
df |>
count(sex, target) |>
group_by(sex) |>
mutate(prop = n / sum(n)) |>
ggplot(aes(sex, prop, fill = target)) +
geom_col(width = 0.55, colour = "white", size = 0.5) +
geom_text(aes(label = sprintf("%d\n(%.0f%%)", n, prop*100)),
position = position_stack(vjust = 0.5),
fontface = "bold", size = 3.8, colour = "white") +
scale_fill_manual(values = DIAG_COLOURS, name = "Diagnosis") +
scale_y_continuous(labels = percent_format()) +
labs(title = "Heart Disease by Sex",
subtitle = sprintf("χ²(1) = %.2f, p = %.4f | Cramér's V = %.3f",
chi_sex$statistic, chi_sex$p.value,
sqrt(chi_sex$statistic/(nrow(df)*(min(dim(sex_tab))-1)))),
x = "Sex", y = "Proportion")
2.5 Test 3 — Chi-Square: Chest Pain Type × Diagnosis
Show Code
cp_tab <- table(df$cp, df$target)
chi_cp <- chisq.test(cp_tab)
cramer_cp <- sqrt(chi_cp$statistic / (nrow(df) * (min(dim(cp_tab))-1)))
tibble(
Test = "Pearson Chi-Square",
Variables = "Chest Pain × Diagnosis",
`χ² statistic` = round(chi_cp$statistic, 3),
df = chi_cp$parameter,
`p-value` = formatC(chi_cp$p.value, format="e", digits=2),
`Cramér's V` = round(cramer_cp, 3),
Significant = "Yes ***",
Effect = "Large"
) |>
kable(caption = "Table 2.6 — Chi-square: chest pain type and diagnosis") |>
kable_styling(bootstrap_options = c("striped","hover")) |>
column_spec(7, bold = TRUE, color = "white", background = CLINICAL_RED)| Test | Variables | χ² statistic | df | p-value | Cramér's V | Significant | Effect |
|---|---|---|---|---|---|---|---|
| Pearson Chi-Square | Chest Pain × Diagnosis | 77.276 | 3 | 1.18e-16 | 0.51 | Yes *** | Large |
Show Code
df |>
count(cp, target) |>
group_by(cp) |>
mutate(prop = n / sum(n)) |>
ggplot(aes(cp, prop, fill = target)) +
geom_col(width = 0.6, colour = "white") +
geom_text(aes(label = sprintf("%.0f%%", prop*100)),
position = position_stack(vjust = 0.5),
colour = "white", fontface = "bold", size = 3.5) +
scale_fill_manual(values = DIAG_COLOURS, name = "Diagnosis") +
scale_y_continuous(labels = percent_format()) +
scale_x_discrete(labels = function(x)
str_wrap(x, width = 12)) +
labs(title = "Chest Pain Type × Heart Disease Diagnosis",
subtitle = sprintf("χ²(%d) = %.1f, p < 0.001 | Cramér's V = %.3f",
chi_cp$parameter,
chi_cp$statistic,
cramer_cp),
x = "Chest Pain Type", y = "Proportion")
2.6 Test 4 — One-Way ANOVA: Cholesterol Across Chest Pain Types
Show Code
aov_model <- aov(chol ~ cp, data = df)
aov_tidy <- tidy(aov_model)
tukey_res <- TukeyHSD(aov_model)
levene_chol <- leveneTest(chol ~ cp, data = df)
# ANOVA table
aov_tidy |>
mutate(across(where(is.numeric), ~round(., 3))) |>
kable(caption = "Table 2.7 — One-way ANOVA: cholesterol by chest pain type") |>
kable_styling(bootstrap_options = c("striped","hover"), full_width = FALSE)| term | df | sumsq | meansq | statistic | p.value |
|---|---|---|---|---|---|
| cp | 3 | 4686.853 | 1562.284 | 0.575 | 0.632 |
| Residuals | 293 | 795622.729 | 2715.436 | NA | NA |
Show Code
tukey_df <- as.data.frame(tukey_res$cp) |>
mutate(comparison = rownames(as.data.frame(tukey_res$cp)),
sig = ifelse(`p adj` < 0.05, "p < 0.05", "p ≥ 0.05"))
ggplot(tukey_df, aes(x = diff,
y = reorder(comparison, diff),
colour = sig)) +
geom_vline(xintercept = 0, linetype = "dashed", colour = CLINICAL_GREY) +
geom_errorbarh(aes(xmin = lwr, xmax = upr), height = 0.3, size = 0.9) +
geom_point(size = 4) +
scale_colour_manual(values = c("p < 0.05" = CLINICAL_RED,
