4  Visualizing Clinical Data

From Raw Numbers to Clear Communication

visualization
ggplot2
data-communication

Lecture slides for this module: Open Slides

4.1 Learning Objectives

By the end of this module, you will be able to:

  1. Identify principles of honest data visualization and recognise common graphical distortions
  2. Select appropriate plot types based on data type and research question
  3. Construct bar charts, histograms, box plots, violin plots, scatter plots, and line graphs using ggplot2
  4. Explain the grammar of graphics as a layered system
  5. Apply Tufte’s principles of data-ink ratio to clinical figures
  6. Design plots that prioritise clarity, accuracy, and accessibility (including colour-blind palettes)
  7. Critique visualizations from published clinical literature

4.2 Clinical Hook: The Pharma Pitch That Fooled Everyone

Same Data, Very Different Impressions

A pharmaceutical company presents a bar chart at a conference claiming their new antihypertensive drug reduces systolic blood pressure “dramatically more” than the competitor. The audience is impressed — the bars look strikingly different.

But look at the y-axis: it starts at 130 mmHg, not at 0. The actual difference? Their drug reduces SBP from 150 to 138 mmHg (12 mmHg reduction) while the competitor reduces it from 150 to 140 mmHg (10 mmHg reduction). A 2 mmHg difference, inflated to look enormous by truncating the axis.

This is not hypothetical — truncated axes are one of the most common distortion techniques in pharmaceutical advertisements, news graphics, and even peer-reviewed journals. In this module, you will learn to see through these tricks and to create honest, effective visualizations.

Code
bp_data <- tibble(
  Drug = c("New Drug", "Competitor"),
  SBP_reduction = c(138, 140)
)

p_trunc <- ggplot(bp_data, aes(x = Drug, y = SBP_reduction, fill = Drug)) +
  geom_col(width = 0.6, show.legend = FALSE) +
  scale_fill_manual(values = c("#e74c3c", "#3498db")) +
  coord_cartesian(ylim = c(130, 145)) +
  labs(
    title = "Truncated Axis (Misleading)",
    subtitle = "Y-axis starts at 130 mmHg",
    y = "Systolic BP after treatment (mmHg)", x = NULL
  ) +
  theme_minimal(base_size = 12) +
  theme(plot.title = element_text(face = "bold", colour = "#dc3545"))

p_honest <- ggplot(bp_data, aes(x = Drug, y = SBP_reduction, fill = Drug)) +
  geom_col(width = 0.6, show.legend = FALSE) +
  scale_fill_manual(values = c("#e74c3c", "#3498db")) +
  coord_cartesian(ylim = c(0, 150)) +
  labs(
    title = "Honest Axis (Accurate)",
    subtitle = "Y-axis starts at 0",
    y = "Systolic BP after treatment (mmHg)", x = NULL
  ) +
  theme_minimal(base_size = 12) +
  theme(plot.title = element_text(face = "bold", colour = "#28a745"))

p_trunc + p_honest +
  plot_annotation(
    title = "Truncated vs. Honest Bar Charts",
    subtitle = "The actual difference is only 2 mmHg — but the truncated axis makes it look enormous",
    theme = theme(
      plot.title = element_text(face = "bold", size = 14),
      plot.subtitle = element_text(colour = "grey40", size = 11)
    )
  )
Figure 4.1: The same data, two very different visual impressions. The truncated axis (left) exaggerates a small difference.

Truncated Axes on Bar Charts

Bar charts encode values as lengths. When the y-axis does not start at zero, the visual length of the bar no longer corresponds to the actual value, violating the most basic principle of bar charts. This is the single most common distortion in medical graphics. Rule: bar charts must start at zero. If the differences are too small to see, use a dot plot or a difference plot instead — do not truncate.

Note: line charts and scatter plots can use truncated axes because they encode position, not length.

