Spider Plots for Tumor Response

Introduction

In oncology clinical trials, tumor response is not simply a single number. A patient's tumor burden can change repeatedly over the course of treatment.

A patient may initially experience substantial tumor shrinkage, remain stable for several months, and then develop tumor growth.

Another patient may experience modest shrinkage that remains stable for a long period.

A third patient may initially progress and subsequently demonstrate tumor shrinkage after a treatment modification.

A spider plot is designed to display these longitudinal patterns.

What Is a Spider Plot?

A spider plot consists of one line for each patient.

Each line begins at approximately:

because baseline tumor burden is defined as the reference value.

Subsequent measurements are expressed as percentage changes from baseline.

where:

• \(S_0\) = baseline tumor measurement
• \(S_t\) = tumor measurement at time \(t\)

A decrease in tumor burden therefore produces a negative percentage. A 30% reduction appears at:

while a 20% increase appears at:

Why Is It Called a Spider Plot?

When many patients are plotted simultaneously, the individual trajectories cross and diverge across time.

The resulting collection of lines can resemble a spider web, particularly when the study contains many patients with different tumor trajectories.

The name therefore refers to the visual appearance rather than to a specific statistical model.

The Basic Structure

A typical oncology spider plot contains:

• X-axis: time since treatment initiation
• Y-axis: percentage change from baseline
• One line: one patient
• Baseline: 0%
• Reference lines: clinically relevant response or progression thresholds

The plot is therefore fundamentally a longitudinal visualization.

An Actual Spider Plot

The following example contains simulated oncology patients with different tumor-response trajectories.

Reading the Spider Plot

The first feature to examine is the overall direction of the trajectories.

Lines moving downward indicate tumor shrinkage.

Lines moving upward indicate increasing tumor burden.

A line that remains near zero indicates relatively stable tumor burden.

Percent Change From Baseline

The fundamental quantity plotted on the y-axis is usually percentage change from baseline.

Suppose a patient's baseline sum of target-lesion diameters is:

At Week 8, suppose the sum is:

Then:

The patient would therefore appear at −30% at Week 8.

Example of Tumor Growth

Suppose another patient begins with:

and subsequently has:

The percentage change is:

The patient therefore appears at +25% at Week 16.

Why the Baseline Is Zero

Because every patient is normalized to their own baseline, the baseline measurement is:

This allows patients with very different absolute tumor burdens to be plotted on the same percentage-change scale.

RECIST 1.1 and the −30% Response Threshold

For target lesions, RECIST 1.1 defines partial response based on at least a 30% decrease in the sum of diameters from baseline.

Therefore, a commonly displayed reference line is:

A trajectory that crosses below this line has reached the magnitude of tumor shrinkage associated with the target-lesion component of partial response.

RECIST 1.1 and Progressive Disease

Progressive disease is more complicated because RECIST 1.1 evaluates tumor growth relative to the nadir, rather than simply relative to baseline.

For target lesions, progression requires at least a 20% increase from the smallest sum of diameters recorded on study, with an absolute increase of at least 5 mm, subject to the full RECIST assessment.

Thus, a spider plot can show a patient falling from:

and later rising to:

The patient has increased by:

relative to the nadir on the percentage-change scale.

Therefore, the apparent return to −15% from a nadir of −40% can represent substantial tumor regrowth even though the tumor burden remains below baseline.

Why a +20% Line Is Only an Approximation on a Spider Plot

This is an important technical point.

If the y-axis is percentage change from baseline, a horizontal +20% line means a 20% increase from baseline.

But RECIST progressive disease is based on an increase of at least 20% from the nadir, not necessarily from baseline.

Therefore, the following line:

is useful as a visual reference but should not be labeled as a complete RECIST progressive-disease boundary.

Longitudinal Response Is the Main Strength

Consider two patients.

Patient A reaches −55% at Week 8 but later returns to −10%.

Patient B reaches −35% at Week 8 and remains near −35% through Week 24.

