15  ggplot2 in R with CDISC ADaM Examples

16 Introduction

This tutorial is a step-by-step guide to ggplot2 in R, designed to take you from beginner to advanced, using CDISC ADaM-style datasets throughout.

• ADSL: Subject-level data
• ADLB: Laboratory data
• ADVS: Vital signs (optional example)
• ADTTE: Time-to-event data (e.g. PFS)

These are simulated examples with ADaM-like structures and variable names. In a real project you would replace them with your actual ADSL/ADLB/ADTTE datasets.

We will cover:

• Basic Grammar of Graphics concepts
• Building plots layer by layer with ggplot()
• Common geoms: points, lines, bars, histograms, boxplots
• Aesthetics: color, shape, size, linetype
• Faceting, scales, themes, coordinates
• Combining dplyr + ggplot2 in a tidy ADaM workflow
• Advanced: annotations, secondary axes, simple KM plot from ADTTE
• Saving plots for reports/presentations

---

17 Creating Simulated ADaM-Style Data

In this section we create small ADaM-like datasets entirely in R so the tutorial is self-contained.

17.1 ADSL: Subject-Level Data

ADSL <- haven::read_sas("./data/adam/adsl.sas7bdat")

ADSL <- ADSL |> 
       dplyr::mutate(TRTA=ARM)

17.2 ADLB: Laboratory Data (e.g. ALT)

We simulate longitudinal ALT values (PARAMCD = "ALT") at multiple visits:

ADLB <- haven::read_sas("./data/adam/adlbhy.sas7bdat") |> 
              dplyr::filter(PARAMCD == "ALT")

17.3 ADTTE: Time-to-Event Data (e.g. PFS)

We simulate a simple ADTTE:

• PARAMCD = "PFS"
• AVAL: time (months)
• CNSR: censor flag (0 = event, 1 = censored)

ADTTE <- ADSL %>%
  select(STUDYID, USUBJID, ARM,TRTA, SEX, AGE) %>%
  mutate(
    PARAMCD = "PFS",
    PARAM   = "Progression-Free Survival (Months)",
    # Arm-specific exponential rates
    rate = case_when(
      ARM == "Placebo" ~ 0.12,
      ARM == "Dose 1"  ~ 0.09,
      TRUE             ~ 0.07
    ),
    AVAL = rexp(n(), rate = rate),
    AVAL = pmin(AVAL, 36),  # administrative censoring at 36 months
    CNSR = if_else(AVAL >= 36, 1, 0)
  ) %>%
  select(-rate)

ADTTE %>% head()

# A tibble: 6 × 10
  STUDYID      USUBJID     ARM      TRTA  SEX     AGE PARAMCD PARAM   AVAL  CNSR
  <chr>        <chr>       <chr>    <chr> <chr> <dbl> <chr>   <chr>  <dbl> <dbl>
1 CDISCPILOT01 01-701-1015 Placebo  Plac… F        63 PFS     Prog…  7.03      0
2 CDISCPILOT01 01-701-1023 Placebo  Plac… M        64 PFS     Prog…  4.81      0
3 CDISCPILOT01 01-701-1028 Xanomel… Xano… M        71 PFS     Prog… 19.0       0
4 CDISCPILOT01 01-701-1033 Xanomel… Xano… M        74 PFS     Prog…  0.451     0
5 CDISCPILOT01 01-701-1034 Xanomel… Xano… F        77 PFS     Prog…  0.803     0
6 CDISCPILOT01 01-701-1047 Placebo  Plac… F        85 PFS     Prog…  2.64      0

With these datasets we can now explore ggplot2 concepts in an ADaM context.

---

18 Grammar of Graphics with ADaM

The basic idea of ggplot2 (Grammar of Graphics):

• Data: ADSL / ADLB / ADTTE
• Aesthetics (aes): how variables map to x, y, color, etc.
• Geoms: geometric objects (points, lines, bars, etc.)
• Scales, facets, coordinates, theme: control look and structure

The general pattern:

ADSL <- haven::read_sas("./data/adam/adsl.sas7bdat")
ggplot(data = ADSL, aes(x = AGE, y = BMIBL)) +
  geom_point()

We will start with subject-level plots from ADSL, then move to longitudinal labs from ADLB, and finally to ADTTE.

---

19 Basic Plots from ADSL (Beginner Level)

19.1 Scatterplot: AGE vs BMIBL

ggplot(ADSL, aes(x = AGE, y = BMIBL)) +
  geom_point() +
  labs(
    title = "Scatterplot of Baseline BMI vs Age",
    x = "Age (Years)",
    y = "Baseline BMI (kg/m^2)"
  )

19.1.1 Mapping vs Setting Aesthetics

• Mapping (inside aes()): color/shape/size depend on data (e.g. ARM).
• Setting (outside aes()): fixed values.

Map ARM to color:

ggplot(ADSL, aes(x = AGE, y = BMIBL, color = TRT01A)) +
  geom_point(size = 2.5, alpha = 0.8) +
  labs(
    title = "Baseline BMI vs Age by Treatment Arm",
    color = "Treatment Arm",
    x = "Age (Years)",
    y = "Baseline BMI (kg/m^2)"
  )

Set color manually (not based on data):

ggplot(ADSL, aes(x = AGE, y = BMIBL)) +
  geom_point(color = "steelblue", size = 2.5, alpha = 0.8) +
  labs(
    title = "Baseline BMI vs Age (All Arms)",
    x = "Age (Years)",
    y = "Baseline BMI (kg/m^2)"
  )

---

19.2 Bar Plot: Subject Counts per Arm

Using ADSL, we can show how many subjects per arm:

ggplot(ADSL, aes(x = TRT01A)) +
  geom_bar(fill = "skyblue", color = "black") +
  labs(
    title = "Number of Subjects per Treatment Arm",
    x = "Treatment Arm",
    y = "Count of Subjects"
  )