4.3 The Grammar of Graphics: How ggplot2 Thinks

Before diving into specific chart types, let us understand the system behind ggplot2. Every plot is built from the same set of components, layered on top of each other like transparencies on an overhead projector.

The Grammar of Graphics (Wilkinson, 2005; Wickham, 2010)

Every ggplot2 figure is composed of seven layers:

  1. Data — the data frame you are plotting
  2. Aesthetics (aes()) — which variables map to x, y, colour, size, shape
  3. Geometries (geom_*) — what visual marks represent the data (points, bars, lines)
  4. Facets (facet_*) — split the plot into panels by a categorical variable
  5. Statistics (stat_*) — compute summaries (means, counts, smoothers)
  6. Coordinates (coord_*) — the coordinate system (Cartesian, polar, flipped)
  7. Theme (theme_*) — the non-data appearance (fonts, colours, grid lines)

The power of this system: you can swap any layer independently. Change geom_point() to geom_line() and the same data becomes a line chart. Add facet_wrap(~group) and you get small multiples. This modularity makes ggplot2 extraordinarily flexible.

Code
layers <- tibble(
  layer = factor(c("Theme", "Coordinates", "Statistics", "Facets",
                    "Geometries", "Aesthetics", "Data"),
                 levels = c("Theme", "Coordinates", "Statistics", "Facets",
                            "Geometries", "Aesthetics", "Data")),
  y = 7:1,
  description = c(
    "Non-data ink: fonts, background, grid",
    "Coordinate system: Cartesian, polar, flip",
    "Computed summaries: means, smoothers",
    "Split into panels by category",
    "Visual marks: points, bars, lines, areas",
    "Map variables to x, y, colour, size",
    "The data frame you are plotting"
  ),
  colour = c("#6c757d", "#17a2b8", "#ffc107", "#28a745",
             "#dc3545", "#6f42c1", "#0077b6")
)

ggplot(layers, aes(x = 1, y = y)) +
  geom_tile(aes(fill = colour), width = 3.5, height = 0.85,
            colour = "white", linewidth = 1.5) +
  geom_text(aes(x = 1, label = layer),
            fontface = "bold", size = 5, colour = "white") +
  geom_text(aes(x = 1, y = y - 0.35, label = description),
            size = 3.2, colour = "grey30") +
  scale_fill_identity() +
  coord_cartesian(xlim = c(-0.5, 2.5), ylim = c(0.2, 7.8)) +
  labs(title = "The Seven Layers of ggplot2") +
  theme_void(base_size = 13) +
  theme(plot.title = element_text(face = "bold", hjust = 0.5, size = 15))
Figure 4.2: The seven layers of the Grammar of Graphics, shown as a stacked diagram.

4.4 Choosing the Right Chart: A Decision Framework

The most important decision in visualization is not how to draw the chart but which chart to draw. The wrong chart type will mislead no matter how pretty it looks.

Code
nodes <- tibble(
  x = c(5,     # Start: What's your goal?
        2, 5, 8,   # Level 2: Comparison, Distribution, Relationship
        1, 3,      # Comparison subtypes
        4, 6,      # Distribution subtypes
        7, 9),     # Relationship subtypes
  y = c(6,
        4, 4, 4,
        2, 2,
        2, 2,
        2, 2),
  label = c("What is your\nresearch question?",
            "Comparing\ngroups", "Showing\ndistribution", "Showing\nrelationship",
            "Bar chart\nDot plot", "Box plot\nViolin plot",
            "Histogram\nDensity plot", "Line chart\n(over time)",
            "Scatter plot", "Heat map\n(many variables)"),
  type = c("question",
           "goal", "goal", "goal",
           "chart", "chart", "chart", "chart", "chart", "chart"),
  fill = c("#495057",
           "#0077b6", "#28a745", "#e74c3c",
           "#0077b6", "#0077b6", "#28a745", "#28a745", "#e74c3c", "#e74c3c")
)