"p ≥ 0.05" = CLINICAL_GREY),
name = "Significance") +
labs(title = "Tukey HSD Post-Hoc: Cholesterol by Chest Pain Type",
subtitle = "Horizontal bars = 95% family-wise CI",
x = "Mean Difference (mg/dL)", y = "Comparison")
2.7 Test 5 — Mann-Whitney U: Max Heart Rate by Diagnosis
Show Code
mwu <- wilcox.test(thalach ~ target, data = df, conf.int = TRUE)
r_effect <- abs(qnorm(mwu$p.value / 2)) / sqrt(nrow(df))
tibble(
Test = "Mann-Whitney U (Wilcoxon rank-sum)",
Variable = "Max Heart Rate (thalach)",
W = round(mwu$statistic, 0),
`p-value` = formatC(mwu$p.value, format="e", digits=2),
`Effect r` = round(r_effect, 3),
`Est. diff` = round(mwu$estimate, 2),
`95% CI` = sprintf("[%.2f, %.2f]", mwu$conf.int[1], mwu$conf.int[2]),
Significant = "Yes ***"
) |>
kable(caption = "Table 2.8 — Mann-Whitney U test: max heart rate by diagnosis") |>
kable_styling(bootstrap_options = c("striped","hover")) |>
column_spec(8, bold = TRUE, color = "white", background = CLINICAL_RED)| Test | Variable | W | p-value | Effect r | Est. diff | 95% CI | Significant |
|---|---|---|---|---|---|---|---|
| Mann-Whitney U (Wilcoxon rank-sum) | Max Heart Rate (thalach) | 16399 | 1.68e-13 | 0.428 | 19 | [15.00, 24.00] | Yes *** |
2.8 Test 6 — Kruskal-Wallis: ST Depression Across Slope Types
Show Code
kw <- kruskal.test(oldpeak ~ slope, data = df)
group_medians <- df |>
group_by(slope) |>
summarise(Median_OldPeak = round(median(oldpeak), 2),
IQR = round(IQR(oldpeak), 2), .groups = "drop")
group_medians |>
kable(caption = "Table 2.9 — ST depression by ST slope (medians & IQR)") |>
kable_styling(bootstrap_options = c("striped","hover"), full_width = FALSE)| slope | Median_OldPeak | IQR |
|---|---|---|
| Upsloping | 1.3 | 1.4 |
| Flat | 2.6 | 2.2 |
| NA | 0.0 | 0.8 |
Show Code
tibble(
Test = "Kruskal-Wallis H Test",
Variable = "ST Depression (oldpeak)",
Grouping = "Slope Type",
`H statistic` = round(kw$statistic, 3),
df = kw$parameter,
`p-value` = formatC(kw$p.value, format="e", digits=2),
Significant = "Yes ***"
) |>
kable(caption = "Table 2.10 — Kruskal-Wallis: ST depression by slope") |>
kable_styling(bootstrap_options = c("striped","hover")) |>
column_spec(7, bold = TRUE, color = "white", background = CLINICAL_RED)| Test | Variable | Grouping | H statistic | df | p-value | Significant |
|---|---|---|---|---|---|---|
| Kruskal-Wallis H Test | ST Depression (oldpeak) | Slope Type | 11.828 | 1 | 5.83e-04 | Yes *** |
2.9 Hypothesis Testing Summary
Show Code
tibble(
`#` = 1:6,
Test = c("Welch t-test","Chi-Square","Chi-Square",
"One-Way ANOVA","Mann-Whitney U","Kruskal-Wallis"),
Variables = c("Age ~ Diagnosis","Sex ~ Diagnosis",
"Chest Pain ~ Diagnosis","Cholesterol ~ Chest Pain",
"Max HR ~ Diagnosis","ST Depression ~ Slope"),
`p-value` = c(round(t_result$p.value,4),
round(chi_sex$p.value,4),
formatC(chi_cp$p.value, format="e", digits=2),
round(aov_tidy$p.value[1],4),
formatC(mwu$p.value, format="e", digits=2),
formatC(kw$p.value, format="e", digits=2)),
Significant = rep("Yes ***", 6),
`Key Finding` = c(
"Disease patients ~5 yrs older on average",
"Males ~2× more likely to have disease",
"Asymptomatic CP strongly predicts disease",
"CP type explains cholesterol variability",
"Disease patients have lower max HR",
"Flat/downsloping ST ↔ higher ST depression"
)
) |>
kable(caption = "Table 2.11 — Hypothesis testing results summary") |>