If you pre-compute counts (e.g. for TLG), use geom_col():

adsl_counts <- ADSL %>%
  count(TRT01A, name = "n")

adsl_counts

# A tibble: 3 × 2
  TRT01A                   n
  <chr>                <int>
1 Placebo                 86
2 Xanomeline High Dose    84
3 Xanomeline Low Dose     84

ggplot(adsl_counts, aes(x = TRT01A, y = n)) +
  geom_col(fill = "steelblue") +
  labs(
    title = "Number of Subjects per Treatment Arm",
    x = "Treatment Arm",
    y = "Count of Subjects"
  )

---

19.3 Histogram and Density: Age Distribution

Histogram of AGE:

ggplot(ADSL, aes(x = AGE)) +
  geom_histogram(binwidth = 5, fill = "lightgreen", color = "white") +
  labs(
    title = "Age Distribution (ADSL)",
    x = "Age (Years)",
    y = "Frequency"
  )

Density plot:

ggplot(ADSL, aes(x = AGE)) +
  geom_density(fill = "orange", alpha = 0.6) +
  labs(
    title = "Age Density (ADSL)",
    x = "Age (Years)",
    y = "Density"
  )

---

19.4 Boxplot: BMIBL by ARM

Boxplots are useful in CSR or listings to compare distributions across arms.

ggplot(ADSL, aes(x = TRT01A, y = BMIBL)) +
  geom_boxplot(fill = "lightgray") +
  labs(
    title = "Baseline BMI by Treatment Arm",
    x = "Treatment Arm",
    y = "Baseline BMI (kg/m^2)"
  )

You can combine boxplot with jittered points:

ggplot(ADSL, aes(x = TRT01A, y = BMIBL)) +
  geom_boxplot(fill = "lightgray", outlier.shape = NA) +
  geom_jitter(width = 0.15, alpha = 0.6, color = "blue") +
  labs(
    title = "Baseline BMI by Treatment Arm (with Individual Subjects)",
    x = "Treatment Arm",
    y = "Baseline BMI (kg/m^2)"
  )

---

20 Longitudinal Plots from ADLB (Intermediate)

Now we use ADLB to illustrate time course plots, which are common in clinical reports.

20.1 Line Plot: Mean ALT Over Time by ARM

First summarise ADLB:

adlb_alt_summary <- ADLB %>%
  group_by(TRTA, AVISITN,AVISIT) %>%
  summarise(
    mean_ALT = mean(AVAL,na.rm=TRUE),
    sd_ALT   = sd(AVAL,na.rm=TRUE),
    n        = n(),
    se_ALT   = sd_ALT / sqrt(n),
    .groups = "drop"
  )

adlb_alt_summary$AVISIT <- factor(adlb_alt_summary$AVISIT, levels=unique(adlb_alt_summary$AVISIT[order(adlb_alt_summary$AVISITN)]))  
adlb_alt_summary

# A tibble: 27 × 7
   TRTA                 AVISITN AVISIT             mean_ALT sd_ALT     n se_ALT
   <chr>                  <dbl> <fct>                 <dbl>  <dbl> <int>  <dbl>
 1 Placebo                    0 "        Baseline"     17.6   9.22    86  0.994
 2 Placebo                    2 "          Week 2"     18.0  12.5     83  1.38 
 3 Placebo                    4 "          Week 4"     18.7  12.9     79  1.45 
 4 Placebo                    6 "          Week 6"     17.0   9.92    73  1.16 
 5 Placebo                    8 "          Week 8"     16.7   9.34    72  1.10 
 6 Placebo                   12 "         Week 12"     18     9.16    67  1.12 
 7 Placebo                   16 "         Week 16"     17.1   7.39    68  0.897
 8 Placebo                   20 "         Week 20"     16.1   6.56    65  0.814
 9 Placebo                   24 "         Week 24"     17.9  15.6     57  2.07 
10 Xanomeline High Dose       0 "        Baseline"     19.2  10.0     84  1.10 
# ℹ 17 more rows

Line plot with error bars:

ggplot(adlb_alt_summary,
       aes(x = AVISIT, y = mean_ALT, group = TRTA, color = TRTA)) +
  geom_line(linewidth = 1.1) +
  geom_point(size = 2.5) +
  geom_errorbar(
    aes(ymin = mean_ALT - se_ALT, ymax = mean_ALT + se_ALT),
    width = 0.15
  ) +
  labs(
    title = "Mean ALT Over Time by Treatment Arm",
    x = "Visit",
    y = "Mean ALT (U/L)",
    color = "Treatment Arm"
  ) +
  theme_minimal()

---

20.2 Individual Profiles (Spaghetti Plot)

To see variability, we can plot individual patients (subset to avoid overcrowding):

ADLB_sub <- ADLB |> 
  head(20)
ADLB_sub$AVISIT <- factor(ADLB_sub$AVISIT, levels=unique(ADLB_sub$AVISIT[order(ADLB_sub$AVISITN)]))  
ggplot(ADLB_sub,
       aes(x = AVISIT, y = AVAL, group = USUBJID, color = TRTA)) +
  geom_line(alpha = 0.6) +
  geom_point(alpha = 0.6, size = 1.5) +
  labs(
    title = "Individual ALT Trajectories ",
    x = "Visit",
    y = "ALT (U/L)",
    color = "Arm"
  ) +
  theme_minimal() +
  theme(legend.position = "bottom")

---

21 Faceting with ADaM Data

Faceting is useful when splitting results by subgroups, lab parameters, or regions.

Here we only have PARAMCD = "ALT", but we can pretend we have multiple parameters.

21.1 Example: Facet ALT by Sex

ggplot(adlb_alt_summary,
       aes(x = AVISIT, y = mean_ALT, group = TRTA, color = TRTA)) +
  geom_line(linewidth = 1) +
  geom_point(size = 2) +
  facet_wrap(~ TRTA) +
  labs(
    title = "Mean ALT Over Time – Faceted by Treatment TRTA",
    x = "Visit",
    y = "Mean ALT (U/L)",
    color = "TRTA"
  ) +
  theme_minimal()

You can similarly facet by SEX, RACE, or any other ADSL variable once merged into ADLB.