edges <- tibble(
  x = c(5, 5, 5, 2, 2, 5, 5, 8, 8),
  y = c(6, 6, 6, 4, 4, 4, 4, 4, 4),
  xend = c(2, 5, 8, 1, 3, 4, 6, 7, 9),
  yend = c(4, 4, 4, 2, 2, 2, 2, 2, 2)
)

ggplot() +
  geom_segment(data = edges,
               aes(x = x, y = y, xend = xend, yend = yend),
               colour = "grey50", linewidth = 0.8,
               arrow = arrow(length = unit(0.15, "cm"), type = "closed")) +
  geom_label(data = nodes,
             aes(x = x, y = y, label = label, fill = fill),
             colour = "white", fontface = "bold", size = 3.5,
             label.padding = unit(0.5, "lines"),
             label.r = unit(0.3, "lines")) +
  scale_fill_identity() +
  coord_cartesian(xlim = c(0, 10), ylim = c(1, 7)) +
  labs(title = "Which Chart Should I Use?",
       subtitle = "Start with the question, not the chart type") +
  theme_void(base_size = 13) +
  theme(
    plot.title = element_text(face = "bold", hjust = 0.5, size = 15),
    plot.subtitle = element_text(hjust = 0.5, colour = "grey40")
  )
Figure 4.3: A decision framework for choosing the right chart type based on your data and research question.

Quick Reference: Data Type to Chart Type

Your data Your question Best chart
1 categorical variable How many in each group? Bar chart
1 continuous variable What does the distribution look like? Histogram, density
1 categorical + 1 continuous Do groups differ? Box plot, violin plot
2 continuous variables Is there a relationship? Scatter plot
Continuous over time How does it change? Line chart
Multiple groups over time Trends by group? Line chart with colour / facets
3+ variables Complex patterns? Heat map, faceted plots

For interactive guidance, explore From Data to Viz — a beautiful flowchart that helps you navigate from your data structure to the ideal chart.

4.5 Chart Types with Clinical Examples

Bar Charts: Counting and Comparing Categories

Bar charts are the workhorses of categorical data — simple, universally understood, and nearly impossible to misread (as long as the axis starts at zero).

Code
blood_groups <- tibble(
  group = factor(c("O+", "A+", "B+", "AB+", "O-", "A-", "B-", "AB-"),
                 levels = c("O+", "A+", "B+", "AB+", "O-", "A-", "B-", "AB-")),
  count = c(175, 110, 120, 30, 30, 18, 12, 5)
)

ggplot(blood_groups, aes(x = group, y = count, fill = group)) +
  geom_col(show.legend = FALSE, width = 0.7) +
  geom_text(aes(label = count), vjust = -0.5, fontface = "bold", size = 3.5) +
  scale_fill_manual(values = c(
    rep("#c0392b", 4), rep("#2c3e50", 4)
  )) +
  labs(
    title = "Blood Group Distribution — Hospital Blood Bank",
    subtitle = "Rh-positive groups dominate; AB- is the rarest group",
    x = "Blood Group", y = "Number of Donors"
  ) +
  theme_minimal(base_size = 12) +
  theme(plot.title = element_text(face = "bold"))
Figure 4.4: Distribution of blood groups in a hospital blood bank (n = 500). Note the y-axis starts at zero.

Dynamite Plots (Bar Chart + Error Bar for Continuous Data)

A common practice in medical journals is to show a bar chart with an error bar on top (mean + SD or SEM). This “dynamite plot” hides the actual distribution — 50 data points or 5000, bimodal or uniform, the bar chart looks the same. Never use bar charts for continuous data. Use box plots, violin plots, or dot plots instead.

Histograms and Density Plots: Seeing the Distribution

Histograms divide continuous data into bins and count observations in each. Density plots smooth the histogram into a continuous curve.