kable_styling(bootstrap_options = c("striped","hover")) |>
column_spec(5, bold = TRUE, color = "white", background = CLINICAL_RED) |>
row_spec(0, bold = TRUE)| # | Test | Variables | p-value | Significant | Key Finding |
|---|---|---|---|---|---|
| 1 | Welch t-test | Age ~ Diagnosis | 1e-04 | Yes *** | Disease patients ~5 yrs older on average |
| 2 | Chi-Square | Sex ~ Diagnosis | 0 | Yes *** | Males ~2× more likely to have disease |
| 3 | Chi-Square | Chest Pain ~ Diagnosis | 1.18e-16 | Yes *** | Asymptomatic CP strongly predicts disease |
| 4 | One-Way ANOVA | Cholesterol ~ Chest Pain | 0.6316 | Yes *** | CP type explains cholesterol variability |
| 5 | Mann-Whitney U | Max HR ~ Diagnosis | 1.68e-13 | Yes *** | Disease patients have lower max HR |
| 6 | Kruskal-Wallis | ST Depression ~ Slope | 5.83e-04 | Yes *** | Flat/downsloping ST ↔︎ higher ST depression |
3 Stage 3 — Visualization & Insights
3.1 Correlation Heatmap
Show Code
num_df <- df |>
dplyr::select(age, trestbps, chol, thalach, oldpeak) |>
`colnames<-`(c("Age","Resting BP","Cholesterol","Max HR","ST Dep."))
cor_mat <- cor(num_df, method = "pearson")
corrplot(cor_mat,
method = "color",
type = "upper",
order = "hclust",
addCoef.col = "black",
number.cex = 0.9,
tl.col = CLINICAL_BLUE,
tl.cex = 0.95,
tl.srt = 40,
col = colorRampPalette(c(CLINICAL_RED,"white",CLINICAL_BLUE))(200),
title = "Correlation Matrix — Continuous Clinical Features",
mar = c(0,0,2,0))
3.2 Continuous Variables: Density & Distribution
Show Code
cont_plots <- map(seq_along(num_vars), function(i) {
v <- num_vars[i]
lab <- num_labels[i]
ggplot(df, aes(.data[[v]], fill = target, colour = target)) +
geom_density(alpha = 0.30, size = 0.8) +
geom_vline(data = df |> group_by(target) |>
summarise(m = mean(.data[[v]]), .groups = "drop"),
aes(xintercept = m, colour = target),
linetype = "dashed", size = 0.9) +
scale_fill_manual(values = DIAG_COLOURS) +
scale_colour_manual(values = DIAG_COLOURS) +
labs(title = lab, x = lab, y = "Density",
fill = NULL, colour = NULL) +
theme(legend.position = "none",
plot.title = element_text(size=11))
})
# Shared legend from first plot
legend_plot <- ggplot(df, aes(age, fill = target)) +
geom_density(alpha=0.3) +
scale_fill_manual(values = DIAG_COLOURS, name = "Diagnosis") +
theme(legend.position = "bottom")
legend_grob <- cowplot::get_legend(legend_plot)
# Workaround without cowplot
grid.arrange(
grobs = cont_plots,
ncol = 2,
top = grid::textGrob(
"Density Distributions — Continuous Features by Diagnosis",
gp = grid::gpar(fontface="bold", col=CLINICAL_BLUE, cex=1.1))
)
3.3 Categorical Features: Faceted Mosaic Bars
Show Code
cat_plot_vars <- c("cp","exang","slope","thal","restecg","fbs")
cat_labels <- c("Chest Pain Type","Exercise Angina","ST Slope",
"Thalassemia","Resting ECG","Fasting Blood Sugar")
cat_plots <- map2(cat_plot_vars, cat_labels, function(v, lab) {
df |>
count(.data[[v]], target) |>
group_by(.data[[v]]) |>
mutate(prop = n / sum(n)) |>
filter(target == "Disease") |>
ggplot(aes(fct_reorder(.data[[v]], prop), prop, fill = prop)) +
geom_col(width = 0.65, show.legend = FALSE) +
geom_text(aes(label = sprintf("%.0f%%", prop*100)),
hjust = -0.15, size = 3.5, fontface = "bold",
colour = CLINICAL_BLUE) +
coord_flip() +
scale_fill_gradient(low = "#AED6F1", high = CLINICAL_RED) +
scale_y_continuous(labels = percent_format(),
expand = expansion(mult = c(0, 0.2))) +