---

22 Scales, Labels, and Themes in ADaM Context

22.1 Changing Axis Labels and Titles

ggplot(ADSL, aes(x = AGE, y = BMIBL, color = TRT01A)) +
  geom_point(alpha = 0.8) +
  labs(
    title = "Baseline BMI vs Age by Treatment TRTA",
    subtitle = "Simulated ADaM ADSL Data",
    x = "Age (Years)",
    y = "Baseline BMI (kg/m^2)",
    color = "Treatment TRTA"
  )

22.2 Controlling Color Scales

For discrete TRTA:

ggplot(ADSL, aes(x = AGE, y = BMIBL, color = TRT01A)) +
  geom_point(size = 2.5) +
  scale_color_brewer(palette = "Set1") +
  labs(
    title = "Custom Color Scale for TRTA",
    color = "TRTA"
  ) +
  theme_minimal()

For continuous values (e.g. ALT):

ggplot(ADLB, aes(x = AGE, y = AVAL, color = AVAL)) +
  geom_point(alpha = 0.7) +
  scale_color_gradient(low = "lightyellow", high = "darkred") +
  labs(
    title = "ALT vs Age (Colored by ALT)",
    x = "Age (Years)",
    y = "ALT (U/L)",
    color = "ALT"
  ) +
  theme_minimal()

22.3 Transforming Scales (e.g., log10 ALT)

ggplot(ADLB, aes(x = AGE, y = AVAL, color = TRTA)) +
  geom_point(alpha = 0.7) +
  scale_y_log10() +
  labs(
    title = "ALT vs Age (Log10 ALT)",
    x = "Age (Years)",
    y = "ALT (U/L) – log10 scale",
    color = "TRTA"
  ) +
  theme_minimal()

---

23 Themes and Publication-Style Graphics

23.1 Using Built-In Themes

p_base <- ggplot(ADSL, aes(x = AGE, y = BMIBL, color = TRT01A)) +
  geom_point(size = 2.2, alpha = 0.8) +
  labs(
    title = "Baseline BMI vs Age by TRTA",
    x = "Age (Years)",
    y = "Baseline BMI"
  )

p_base + theme_bw()

p_base + theme_minimal()

p_base + theme_classic()

23.2 Custom Theme for Clinical Reports

theme_adam_pub <- function() {
  theme_minimal(base_size = 11) +
    theme(
      plot.title = element_text(face = "bold", hjust = 0.5, size = 13),
      axis.title = element_text(face = "bold"),
      panel.grid.major = element_line(color = "grey85"),
      panel.grid.minor = element_blank(),
      legend.position = "bottom",
      legend.title = element_text(face = "bold")
    )
}

ADLB$AVISIT <- factor(ADLB$AVISIT, levels=unique(ADLB$AVISIT[order(ADLB$AVISITN)])) 
ggplot(ADLB, aes(x = AVISIT, y = AVAL, color = TRTA)) +
  geom_boxplot() +
  labs(
    title = "ALT Distribution by Visit and TRTA",
    x = "Visit",
    y = "ALT (U/L)",
    color = "TRTA"
  ) +
  theme_adam_pub()

---

24 Coordinates and Positions

24.1 Flipping Coordinates: Boxplot of BMI by Sex

ggplot(ADSL, aes(x = SEX, y = BMIBL)) +
  geom_boxplot(fill = "lightblue") +
  coord_flip() +
  labs(
    title = "Baseline BMI by Sex",
    x = "Sex",
    y = "Baseline BMI"
  )

24.2 Stacked vs Dodged Bar Plots: TRTA by SEX

ggplot(ADSL, aes(x = TRTA, fill = SEX)) +
  geom_bar(position = "stack") +
  labs(
    title = "Subjects by Treatment and Sex (Stacked)",
    x = "TRTA",
    y = "Count",
    fill = "Sex"
  )

ggplot(ADSL, aes(x = TRTA, fill = SEX)) +
  geom_bar(position = position_dodge(width = 0.8)) +
  labs(
    title = "Subjects by Treatment and Sex (Dodged)",
    x = "TRTA",
    y = "Count",
    fill = "Sex"
  )

---

25 Adding Statistical Layers

25.1 Linear Regression: BMIBL vs AGE

ggplot(ADSL, aes(x = AGE, y = BMIBL)) +
  geom_point(alpha = 0.7) +
  stat_smooth(method = "lm", se = TRUE, color = "red") +
  labs(
    title = "Regression of Baseline BMI on Age",
    x = "Age (Years)",
    y = "Baseline BMI"
  ) +
  theme_minimal()

25.2 Summaries: Mean ALT ± SD per Visit and TRTA

We already computed adlb_alt_summary. Plot mean ± SD:

ggplot(adlb_alt_summary,
       aes(x = AVISIT, y = mean_ALT, color = TRTA, group = TRTA)) +
  geom_point(size = 2.5) +
  geom_line() +
  geom_errorbar(
    aes(ymin = mean_ALT - sd_ALT, ymax = mean_ALT + sd_ALT),
    width = 0.2, alpha = 0.7
  ) +
  labs(
    title = "Mean ALT ± SD by Visit and TRTA",
    x = "Visit",
    y = "ALT (U/L)",
    color = "TRTA"
  ) +
  theme_minimal()

---

26 Tidy Workflow: dplyr + ggplot2 with ADaM

A very common pattern in clinical reporting:

1. Start from ADaM dataset
2. Filter / summarise with dplyr
3. Plot with ggplot2

Example: Mean BMI by TRTA and SEX:

ADSL %>%
  group_by(TRTA, SEX) %>%
  summarise(
    mean_BMIBL = mean(BMIBL),
    sd_BMIBL   = sd(BMIBL),
    n          = n(),
    .groups = "drop"
  ) %>%
  ggplot(aes(x = TRTA, y = mean_BMIBL, fill = SEX)) +
  geom_col(position = position_dodge(width = 0.8)) +
  geom_errorbar(
    aes(ymin = mean_BMIBL - sd_BMIBL, ymax = mean_BMIBL + sd_BMIBL),
    position = position_dodge(width = 0.8),
    width = 0.2
  ) +
  labs(
    title = "Mean Baseline BMI by Treatment TRTA and Sex",
    x = "Treatment TRTA",
    y = "Mean Baseline BMI",
    fill = "Sex"
  ) +
  theme_adam_pub()

---

27 Advanced: Simple Kaplan–Meier Plot from ADTTE

In ADaM, ADTTE contains time-to-event information (e.g. PFS, OS).