Code
hb_data <- tibble(
  hb = c(rnorm(400, mean = 12.5, sd = 1.2),
         rnorm(100, mean = 9.0, sd = 1.0))
)

p_hist <- ggplot(hb_data, aes(x = hb)) +
  geom_histogram(binwidth = 0.5, fill = "#3498db", colour = "white", alpha = 0.8) +
  geom_vline(xintercept = 12, linetype = "dashed", colour = "#e74c3c", linewidth = 0.8) +
  annotate("text", x = 12.1, y = Inf, vjust = 2, hjust = 0,
           label = "WHO cutoff\n(12 g/dL)", colour = "#e74c3c", size = 3.5) +
  labs(title = "Histogram", x = "Hemoglobin (g/dL)", y = "Count") +
  theme_minimal(base_size = 12) +
  theme(plot.title = element_text(face = "bold"))

p_dens <- ggplot(hb_data, aes(x = hb)) +
  geom_density(fill = "#3498db", alpha = 0.4, colour = "#2c3e50") +
  geom_vline(xintercept = 12, linetype = "dashed", colour = "#e74c3c", linewidth = 0.8) +
  annotate("text", x = 12.1, y = Inf, vjust = 2, hjust = 0,
           label = "WHO cutoff\n(12 g/dL)", colour = "#e74c3c", size = 3.5) +
  labs(title = "Density Plot", x = "Hemoglobin (g/dL)", y = "Density") +
  theme_minimal(base_size = 12) +
  theme(plot.title = element_text(face = "bold"))

p_hist + p_dens +
  plot_annotation(
    title = "Hemoglobin Distribution — Adult Women (n = 500)",
    subtitle = "Notice the secondary peak around 9 g/dL — a subgroup with iron deficiency anaemia",
    theme = theme(
      plot.title = element_text(face = "bold", size = 14),
      plot.subtitle = element_text(colour = "grey40", size = 11)
    )
  )
Figure 4.5: Hemoglobin distribution in 500 adult women — histogram and density plot show the same data in different ways.

Bin Width Matters

The choice of bin width in a histogram dramatically affects what you see. Too wide: you lose detail and miss subgroups. Too narrow: you see random noise. A good rule of thumb: start with the Freedman-Diaconis rule (the ggplot2 default) and then adjust manually if needed. Always try at least 2-3 bin widths before settling on one.

Box Plots and Violin Plots: Comparing Distributions Across Groups

When you want to compare a continuous variable across groups, box plots and violin plots are your best options. We covered these in Module 2 with CRP data — here we see additional clinical examples and discuss when to choose one over the other.

Code
treatment_data <- tibble(
  group = rep(c("Standard Care", "Drug A", "Drug B"), each = 100),
  los = c(
    rgamma(100, shape = 3, rate = 0.5),     # Standard: right skewed
    rnorm(100, mean = 5, sd = 1.2),          # Drug A: roughly symmetric
    c(rnorm(60, mean = 4, sd = 0.8),         # Drug B: bimodal
      rnorm(40, mean = 8, sd = 0.8))
  )
) %>%
  mutate(group = factor(group, levels = c("Standard Care", "Drug A", "Drug B")))

p_points <- ggplot(treatment_data, aes(x = group, y = los, colour = group)) +
  geom_jitter(width = 0.2, alpha = 0.4, size = 1.5) +
  scale_colour_manual(values = c("#0077b6", "#28a745", "#e74c3c")) +
  labs(title = "Raw Data", y = "Length of Stay (days)", x = NULL) +
  theme_minimal(base_size = 11) +
  theme(plot.title = element_text(face = "bold"), legend.position = "none")

p_box <- ggplot(treatment_data, aes(x = group, y = los, fill = group)) +
  geom_boxplot(alpha = 0.7, outlier.shape = 21) +
  scale_fill_manual(values = c("#0077b6", "#28a745", "#e74c3c")) +
  labs(title = "Box Plot", y = "Length of Stay (days)", x = NULL) +
  theme_minimal(base_size = 11) +
  theme(plot.title = element_text(face = "bold"), legend.position = "none")