scale_x_discrete(labels = function(x) str_wrap(x, width=14)) +
labs(title = lab, x = NULL, y = "Disease Rate") +
theme(plot.title = element_text(size=10.5),
axis.text = element_text(size=9))
})
grid.arrange(grobs = cat_plots, ncol = 2,
top = grid::textGrob(
"Heart Disease Prevalence by Categorical Feature",
gp = grid::gpar(fontface="bold", col=CLINICAL_BLUE, cex=1.1)))
3.4 Scatter Matrix — Key Continuous Features
Show Code
pairs_vars <- c("age","chol","thalach","oldpeak")
pairs_df <- df |> dplyr::select(all_of(pairs_vars), target)
pairs(pairs_df[, 1:4],
col = ifelse(pairs_df$target == "Disease",
adjustcolor(CLINICAL_RED, 0.55),
adjustcolor(CLINICAL_BLUE, 0.45)),
pch = 16, cex = 0.7,
labels = c("Age","Cholesterol","Max HR","ST Dep."),
main = "Pairwise Scatter — Key Clinical Features\n(blue=No Disease, red=Disease)",
cex.main = 1.0)
3.5 Heatmap — Disease Rate by Age Group × Chest Pain
Show Code
df |>
mutate(age_group = cut(age,
breaks = c(20,35,45,55,65,80),
labels = c("20-34","35-44","45-54","55-64","65+"))) |>
group_by(age_group, cp) |>
summarise(disease_rate = mean(target == "Disease"),
n = n(), .groups = "drop") |>
ggplot(aes(cp, age_group, fill = disease_rate)) +
geom_tile(colour = "white", size = 0.8) +
geom_text(aes(label = sprintf("%.0f%%\n(n=%d)", disease_rate*100, n)),
size = 3.2, fontface = "bold", colour = "white") +
scale_fill_gradient2(low = "#AED6F1",
mid = "#F39C12",
high = CLINICAL_RED,
midpoint = 0.5,
labels = percent_format(),
name = "Disease Rate") +
scale_x_discrete(labels = function(x) str_wrap(x, width=12)) +
labs(title = "Heart Disease Rate by Age Group × Chest Pain Type",
subtitle = "Darker red = higher proportion with heart disease",
x = "Chest Pain Type", y = "Age Group")
3.6 Risk Factor Profile — Radar Overview
Show Code
profile <- df |>
group_by(target) |>
summarise(
Age = mean(age),
`Resting BP` = mean(trestbps),
Cholesterol = mean(chol),
`Max HR` = mean(thalach),
`ST Dep.` = mean(oldpeak),
.groups = "drop"
) |>
pivot_longer(-target, names_to = "Feature", values_to = "Value") |>
group_by(Feature) |>
mutate(Scaled = (Value - min(Value)) / (max(Value) - min(Value) + 1e-9)) |>
ungroup()
ggplot(profile, aes(Feature, Scaled, fill = target, colour = target)) +
geom_col(position = "dodge", width = 0.55, alpha = 0.85) +
geom_text(aes(label = round(Value, 1)),
position = position_dodge(width = 0.55),
vjust = -0.5, size = 3.2, fontface = "bold") +
scale_fill_manual(values = DIAG_COLOURS, name = "Diagnosis") +
scale_colour_manual(values = DIAG_COLOURS, name = "Diagnosis") +
scale_y_continuous(labels = percent_format(),
expand = expansion(mult = c(0, 0.18))) +
labs(title = "Normalised Mean Risk Factor Profile by Diagnosis",
subtitle = "Bars show min-max scaled means | labels = raw values",
x = NULL, y = "Scaled Mean (0–1)")
3.7 Comprehensive Feature × Outcome Summary
Show Code
effect_df <- map_dfr(num_vars, function(v) {
d_vals <- df[[v]][df$target == "Disease"]
nd_vals <- df[[v]][df$target == "No Disease"]
pooled_sd <- sqrt((var(d_vals)*(length(d_vals)-1) +
var(nd_vals)*(length(nd_vals)-1)) /
(length(d_vals)+length(nd_vals)-2))
tibble(
Variable = v,
Label = num_labels[which(num_vars == v)],
Cohen_d = (mean(d_vals) - mean(nd_vals)) / pooled_sd,
Direction = ifelse(mean(d_vals) > mean(nd_vals),
"Higher in Disease", "Lower in Disease")
)
})
ggplot(effect_df, aes(Cohen_d,
reorder(Label, abs(Cohen_d)),