A waterfall plot may rank both according to their best response:

• Patient A: −55%
• Patient B: −35%

Patient A therefore looks more impressive on the waterfall plot.

But the spider plot reveals that Patient B may have a much more durable response.

Spider Plot vs. Waterfall Plot

Spider and waterfall plots are complementary rather than competing displays.

The Two Figures Answer Different Questions

A waterfall plot asks:

A spider plot asks:

Together, the two figures provide a much more complete description of tumor response.

An Example of Different Patient Trajectories

Consider the following simulated patients.

A spider plot makes the differences immediately visible.

Patient 001 has a deep and durable response. Patient 004 has progressive tumor growth. Patient 005 has a deep but transient response.

Response Categories and Trajectory Patterns

Several common patterns occur in oncology trials.

Deep and Durable Response

The trajectory falls substantially below baseline and remains low.

For example:

This pattern is visually easy to recognize.

Deep but Transient Response

The trajectory initially falls substantially but subsequently rises.

The waterfall plot might emphasize the −55% best response, whereas the spider plot shows that the response was not sustained.

Stable Disease

The trajectory remains relatively close to baseline without meeting criteria for meaningful shrinkage or progression.

Progressive Disease

The trajectory rises over time, potentially following an initial period of stability or shrinkage.

Initial Progression Followed by Shrinkage

Some patients may show unusual trajectories.

The scientific interpretation depends on the treatment, disease, assessment schedule, and prespecified response criteria.

Time on the X-Axis

Time is commonly displayed as:

• Days from first dose
• Weeks from first dose
• Months from first dose
• Study visit number

Weeks from first dose are particularly common for oncology visualizations.

For example:

may correspond to baseline and assessments every four weeks.

Irregular Assessment Times

Patients do not always have tumor assessments exactly on the planned visit days.

One patient might be assessed on Day 56 while another is assessed on Day 59.

A spider plot can either:

• Use the actual assessment time
• Map assessments to nominal study visits

The preferred choice depends on the analysis convention and the purpose of the figure.

Missing Tumor Assessments

Real clinical trial datasets frequently contain missing assessments.

A patient might have:

The plot should not automatically connect the measurements across the missing assessment if doing so could imply that a measurement was actually obtained.

Instead, the graphical implementation should generally preserve the missing interval or otherwise follow the prespecified data-display convention.

Patients With Early Discontinuation

Some patients leave treatment early because of:

• Adverse events
• Progressive disease
• Withdrawal of consent
• Protocol deviations
• Clinical deterioration
• Investigator decision

Their trajectories may therefore terminate before the end of the displayed follow-up period.

This is not necessarily a plotting error.

The absence of later measurements may itself be clinically informative.

Why Spider Plots Can Become Difficult to Read

A major limitation is overplotting.

If a study contains 500 patients, plotting 500 individual lines can produce a dense visual display.

Possible approaches include:

• Using thin semi-transparent trajectories
• Separating treatment groups into panels
• Showing only a defined analysis population
• Using interactive patient selection
• Adding response-category annotations
• Displaying a representative subset for exploratory figures

However, selectively removing patients from a primary clinical-trial figure should be carefully justified.

A Second Actual Spider Plot: Response Durability

The next example emphasizes an important distinction between depth of response and duration of response.

How to Calculate the Nadir

For a patient with observed percentage changes:

the nadir is the smallest observed percentage change:

In terms of absolute tumor burden, the nadir corresponds to the smallest on-study tumor measurement used for the relevant RECIST assessment.

Nadir Is Not the Same as Best Overall Response

This distinction is important.

The nadir is a measurement-based concept.

Best overall response is a formal response category determined using the prespecified response criteria.

For example, a patient can reach substantial tumor shrinkage but still not qualify for a formal partial response because of other disease components, new lesions, confirmation requirements where applicable, or other aspects of the response algorithm.