p_violin <- ggplot(treatment_data, aes(x = group, y = los, fill = group)) +
  geom_violin(alpha = 0.7, colour = "grey30") +
  geom_boxplot(width = 0.15, fill = "white", alpha = 0.8, outlier.shape = NA) +
  scale_fill_manual(values = c("#0077b6", "#28a745", "#e74c3c")) +
  labs(title = "Violin + Box Plot", y = "Length of Stay (days)", x = NULL) +
  theme_minimal(base_size = 11) +
  theme(plot.title = element_text(face = "bold"), legend.position = "none")

p_points + p_box + p_violin +
  plot_annotation(
    title = "Hospital Length of Stay by Treatment Group (n = 100 per group)",
    subtitle = "Drug B has a bimodal distribution — only the violin plot reveals this",
    theme = theme(
      plot.title = element_text(face = "bold", size = 14),
      plot.subtitle = element_text(colour = "grey40", size = 11)
    )
  )
Figure 4.6: Three representations of the same data: raw points, box plot, and violin plot. Each reveals different aspects of the distribution.

Box Plot vs. Violin Plot: When to Use Which

Feature Box plot Violin plot
Shows median and quartiles Yes (explicitly) Only if overlaid
Shows full distribution shape No Yes
Reveals bimodality No Yes
Easy to compare many groups Yes Gets crowded beyond 5-6 groups
Familiar to clinical audiences Yes Less familiar
Best used when… You need a quick, standard summary Distribution shape matters for interpretation

Best practice: overlay a narrow box plot inside the violin to get the best of both worlds, as shown in the right panel above.

Scatter Plots: Revealing Relationships

Scatter plots show the relationship between two continuous variables — the foundation for correlation and regression analysis.

Code
gfr_data <- tibble(
  age = runif(200, 25, 85),
  egfr = 120 - 0.8 * age + rnorm(200, 0, 12),
  ckd_stage = case_when(
    egfr >= 90 ~ "G1 (Normal)",
    egfr >= 60 ~ "G2 (Mild)",
    egfr >= 30 ~ "G3 (Moderate)",
    TRUE ~ "G4-5 (Severe)"
  )
) %>%
  mutate(ckd_stage = factor(ckd_stage,
                            levels = c("G1 (Normal)", "G2 (Mild)",
                                       "G3 (Moderate)", "G4-5 (Severe)")))

ggplot(gfr_data, aes(x = age, y = egfr)) +
  geom_point(aes(colour = ckd_stage), alpha = 0.6, size = 2) +
  geom_smooth(method = "lm", colour = "#2c3e50", linewidth = 1, se = TRUE) +
  geom_hline(yintercept = c(90, 60, 30), linetype = "dashed",
             colour = c("#ffc107", "#e67e22", "#dc3545"), linewidth = 0.6) +
  annotate("text", x = 26, y = 93, label = "G1/G2 cutoff", hjust = 0,
           colour = "#ffc107", size = 3, fontface = "bold") +
  annotate("text", x = 26, y = 63, label = "G2/G3 cutoff", hjust = 0,
           colour = "#e67e22", size = 3, fontface = "bold") +
  annotate("text", x = 26, y = 33, label = "G3/G4 cutoff", hjust = 0,
           colour = "#dc3545", size = 3, fontface = "bold") +
  scale_colour_manual(values = c("#28a745", "#ffc107", "#e67e22", "#dc3545")) +
  labs(
    title = "eGFR Declines with Age — But with Substantial Individual Variability",
    subtitle = "Blue band = 95% confidence interval for the regression line",
    x = "Age (years)", y = "eGFR (mL/min/1.73m2)",
    colour = "CKD Stage"
  ) +
  theme_minimal(base_size = 12) +
  theme(
    plot.title = element_text(face = "bold"),
    legend.position = "bottom"
  )
Figure 4.7: Age vs. estimated GFR (eGFR) in 200 patients. The decline with age is clearly visible, with considerable person-to-person variability.