Here we show a simple KM plot using survminer::ggsurvplot() which uses ggplot2 under the hood.

adtte_pfs <- ADTTE %>%
  filter(PARAMCD == "PFS")

fit_pfs <- survfit(Surv(AVAL, 1 - CNSR) ~ TRTA, data = adtte_pfs)

km_plot <- ggsurvplot(
  fit_pfs,
  data = adtte_pfs,
  conf.int = TRUE,
  risk.table = TRUE,
  risk.table.height = 0.25,
  ggtheme = theme_minimal(),
  legend.title = "TRTA",
  legend.labs = levels(factor(adtte_pfs$TRTA)),
  title = "Kaplan–Meier Curve for PFS by Treatment TRTA",
  xlab = "Time (Months)",
  ylab = "Progression-Free Survival Probability"
)

km_plot

For a pure ggplot2 approach, you could extract the survfit summary into a data frame and use geom_step() manually, but ggsurvplot() is convenient and still fully compatible with ggplot theming.

---

28 Advanced: Annotations and Secondary Axes

28.1 Annotations on an ADaM Plot

Example: highlight older patients with high BMI:

ggplot(ADSL, aes(x = AGE, y = BMIBL)) +
  geom_point(alpha = 0.7) +
  annotate(
    "rect", xmin = 70, xmax = 80,
    ymin = 30, ymax = 40,
    alpha = 0.1, fill = "red"
  ) +
  annotate(
    "text", x = 75, y = 41,
    label = "Older, higher BMI
(Example region)",
    size = 3, color = "red"
  ) +
  labs(
    title = "Annotations on ADSL Scatterplot",
    x = "Age (Years)",
    y = "Baseline BMI"
  ) +
  theme_minimal()

28.2 Secondary Axis (Use with Care)

ggplot2 allows secondary axes only if they are a simple transformation of the primary axis.

Example (purely illustrative) converting ALT to approximate multiple:

ggplot(adlb_alt_summary, aes(x = AVISIT, y = mean_ALT, group = TRTA, color = TRTA)) +
  geom_line() +
  geom_point() +
  scale_y_continuous(
    name = "ALT (U/L)",
    sec.axis = sec_axis(~ . / 40, name = "ALT (x Upper Normal Limit)")
  ) +
  labs(
    title = "ALT Mean Over Time with Secondary Axis",
    x = "Visit",
    color = "TRTA"
  ) +
  theme_minimal()

In production, ensure the secondary axis has a clinically correct and meaningful transformation.

---

29 Writing Reusable Plot Functions for ADaM

You can encapsulate standard plots (e.g., lab time course by TRTA) into functions.

plot_lab_timecourse <- function(adlb_df, paramcd = "ALT") {
  df <- adlb_df %>%
    filter(PARAMCD == paramcd) %>%
    group_by(TRTA, AVISIT) %>%
    summarise(
      mean_aval = mean(AVAL),
      se_aval   = sd(AVAL) / sqrt(n()),
      .groups = "drop"
    )

  ggplot(df, aes(x = AVISIT, y = mean_aval, color = TRTA, group = TRTA)) +
    geom_line(linewidth = 1) +
    geom_point(size = 2.5) +
    geom_errorbar(
      aes(ymin = mean_aval - se_aval, ymax = mean_aval + se_aval),
      width = 0.15
    ) +
    labs(
      title = paste("Mean", paramcd, "Over Time by TRTA"),
      x = "Visit",
      y = paste("Mean", paramcd),
      color = "TRTA"
    ) +
    theme_adam_pub()
}

plot_lab_timecourse(ADLB, paramcd = "ALT")

This approach is very powerful for building standardized plotting utilities for your ADaM pipeline.

---

30 Saving Plots for Reports

Use ggsave() to save plots for use in RTF, Word, or PowerPoint.

p <- ggplot(ADSL, aes(x = AGE, y = BMIBL, color = TRTA)) +
  geom_point() +
  labs(
    title = "Baseline BMI vs Age by TRTA",
    x = "Age (Years)",
    y = "Baseline BMI"
  ) +
  theme_adam_pub()

# Save as PNG
ggsave("bmi_vs_age_by_TRTA.png", plot = p, width = 6, height = 4, dpi = 300)

# Save as PDF
ggsave("bmi_vs_age_by_TRTA.pdf", plot = p, width = 6, height = 4)

31 Introduction

This document walks through six common oncology graphs step by step, using the exact code you provided:

1. Bar graph of sum of lesion diameters plus individual lesions
   
2. Waterfall plot of best percentage change from baseline
   
3. Swimmer plot for duration of response
   
4. Spider plot (tumor trajectories over time)
   
5. Kaplan–Meier survival curve
   
6. Forest plot for subgroup hazard ratios

For each figure, we will:

• Build a simple example dataset in R (similar structure to what you’d get from ADaM like ADTR/ADTTE).
• Explain the main ggplot2 layers and options.
• Show the final code to generate the plot.

In a real clinical project, you would replace the toy data here with your ADaM datasets, but the plotting logic stays the same.

---

32 Figure 1 – Bar Graph with Individual Lesion Lines

This plot shows, for a single subject:

• Bars: Sum of Lesion Diameters (SUMLD) at each visit
  
• Lines/points: each individual lesion (tumor) over time

Clinically, this helps you see how the total tumor burden and individual lesions evolve across visits.