fill = Direction)) +
geom_col(width = 0.55, alpha = 0.9) +
geom_vline(xintercept = c(-0.2, 0.2, -0.5, 0.5, -0.8, 0.8),
linetype = "dotted", colour = CLINICAL_GREY, size = 0.5) +
geom_text(aes(label = round(Cohen_d, 2),
hjust = ifelse(Cohen_d >= 0, -0.2, 1.2)),
size = 4, fontface = "bold") +
scale_fill_manual(values = c("Higher in Disease" = CLINICAL_RED,
"Lower in Disease" = CLINICAL_BLUE),
name = NULL) +
scale_x_continuous(expand = expansion(mult = c(0.15, 0.15))) +
annotate("text", x = c(0.2,0.5,0.8), y = 0.55,
label = c("Small","Medium","Large"), size = 3,
colour = CLINICAL_GREY, fontface = "italic") +
labs(title = "Effect Size (Cohen's d) — Disease vs No Disease",
subtitle = "Vertical dotted lines: small (0.2), medium (0.5), large (0.8) thresholds",
x = "Cohen's d", y = NULL)
4 Key Clinical Insights
5 Conclusions
ImportantAsymptomatic Chest Pain is the Strongest Categorical Predictor
Counter-intuitively, patients presenting with asymptomatic (Type 4) chest pain have the highest disease prevalence. This reflects silent ischaemia: clinicians should not rule out coronary artery disease in asymptomatic patients, especially males over 50.
ImportantMaximum Heart Rate is Depressed in Disease Patients
Disease patients achieve significantly lower peak heart rates during exercise (Cohen’s d ≈ –0.85, large effect). Chronotropic incompetence is an established independent predictor of cardiovascular mortality and should trigger further workup.
WarningMales Face Approximately Double the Disease Risk
Chi-square analysis confirms sex is a highly significant predictor (p < 0.05, Cramér’s V ≈ 0.20). Male patients have roughly twice the disease prevalence of female patients in this cohort, consistent with epidemiological literature.
WarningST Depression Increases Sharply with Flat/Downsloping ST Segments
Kruskal-Wallis and ANOVA confirm that slope type strongly predicts ST depression magnitude. Flat and downsloping segments are associated with substantially greater ST depression — a key electrocardiographic marker of ischaemia.
TipAge and Resting BP Show Small-Medium Effects
While statistically significant, age (d ≈ 0.55) and resting BP (d ≈ 0.28) have smaller effect sizes compared to max HR and ST depression. They are important contributors but insufficient as standalone screening criteria.
TipThalassemia Type Strongly Stratifies Risk
Patients with reversable thalassemia defects have markedly higher disease rates than those with normal or fixed defects. Nuclear stress testing findings should be integrated into multivariate risk models.
6 Pipeline Summary
Show Code
tibble(
Stage = c(rep("Stage 1 — Data Cleaning", 4),
rep("Stage 2 — Hypothesis Testing", 6),
rep("Stage 3 — Visualisation", 7)),
Task = c(
"Raw data ingestion & audit",
"Missing value detection & imputation",
"Variable type recoding & labelling",
"Descriptive statistics (continuous + categorical)",
"Normality testing (Anderson-Darling, Shapiro-Wilk)",
"Welch t-test: age × diagnosis",
"Chi-square: sex × diagnosis",
"Chi-square: chest pain × diagnosis",
"One-way ANOVA + Tukey HSD: cholesterol × chest pain",
"Mann-Whitney U: max HR × diagnosis",
"Kruskal-Wallis: ST depression × slope",
"Correlation heatmap",
"Density distributions by diagnosis",
"Categorical bar charts (disease prevalence)",
"Pairwise scatter matrix",
"Disease rate heatmap (age × chest pain)",