Scatter Plot Enhancements for Large Datasets

When you have thousands of points, a standard scatter plot becomes an unreadable blob (overplotting). Solutions include reducing alpha (transparency), using hexagonal binning (geom_hex()), 2D density contours (geom_density_2d()), or marginal distributions with the ggExtra package. Always consider which technique reveals the most useful pattern in your particular dataset.

4.6 Principles of Honest Visualization

Edward Tufte, the pioneer of data visualization, established principles that remain the gold standard for scientific graphics [1].

Data-Ink Ratio

Tufte’s Data-Ink Ratio

\[\text{Data-ink ratio} = \frac{\text{Ink used to display data}}{\text{Total ink used in the graphic}}\]

The goal is to maximise this ratio — every drop of ink should represent data. Remove grid lines, background colours, borders, and decorative elements that do not convey information. A figure should be as simple as possible, but not simpler.

Code
demo_data <- tibble(
  category = c("A", "B", "C", "D"),
  value = c(23, 45, 31, 52)
)

p_cluttered <- ggplot(demo_data, aes(x = category, y = value, fill = category)) +
  geom_col(show.legend = FALSE, colour = "black", linewidth = 1) +
  scale_fill_manual(values = c("#ff6b6b", "#4ecdc4", "#45b7d1", "#96ceb4")) +
  labs(title = "Low Data-Ink Ratio", y = "Value", x = "Category") +
  theme(
    panel.background = element_rect(fill = "#f0f0f0"),
    panel.grid.major = element_line(colour = "white", linewidth = 1.2),
    panel.grid.minor = element_line(colour = "white", linewidth = 0.6),
    panel.border = element_rect(colour = "black", fill = NA, linewidth = 2),
    plot.title = element_text(face = "bold"),
    axis.text = element_text(size = 10)
  )

p_clean <- ggplot(demo_data, aes(x = category, y = value)) +
  geom_col(fill = "#2c3e50", width = 0.6) +
  geom_text(aes(label = value), vjust = -0.5, fontface = "bold", size = 4) +
  labs(title = "High Data-Ink Ratio", y = "Value", x = NULL) +
  theme_minimal(base_size = 12) +
  theme(
    panel.grid.major.x = element_blank(),
    panel.grid.minor = element_blank(),
    plot.title = element_text(face = "bold")
  )

p_cluttered + p_clean +
  plot_annotation(
    title = "Data-Ink Ratio: Less Decoration, More Information",
    theme = theme(plot.title = element_text(face = "bold", size = 14))
  )
Figure 4.9: The same data with low data-ink ratio (left, cluttered with non-data elements) and high data-ink ratio (right, clean and focused).

Common Distortion Techniques to Watch For

Table 4.1: Common graphical distortions and how to spot them
Distortion How It Misleads How to Fix It
Truncated axis Exaggerates small differences by not starting the axis at zero Start bar chart axes at zero; use dot plots for small differences
Dual y-axes Creates false apparent correlations between unrelated variables Use two separate panels or normalise to a common scale
3D effects Perspective distorts bar heights and slice sizes Use 2D plots — always
Area/volume encoding Doubling a circle's radius quadruples its area, exaggerating differences Map values to position (dot plot) or length (bar chart), not area
Cherry-picked time range Selecting a specific time window to show a trend that disappears in the full data Show the full time range; use annotations to highlight the period of interest
Unlabelled axes Without units or scale, any difference can look large or small Always label both axes with variable name and units

4.7 Accessibility: Designing for Everyone

Approximately 8% of men and 0.5% of women have some form of colour vision deficiency. A red-green colour scheme — the most common in medical graphics — is the worst possible choice for accessibility.