32.1 1.1 Create bar-graph data

We first build a small dataset with:

• visit: visit label
• SUMLD: sum of lesion diameters
• trstresn_tumor1/2/3: individual lesion diameters

bar_data <- data.frame(
  visit = c("Baseline", "Week 4", "Week 8", "Week 12", "Week 16"),
  SUMLD = c(100, 85, 70, 65, 60),  # Sum of Lesion Diameters
  trstresn_tumor1 = c(50, 45, 35, 30, 28),
  trstresn_tumor2 = c(30, 25, 20, 20, 18),
  trstresn_tumor3 = c(20, 15, 15, 15, 14)
)

bar_data

     visit SUMLD trstresn_tumor1 trstresn_tumor2 trstresn_tumor3
1 Baseline   100              50              30              20
2   Week 4    85              45              25              15
3   Week 8    70              35              20              15
4  Week 12    65              30              20              15
5  Week 16    60              28              18              14

In real ADaM:

• visit could map to AVISIT,
• SUMLD would be a derived variable (sum of target lesions),
• trstresn_tumorX are separate lesions (often separate rows in ADTR; here we keep them as columns for simplicity).

32.2 1.2 Reshape to long format for lines

To draw multiple lesions as separate lines, we reshape the wide columns (trstresn_tumor1/2/3) into a long format using pivot_longer():

bar_data_long <- bar_data %>%
  pivot_longer(
    cols = starts_with("trstresn"), 
    names_to = "tuloc", 
    values_to = "trstresn"
  )

bar_data_long

# A tibble: 15 × 4
   visit    SUMLD tuloc           trstresn
   <chr>    <dbl> <chr>              <dbl>
 1 Baseline   100 trstresn_tumor1       50
 2 Baseline   100 trstresn_tumor2       30
 3 Baseline   100 trstresn_tumor3       20
 4 Week 4      85 trstresn_tumor1       45
 5 Week 4      85 trstresn_tumor2       25
 6 Week 4      85 trstresn_tumor3       15
 7 Week 8      70 trstresn_tumor1       35
 8 Week 8      70 trstresn_tumor2       20
 9 Week 8      70 trstresn_tumor3       15
10 Week 12     65 trstresn_tumor1       30
11 Week 12     65 trstresn_tumor2       20
12 Week 12     65 trstresn_tumor3       15
13 Week 16     60 trstresn_tumor1       28
14 Week 16     60 trstresn_tumor2       18
15 Week 16     60 trstresn_tumor3       14

• tuloc acts like tumor location / lesion identifier.
  
• trstresn holds the lesion diameter for each visit–lesion combination.

32.3 1.3 Build the bar + line plot

Now we combine:

• geom_col() for SUMLD (bars)
• geom_line() and geom_point() for individual lesions
• scale_y_continuous() with a secondary axis label (note: both axes use same numeric scale, only labels differ)

p1 <- ggplot() +
  # Bars = Sum of lesion diameters per visit
  geom_col(
    data = bar_data, 
    aes(x = visit, y = SUMLD), 
    fill = "#4472C4", 
    width = 0.6,
    alpha = 0.7
  ) +
  # Lines = individual lesions
  geom_line(
    data = bar_data_long, 
    aes(x = visit, y = trstresn, 
        group = tuloc, color = tuloc),
    size = 1.2
  ) +
  # Points = individual lesion measurements
  geom_point(
    data = bar_data_long, 
    aes(x = visit, y = trstresn, 
        color = tuloc),
    size = 3
  ) +
  # Primary and secondary y-axes (same numeric scale)
  scale_y_continuous(
    name = "Lesion Diameter (mm)",
    sec.axis = sec_axis(~., name = "Sum of Lesion Diameters (mm)")
  ) +
  labs(
    title = "Figure 1
Tumor Response in Individual Patients with Measurable Disease",
    subtitle = "(ITT Population)
Subject: 1001",
    x = "Visit",
    color = "Tumor Location"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold", size = 14),
    plot.subtitle = element_text(hjust = 0.5, size = 10),
    axis.title = element_text(face = "bold", size = 11),
    legend.position = "bottom"
  )

p1

Key points:

• We used an empty ggplot() and passed data separately to each layer.
  
• This is useful when we have different data frames for bars (bar_data) and lines (bar_data_long).

---

33 Figure 2 – Waterfall Plot of Best Percentage Change from Baseline

Waterfall plots show each patient’s best % change from baseline in tumor size.
They’re very common in oncology to visualize response categories like CR/PR/SD/PD.

33.1 2.1 Create waterfall data

We simulate:

• patientid: patient identifier
• pchg: best % change from baseline in sum of diameters
• response: categorical RECIST-like response (CR/PR/SD/PD)

waterfall_data <- data.frame(
  patientid = paste0("PT", sprintf("%03d", 1:30)),
  pchg = c(-65, -55, -48, -42, -38, -35, -30, -28, -25, -22,
           -18, -15, -12, -10, -8, -5, -2, 5, 8, 12,
           15, 20, 25, 30, 35, 40, 45, 50, 55, 60),
  response = c(rep("CR", 2), rep("PR", 13), rep("SD", 10), rep("PD", 5))
)

waterfall_data

   patientid pchg response
1      PT001  -65       CR
2      PT002  -55       CR
3      PT003  -48       PR
4      PT004  -42       PR
5      PT005  -38       PR
6      PT006  -35       PR
7      PT007  -30       PR
8      PT008  -28       PR
9      PT009  -25       PR
10     PT010  -22       PR
11     PT011  -18       PR
12     PT012  -15       PR
13     PT013  -12       PR
14     PT014  -10       PR
15     PT015   -8       PR
16     PT016   -5       SD
17     PT017   -2       SD
18     PT018    5       SD
19     PT019    8       SD
20     PT020   12       SD
21     PT021   15       SD
22     PT022   20       SD
23     PT023   25       SD
24     PT024   30       SD
25     PT025   35       SD
26     PT026   40       PD
27     PT027   45       PD
28     PT028   50       PD
29     PT029   55       PD
30     PT030   60       PD