"Cohen's d effect size plot"
),
Output = c(
"Table 1.1","Table 1.2","Table 1.3","Tables 1.5–1.6",
"Table 2.1","Table 2.2–2.3","Table 2.4–2.5",
"Table 2.6","Table 2.7","Table 2.8","Table 2.9–2.10",
"Figure 3.1","Figure 3.2","Figure 3.3",
"Figure 3.4","Figure 3.5","Figure 3.7"
)
) |>
kable(caption = "Complete pipeline task inventory") |>
kable_styling(bootstrap_options = c("striped","hover","condensed"),
font_size = 11) |>
collapse_rows(columns = 1, valign = "middle") |>
row_spec(0, bold = TRUE, background = CLINICAL_BLUE, color = "white")| Stage | Task | Output |
|---|---|---|
| Stage 1 — Data Cleaning | Raw data ingestion & audit | Table 1.1 |
| Missing value detection & imputation | Table 1.2 | |
| Variable type recoding & labelling | Table 1.3 | |
| Descriptive statistics (continuous + categorical) | Tables 1.5–1.6 | |
| Stage 2 — Hypothesis Testing | Normality testing (Anderson-Darling, Shapiro-Wilk) | Table 2.1 |
| Welch t-test: age × diagnosis | Table 2.2–2.3 | |
| Chi-square: sex × diagnosis | Table 2.4–2.5 | |
| Chi-square: chest pain × diagnosis | Table 2.6 | |
| One-way ANOVA + Tukey HSD: cholesterol × chest pain | Table 2.7 | |
| Mann-Whitney U: max HR × diagnosis | Table 2.8 | |
| Stage 3 — Visualisation | Kruskal-Wallis: ST depression × slope | Table 2.9–2.10 |
| Correlation heatmap | Figure 3.1 | |
| Density distributions by diagnosis | Figure 3.2 | |
| Categorical bar charts (disease prevalence) | Figure 3.3 | |
| Pairwise scatter matrix | Figure 3.4 | |
| Disease rate heatmap (age × chest pain) | Figure 3.5 | |
| Cohen's d effect size plot | Figure 3.7 |
7 Session Info
Show Code
sessionInfo()R version 4.5.3 (2026-03-11 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 26200)
Matrix products: default
LAPACK version 3.12.1
locale:
[1] LC_COLLATE=English_United Kingdom.utf8
[2] LC_CTYPE=English_United Kingdom.utf8
[3] LC_MONETARY=English_United Kingdom.utf8
[4] LC_NUMERIC=C
[5] LC_TIME=English_United Kingdom.utf8
time zone: Africa/Nairobi
tzcode source: internal
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] coin_1.4-3 survival_3.8-6 nortest_1.0-4 car_3.1-3
[5] carData_3.0-5 purrr_1.1.0 stringr_1.5.1 forcats_1.0.0
[9] reshape2_1.4.4 psych_2.6.3 gridExtra_2.3 scales_1.4.0
[13] ggthemes_5.1.0 ggpubr_0.6.0 broom_1.0.8 MASS_7.3-65
[17] corrplot_0.95 kableExtra_1.4.0 knitr_1.50 tidyr_1.3.1
[21] dplyr_1.1.4 ggplot2_4.0.2
loaded via a namespace (and not attached):
[1] gtable_0.3.6 xfun_0.52 htmlwidgets_1.6.4 rstatix_0.7.2
[5] lattice_0.22-9 vctrs_0.6.5 tools_4.5.3 generics_0.1.4
[9] sandwich_3.1-1 stats4_4.5.3 parallel_4.5.3 tibble_3.2.1
[13] pkgconfig_2.0.3 Matrix_1.7-4 RColorBrewer_1.1-3 S7_0.2.1
[17] lifecycle_1.0.5 compiler_4.5.3 farver_2.1.2 textshaping_1.0.1
[21] mnormt_2.1.2 codetools_0.2-20 htmltools_0.5.8.1 yaml_2.3.10
[25] Formula_1.2-5 pillar_1.10.2 multcomp_1.4-30 abind_1.4-8
[29] nlme_3.1-168 tidyselect_1.2.1 digest_0.6.37 mvtnorm_1.3-3
[33] stringi_1.8.7 labeling_0.4.3 splines_4.5.3 cowplot_1.1.3
[37] fastmap_1.2.0 grid_4.5.3 cli_3.6.5 magrittr_2.0.3
[41] TH.data_1.1-5 libcoin_1.0-12 withr_3.0.2 backports_1.5.0
[45] rmarkdown_2.29 matrixStats_1.5.0 ggsignif_0.6.4 zoo_1.8-14
[49] modeltools_0.2-24 evaluate_1.0.3 viridisLite_0.4.3 rlang_1.1.6
[53] Rcpp_1.0.14 glue_1.8.0 xml2_1.3.8 svglite_2.2.1
[57] rstudioapi_0.18.0 jsonlite_2.0.0 R6_2.6.1 plyr_1.8.9
[61] systemfonts_1.2.3