Code
set.seed(99)
cb_data <- tibble(
  x = rep(1:5, each = 4),
  group = rep(c("Group A", "Group B", "Group C", "Group D"), 5),
  y = rnorm(20, mean = rep(c(10, 15, 12, 18), 5), sd = 2)
)

p_rg <- ggplot(cb_data, aes(x = factor(x), y = y, fill = group)) +
  geom_col(position = "dodge") +
  scale_fill_manual(values = c("#d62728", "#2ca02c", "#ff7f0e", "#1f77b4")) +
  labs(title = "Red-Green Palette", subtitle = "Problematic for ~8% of men",
       x = "Time Point", y = "Value") +
  theme_minimal(base_size = 11) +
  theme(plot.title = element_text(face = "bold"), legend.position = "bottom")

p_viridis <- ggplot(cb_data, aes(x = factor(x), y = y, fill = group)) +
  geom_col(position = "dodge") +
  scale_fill_viridis_d(option = "D") +
  labs(title = "Viridis Palette", subtitle = "Safe for all viewers",
       x = "Time Point", y = "Value") +
  theme_minimal(base_size = 11) +
  theme(plot.title = element_text(face = "bold"), legend.position = "bottom")

p_rg + p_viridis +
  plot_annotation(
    title = "Colour-Blind Friendly Palettes",
    theme = theme(plot.title = element_text(face = "bold", size = 14))
  )
Figure 4.10: The viridis palette (right) remains distinguishable under all forms of colour vision deficiency. The red-green palette (left) does not.

Accessibility Checklist for Clinical Figures

  1. Use colour-blind friendly palettes: viridis, cividis, or okabe_ito (available in scales::pal_okabe_ito())
  2. Don’t rely on colour alone: use different shapes, line types, or direct labels in addition to colour
  3. Ensure sufficient contrast: light text on dark backgrounds (or vice versa) with a contrast ratio of at least 4.5:1
  4. Use readable font sizes: axis labels at least 10pt, titles at least 12pt in the final figure
  5. Add alt-text: in Quarto, use #| fig-alt: to describe the figure for screen readers
  6. Test your palette: the colorblindr R package can simulate how your figure appears under different forms of colour vision deficiency

4.8 Putting It All Together: A Publication-Ready Figure

Let us build a complete clinical figure step by step, applying everything we have learned — grammar of graphics, honest axes, clean design, and accessible colours.

Code
set.seed(42)
bmi_data <- tibble(
  diabetes = rep(c("No Diabetes", "Pre-Diabetes", "Type 2 Diabetes"), c(300, 150, 100)),
  bmi = c(
    rnorm(300, 24, 3.5),
    rnorm(150, 28, 3.0),
    rnorm(100, 32, 4.0)
  )
) %>%
  mutate(diabetes = factor(diabetes,
                           levels = c("No Diabetes", "Pre-Diabetes", "Type 2 Diabetes")))

ggplot(bmi_data, aes(x = diabetes, y = bmi, fill = diabetes)) +
  geom_violin(alpha = 0.7, colour = "grey30", trim = FALSE) +
  geom_boxplot(width = 0.15, fill = "white", alpha = 0.8, outlier.shape = NA) +
  stat_summary(fun = mean, geom = "point", shape = 18, size = 3, colour = "#2c3e50") +
  geom_hline(yintercept = c(25, 30), linetype = "dashed",
             colour = c("#e67e22", "#dc3545"), linewidth = 0.5) +
  annotate("text", x = 3.45, y = 25.3,
           label = "Overweight (25)", colour = "#e67e22",
           size = 3, fontface = "italic", hjust = 1) +
  annotate("text", x = 3.45, y = 30.3,
           label = "Obese (30)", colour = "#dc3545",
           size = 3, fontface = "italic", hjust = 1) +
  scale_fill_viridis_d(option = "D", begin = 0.3, end = 0.9) +
  labs(
    title = "BMI Distribution by Diabetes Status",
    subtitle = "Diamond = mean | Box = median and IQR | Violin = full distribution",
    x = NULL, y = expression("Body Mass Index (kg/m"^2*")"),
    caption = "Simulated data for illustration | n = 550"
  ) +
  theme_minimal(base_size = 12) +
  theme(
    plot.title = element_text(face = "bold", size = 14),
    plot.subtitle = element_text(colour = "grey40", size = 10),
    plot.caption = element_text(colour = "grey50", face = "italic"),
    legend.position = "none",
    panel.grid.major.x = element_blank()
  )
Figure 4.11: A publication-ready figure showing BMI distribution across diabetes status, with appropriate annotations and clean design.