33.2 2.2 Sort patients by change

To get the typical waterfall shape (bars sorted from best to worst response), we sort by pchg and then set factor levels:

waterfall_data <- waterfall_data %>%
  arrange(pchg) %>%
  mutate(patientid = factor(patientid, levels = patientid))

response_colors <- c(
  "CR" = "#00B050",
  "PR" = "#92D050", 
  "SD" = "#FFC000",
  "PD" = "#FF0000"
)

waterfall_data

   patientid pchg response
1      PT001  -65       CR
2      PT002  -55       CR
3      PT003  -48       PR
4      PT004  -42       PR
5      PT005  -38       PR
6      PT006  -35       PR
7      PT007  -30       PR
8      PT008  -28       PR
9      PT009  -25       PR
10     PT010  -22       PR
11     PT011  -18       PR
12     PT012  -15       PR
13     PT013  -12       PR
14     PT014  -10       PR
15     PT015   -8       PR
16     PT016   -5       SD
17     PT017   -2       SD
18     PT018    5       SD
19     PT019    8       SD
20     PT020   12       SD
21     PT021   15       SD
22     PT022   20       SD
23     PT023   25       SD
24     PT024   30       SD
25     PT025   35       SD
26     PT026   40       PD
27     PT027   45       PD
28     PT028   50       PD
29     PT029   55       PD
30     PT030   60       PD

33.3 2.3 Build the waterfall plot

• geom_col() for bars
  
• geom_hline() to draw RECIST cut-off lines (e.g. -30%, +20%)
  
• coord_cartesian() to fix y-axis limits

p2 <- ggplot(waterfall_data, aes(x = patientid, y = pchg, fill = response)) +
  geom_col(width = 0.8) +
  # Baseline line at 0% change
  geom_hline(yintercept = 0, linetype = "solid", color = "black", size = 0.8) +
  # Typical RECIST thresholds
  geom_hline(yintercept = -30, linetype = "dashed", color = "blue", size = 0.6) +
  geom_hline(yintercept = 20, linetype = "dashed", color = "red", size = 0.6) +
  scale_fill_manual(values = response_colors) +
  labs(
    title = "Figure 2
Waterfall Plot of Best Percentage Change from Baseline",
    subtitle = "in Target Lesion per RECIST 1.1",
    x = "Patient ID",
    y = "Best % Change from Baseline in Sum of Diameters",
    fill = "Response"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold", size = 14),
    plot.subtitle = element_text(hjust = 0.5, size = 10),
    axis.title = element_text(face = "bold", size = 11),
    axis.text.x = element_blank(),
    axis.ticks.x = element_blank(),
    legend.position = "bottom"
  ) +
  coord_cartesian(ylim = c(-80, 80))

p2

In real ADaM:

• You’d compute pchg from ADTR using baseline and best post-baseline sum of diameters.

---

34 Figure 3 – Swimmer Plot: Duration of Response

Swimmer plots show time on treatment / response for each patient:

• Horizontal bar per patient: treatment duration
  
• Color: response category
  
• Symbols/markers: progression, death, etc.

34.1 3.1 Create swimmer data

We simulate:

• start_time / end_time: treatment/response duration (months)
• response: CR/PR/SD/PD
• progression / death: logical indicators

set.seed(123)  # for reproducibility

swimmer_data <- data.frame(
  patientid = paste0("PT", sprintf("%03d", 1:20)),
  start_time = rep(0, 20),
  end_time = c(12, 15, 8, 20, 18, 14, 10, 22, 16, 19,
               7, 25, 13, 17, 11, 9, 21, 24, 15, 18),
  response = sample(c("CR", "PR", "SD", "PD"), 20, replace = TRUE),
  progression = c(TRUE, FALSE, TRUE, FALSE, TRUE, FALSE, TRUE, FALSE, TRUE, FALSE,
                  TRUE, FALSE, TRUE, FALSE, TRUE, FALSE, TRUE, FALSE, TRUE, FALSE),
  death = c(FALSE, TRUE, FALSE, FALSE, FALSE, TRUE, FALSE, FALSE, FALSE, TRUE,
            FALSE, FALSE, FALSE, TRUE, FALSE, FALSE, FALSE, TRUE, FALSE, FALSE)
)

# Sort by duration so longest bars at top
swimmer_data <- swimmer_data %>%
  arrange(desc(end_time)) %>%
  mutate(patientid = factor(patientid, levels = patientid))

swimmer_data

   patientid start_time end_time response progression death
1      PT012          0       25       PR       FALSE FALSE
2      PT018          0       24       CR       FALSE  TRUE
3      PT008          0       22       PR       FALSE FALSE
4      PT017          0       21       PD        TRUE FALSE
5      PT004          0       20       PR       FALSE FALSE
6      PT010          0       19       CR       FALSE  TRUE
7      PT005          0       18       SD        TRUE FALSE
8      PT020          0       18       SD       FALSE FALSE
9      PT014          0       17       CR       FALSE  TRUE
10     PT009          0       16       SD        TRUE FALSE
11     PT002          0       15       SD       FALSE  TRUE
12     PT019          0       15       SD        TRUE FALSE
13     PT006          0       14       PR       FALSE  TRUE
14     PT013          0       13       PR        TRUE FALSE
15     PT001          0       12       SD        TRUE FALSE
16     PT015          0       11       PR        TRUE FALSE
17     PT007          0       10       PR        TRUE FALSE
18     PT016          0        9       SD       FALSE FALSE
19     PT003          0        8       SD        TRUE FALSE
20     PT011          0        7       PD        TRUE FALSE

34.2 3.2 Build the swimmer plot

• geom_segment() draws horizontal bars
  
• Triangles (shape = 17) mark progression
  
• Crosses (shape = 4) mark death

p3 <- ggplot(swimmer_data, aes(y = patientid)) +
  # Treatment/response duration
  geom_segment(
    aes(x = start_time, xend = end_time, 
        yend = patientid, color = response),
    size = 3
  ) +
  # ▲ progression (black triangle)
  geom_point(
    data = swimmer_data %>% filter(progression),
    aes(x = end_time), 
    shape = 17, size = 4, color = "black"
  ) +
  # ✕ death (red cross)
  geom_point(
    data = swimmer_data %>% filter(death),
    aes(x = end_time), 
    shape = 4, size = 4, color = "red", stroke = 2
  ) +
  scale_color_manual(values = c(
    "CR" = "#00B050", "PR" = "#92D050",
    "SD" = "#FFC000", "PD" = "#FF0000"
  )) +
  labs(
    title = "Figure 3
Swimmer Plot - Duration of Response",
    subtitle = "Individual Patient Treatment Duration and Response",
    x = "Time (Months)",
    y = "Patient ID",
    color = "Response"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold", size = 14),
    plot.subtitle = element_text(hjust = 0.5, size = 10),
    axis.title = element_text(face = "bold", size = 11),
    legend.position = "bottom"
  ) +
  annotate(
    "text", x = 26, y = 21, 
    label = "▼ = Progression
✕ = Death", 
    hjust = 0, size = 3
  )

p3

In an ADaM workflow, these durations often come from ADTTE / custom response datasets (first response date, progression date, death date, etc.).

---

35 Figure 4 – Spider Plot: Tumor Trajectories Over Time

Spider plots show multiple patients’ tumor trajectories on the same axes:

• x-axis: time (e.g. months from baseline scan)
  
• y-axis: relative tumor size (ratio vs baseline)

Each line represents one patient.

35.1 4.1 Create spider data

We create data for 15 patients, with tumor size values over time:

set.seed(123)

spider_data <- expand.grid(
  patientid = paste0("PT", sprintf("%03d", 1:15)),
  dur = seq(-10, 100, by = 10)
) %>%
  group_by(patientid) %>%
  mutate(
    # Simple random trend around baseline = 1
    size = 1 + rnorm(1, 0, 0.3) + (dur/100) * rnorm(1, -0.5, 0.4)
  ) %>%
  ungroup()

spider_data

# A tibble: 180 × 3
   patientid   dur  size
   <fct>     <dbl> <dbl>
 1 PT001       -10 0.891
 2 PT002       -10 1.51 
 3 PT003       -10 1.02 
 4 PT004       -10 1.24 
 5 PT005       -10 0.862
 6 PT006       -10 1.40 
 7 PT007       -10 1.17 
 8 PT008       -10 0.812
 9 PT009       -10 1.28 
10 PT010       -10 1.28 
# ℹ 170 more rows

• At dur = 0 we are around size = 1 (baseline).
  
• Later timepoints can go up (> 1) or down (< 1).

35.2 4.2 Build the spider plot

p4 <- ggplot(
  spider_data, 
  aes(
    x = dur, y = size, 
    group = patientid, 
    color = patientid
  )
) +
  geom_line(size = 1.2, alpha = 0.7) +
  geom_point(size = 2, alpha = 0.8) +
  # Horizontal line at baseline (1.0)
  geom_hline(
    yintercept = 1.0, linetype = "solid", 
    color = "black", size = 0.8
  ) +
  scale_y_continuous(breaks = seq(0, 2, by = 0.20)) +
  scale_x_continuous(breaks = seq(-100, 100, by = 10)) +
  labs(
    title = "Figure 4
Relative Change in Tumor Size",
    x = "Months from Baseline Scan (Time 0)",
    y = "Tumor Size Relative to Baseline"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(hjust = 0.5, face = "bold", size = 14),
    axis.title = element_text(face = "bold", size = 11),
    legend.position = "none"
  ) +
  coord_cartesian(xlim = c(-10, 100), ylim = c(0, 2))

p4

In practice, this would also come from ADTR (longitudinal tumor measurements) with a derived relative_size = AVAL / BASE.

---

36 Figure 5 – Kaplan–Meier Survival Curve

Kaplan–Meier (KM) plots visualize time-to-event endpoints like PFS or OS:

• Step function for survival probability over time
  
• Separate curve for each treatment arm
  
• Optionally: confidence intervals and number-at-risk table

36.1 5.1 Create KM data

We simulate:

• time: event/censoring time
  
• status: 1 = event, 0 = censored
  
• treatment: Treatment A / Treatment B

set.seed(456)

km_data <- data.frame(
  time = c(rexp(100, 0.05), rexp(100, 0.03)),
  status = c(rbinom(100, 1, 0.7), rbinom(100, 1, 0.6)),
  treatment = rep(c("Treatment A", "Treatment B"), each = 100)
)

head(km_data)

       time status   treatment
1 50.245343      1 Treatment A
2 41.407858      0 Treatment A
3  9.318211      1 Treatment A
4  4.601942      1 Treatment A
5 47.967306      1 Treatment A
6 16.705099      1 Treatment A

36.2 5.2 Fit the KM model and plot

We use survival::survfit() and survminer::ggsurvplot():

km_fit <- survfit(Surv(time, status) ~ treatment, data = km_data)

p5 <- ggsurvplot(
  km_fit,
  data = km_data,
  pval = TRUE,           # log-rank p-value
  conf.int = TRUE,       # confidence interval bands
  risk.table = TRUE,     # number at risk under the plot
  risk.table.height = 0.25,
  ggtheme = theme_minimal(),
  palette = c("#E7B800", "#2E9FDF"),
  title = "Figure 5
Kaplan-Meier Survival Curve",
  xlab = "Time (Months)",
  ylab = "Survival Probability",
  legend.title = "Treatment Group",
  legend.labs = c("Treatment A", "Treatment B"),
  font.main = c(14, "bold"),
  font.x = c(11, "bold"),
  font.y = c(11, "bold"),
  font.legend = c(10)
)

p5

In ADaM:

• This would typically be based on ADTTE, using variables like AVAL (time), CNSR (censor flag), and TRT01A or similar.

---

37 Figure 6 – Forest Plot for Subgroup Hazard Ratios

Forest plots summarize effect estimates (e.g. hazard ratios) across subgroups:

• Each row: one subgroup
  
• Point: HR
  
• Horizontal line: 95% CI
  
• Vertical reference line at HR = 1

Here we use the forestploter package and its example data.