4.9 Visualization Tools and Resources

Books:

  • Edward Tufte, The Visual Display of Quantitative Information (2nd ed.) — the bible of data visualization
  • Kieran Healy, Data Visualization: A Practical Introduction — modern, R-focused, free online
  • Claus Wilke, Fundamentals of Data Visualization — open-access textbook with excellent examples
  • Hadley Wickham, ggplot2: Elegant Graphics for Data Analysis (3rd ed.) — the definitive ggplot2 reference

Videos:

R Packages:

  • ggplot2 — core plotting (you already know this one)
  • patchwork — combine multiple plots into one figure
  • scales — format axes (percent, comma, scientific notation)
  • ggrepel — smart text labels that avoid overlapping
  • viridis — colour-blind friendly palettes
  • ggExtra — marginal histograms/density on scatter plots
  • plotly — convert ggplot2 figures to interactive HTML

Graph Galleries and Chart Selection Guides:

Colour and Accessibility:

4.10 NEET PG Practice MCQs

Q1. A pharmaceutical company shows a bar chart comparing mean systolic BP reduction: Drug X = 14 mmHg, Drug Y = 12 mmHg. The y-axis starts at 10 mmHg. What is the primary concern with this figure?

Q2. A researcher presents box plots comparing serum uric acid levels between gout patients and healthy controls. The gout group’s box is very short with many outliers above the upper whisker. What does this pattern indicate?

Q3. You want to show the relationship between age and estimated GFR (eGFR) in 5000 patients. A standard scatter plot shows a dense blob of overlapping points. Which technique would best address this?

Q4. According to Tufte’s data-ink ratio principle, which of the following should generally be REMOVED from a clinical figure?

Q5. A colleague creates a line chart showing average patient satisfaction scores across 5 different hospital wards (Surgery, Medicine, Paediatrics, Orthopaedics, Obstetrics). What is wrong with this visualization?

Q6. Approximately 8% of men have red-green colour vision deficiency. Which of the following strategies BEST ensures your clinical figure is accessible to these viewers?

Q7. A study reports HbA1c values at 0, 3, 6, and 12 months for two treatment arms. The authors show a line chart with ribbons representing 95% confidence intervals. At month 6, the ribbons of the two arms overlap slightly but the p-value is reported as 0.02. Which statement is correct?

4.11 Summary

This module covered the principles and practice of clinical data visualization:

  1. Choose the chart for the question, not the other way around — bar charts for counts, histograms for distributions, box/violin plots for group comparisons, scatter plots for relationships, and line charts for trends over time.

  2. The Grammar of Graphics gives ggplot2 its power: data, aesthetics, geometries, facets, statistics, coordinates, and themes are independent, swappable layers.

  3. Honest visualization means zero-baseline bar charts, no 3D effects, no truncated axes, and no dual y-axes. Every drop of ink should represent data (Tufte’s data-ink ratio).

  4. Accessibility is not optional — use colour-blind safe palettes (viridis), encode information redundantly (colour + shape), and ensure adequate contrast and font sizes.

  5. Never trust summary statistics alone — always visualise your data. As Anscombe’s quartet showed us in Module 2, identical numbers can hide completely different patterns.

In the next module, we move from describing data to reasoning about uncertainty — the foundations of probability that underpin all of clinical inference.

4.12 References