37.1 6.1 Load example data and prepare

dt <- read.csv(system.file("extdata", "example_data.csv", package = "forestploter"))

head(dt)

      Subgroup Treatment Placebo      est        low       hi   low_gp1
1 All Patients       781     780 1.869694 0.13245636 3.606932 0.1507971
2          Sex        NA      NA       NA         NA       NA        NA
3         Male       535     548 1.449472 0.06834426 2.830600 1.9149515
4       Female       246     232 2.275120 0.50768005 4.042560 0.6336414
5          Age        NA      NA       NA         NA       NA        NA
6       <65 yr       297     333 1.509242 0.67029394 2.348190 1.7679431
     low_gp2   low_gp3   low_gp4  est_gp1   est_gp2  est_gp3  est_gp4   hi_gp1
1 0.35443249 0.3939730 1.1515801 1.181926 1.5163554 1.612725 1.949433 2.924330
2         NA        NA        NA       NA        NA       NA       NA       NA
3 0.09953409 0.3803214 0.3213258 3.266615 0.8291053 1.642294 1.607950 3.996409
4 2.57367694 1.0229365 0.3510777 1.971712 3.9719741 1.751471 1.742973 3.965617
5         NA        NA        NA       NA        NA       NA       NA       NA
6 0.57329716 0.8433183 2.1637576 2.447051 1.9990500 1.390913 3.136695 3.674451
    hi_gp2   hi_gp3   hi_gp4
1 2.919964 3.485182 2.862447
2       NA       NA       NA
3 1.553593 2.573538 2.228632
4 5.439582 3.674689 3.249342
5       NA       NA       NA
6 3.399146 2.346744 4.299211

We:

• Indent subgroup labels when there is a number in Placebo column (for visual hierarchy).
  
• Replace NA with empty strings for display columns.
  
• Compute standard error (se) from CI.
  
• Add a blank column (" ") for the plot.
  
• Create a text column for HR (95% CI).

# Indent subgroup if there is a number in the placebo column
dt$Subgroup <- ifelse(
  is.na(dt$Placebo), 
  dt$Subgroup,
  paste0("   ", dt$Subgroup)
)

# NA to blank for display columns
dt$Treatment <- ifelse(is.na(dt$Treatment), "", dt$Treatment)
dt$Placebo   <- ifelse(is.na(dt$Placebo), "", dt$Placebo)

# Standard error derived from CI on log scale
dt$se <- (log(dt$hi) - log(dt$est)) / 1.96

# Blank column used by forestploter for CI area
dt$` ` <- paste(rep(" ", 20), collapse = " ")

# Display column for HR (95% CI)
dt$`HR (95% CI)` <- ifelse(
  is.na(dt$se), "",
  sprintf("%.2f (%.2f to %.2f)", dt$est, dt$low, dt$hi)
)

head(dt)

         Subgroup Treatment Placebo      est        low       hi   low_gp1
1    All Patients       781     780 1.869694 0.13245636 3.606932 0.1507971
2             Sex                         NA         NA       NA        NA
3            Male       535     548 1.449472 0.06834426 2.830600 1.9149515
4          Female       246     232 2.275120 0.50768005 4.042560 0.6336414
5             Age                         NA         NA       NA        NA
6          <65 yr       297     333 1.509242 0.67029394 2.348190 1.7679431
     low_gp2   low_gp3   low_gp4  est_gp1   est_gp2  est_gp3  est_gp4   hi_gp1
1 0.35443249 0.3939730 1.1515801 1.181926 1.5163554 1.612725 1.949433 2.924330
2         NA        NA        NA       NA        NA       NA       NA       NA
3 0.09953409 0.3803214 0.3213258 3.266615 0.8291053 1.642294 1.607950 3.996409
4 2.57367694 1.0229365 0.3510777 1.971712 3.9719741 1.751471 1.742973 3.965617
5         NA        NA        NA       NA        NA       NA       NA       NA
6 0.57329716 0.8433183 2.1637576 2.447051 1.9990500 1.390913 3.136695 3.674451
    hi_gp2   hi_gp3   hi_gp4        se                                        
1 2.919964 3.485182 2.862447 0.3352463                                        
2       NA       NA       NA        NA                                        
3 1.553593 2.573538 2.228632 0.3414741                                        
4 5.439582 3.674689 3.249342 0.2932884                                        
5       NA       NA       NA        NA                                        
6 3.399146 2.346744 4.299211 0.2255292                                        
          HR (95% CI)
1 1.87 (0.13 to 3.61)
2                    
3 1.45 (0.07 to 2.83)
4 2.28 (0.51 to 4.04)
5                    
6 1.51 (0.67 to 2.35)

37.2 6.2 Define forest theme and build the plot

We define a theme (fonts, refline color, arrow labels) and then call forest():

tm <- forest_theme(
  base_size = 10,
  refline_col = "red",
  arrow_type = "closed",
  footnote_gp = gpar(col = "blue", cex = 0.6)
)

p <- forest(
  dt[, c(1:3, 20:21)],
  est     = dt$est,
  lower   = dt$low, 
  upper   = dt$hi,
  sizes   = dt$se,
  ci_column = 4,
  ref_line  = 1,
  arrow_lab = c("Placebo Better", "Treatment Better"),
  xlim      = c(0, 4),
  ticks_at  = c(0.5, 1, 2, 3),
  footnote  = "This is the demo data. Please feel free to change
anything you want.",
  theme     = tm
)

plot(p)

Interpretation:

• Values < 1 favor treatment, values > 1 favor placebo (depending on how you define HR).
  
• The arrow labels at the bottom make this direction explicit.

In real analyses, the columns est, low, hi would be generated from Cox models or other regressions per subgroup.

---

38 Summary and Next Steps

In this tutorial, you:

• Learned the ggplot2 grammar of graphics using ADSL, ADLB, and ADTTE style data
• Built common clinical plots:
  • Subject counts by arm
  • Histograms/boxplots of demographics
  • Longitudinal lab plots with summary statistics
  • Simple KM curves from ADTTE
• Customized scales, themes, coordinates, and annotations
• Combined dplyr + ggplot2 for tidy ADaM workflows
• Wrote a small reusable plotting function tailored to ADaM

From here, you can:

• Swap the simulated data with your real ADaM datasets
• Extend functions for specific TLFs (e.g., safety labs, efficacy endpoints)
• Integrate these plots into R Markdown/Quarto CSRs or Shiny dashboards for interactive review.

Happy plotting with ggplot2 and ADaM!