Spider Plots and Complete Response

A complete response is not simply "the line reached −100%."

For target lesions, complete response involves disappearance of all target lesions, together with the complete RECIST assessment of non-target disease and new lesions.

Therefore, a spider plot should not generally use −100% as a simplistic substitute for the RECIST complete-response category.

Spider Plots in Randomized Trials

Spider plots can be particularly useful when comparing treatment arms.

One common approach is to create separate panels for:

• Experimental treatment
• Control treatment

The same y-axis scale should generally be maintained across panels so that the visual comparison is meaningful.

Example of Treatment-Arm Interpretation

These patterns may provide useful descriptive context, but they should not replace the prespecified treatment-effect analysis.

Spider Plots Are Descriptive, Not Inferential

A spider plot is primarily a graphical descriptive tool.

It does not by itself provide:

• A hypothesis test
• A confidence interval
• A treatment-effect estimate
• A p-value
• A multiplicity-adjusted comparison

Instead, it helps investigators understand the distribution and evolution of individual patient tumor responses.

What a Spider Plot Can Reveal That a Summary Table Cannot

Suppose two treatment groups both have an objective response rate of 40%.

The summary table suggests identical response rates.

But the spider plots might show:

• One arm has many durable responses.
• The other arm has transient responses.
• One arm has more early progression.
• One arm has deeper tumor shrinkage.

Thus, the spider plot provides a qualitative description of the underlying patient-level response behavior.

Spider Plot Data Structure

A basic dataset requires at least:

For a CDISC-oriented clinical-trial analysis, the exact source variables depend on the study's analysis datasets and derivation conventions.

Deriving Percentage Change

Suppose the analysis dataset contains baseline and post-baseline tumor burden.

The standard percentage-change calculation is:

For example, if:

then:

Data Preparation Workflow

R Implementation

A spider plot can be created using ggplot2.

library(ggplot2)

ggplot(
  data,
  aes(
    x = ADY,
    y = PCHG,
    group = USUBJID
  )
) +
  geom_line(
    alpha = 0.55,
    linewidth = 0.7
  ) +
  geom_hline(
    yintercept = 0,
    linewidth = 0.6
  ) +
  geom_hline(
    yintercept = -30,
    linetype = "dashed"
  ) +
  geom_hline(
    yintercept = 20,
    linetype = "dashed"
  ) +
  labs(
    x = "Days from First Dose",
    y = "Percent Change from Baseline"
  ) +
  theme_classic()

Adding Treatment Groups

If the study has multiple treatment groups, treatment can be mapped to color and patients can be grouped separately.

ggplot(
  data,
  aes(
    x = ADY,
    y = PCHG,
    group = USUBJID,
    color = TRT01P
  )
) +
  geom_line(
    alpha = 0.5,
    linewidth = 0.7
  ) +
  geom_hline(
    yintercept = 0,
    linewidth = 0.6
  ) +
  geom_hline(
    yintercept = -30,
    linetype = "dashed"
  ) +
  geom_hline(
    yintercept = 20,
    linetype = "dashed"
  ) +
  labs(
    x = "Days from First Dose",
    y = "Percent Change from Baseline",
    color = "Treatment"
  ) +
  theme_classic()

Faceting by Treatment Group

An alternative is to create separate panels.

ggplot(
  data,
  aes(
    x = ADY,
    y = PCHG,
    group = USUBJID
  )
) +
  geom_line(
    alpha = 0.55,
    linewidth = 0.7
  ) +
  geom_hline(
    yintercept = 0,
    linewidth = 0.6
  ) +
  geom_hline(
    yintercept = -30,
    linetype = "dashed"
  ) +
  geom_hline(
    yintercept = 20,
    linetype = "dashed"
  ) +
  facet_wrap(~TRT01P) +
  labs(
    x = "Days from First Dose",
    y = "Percent Change from Baseline"
  ) +
  theme_classic()

Adding Patient Labels

Patient labels can be useful in exploratory analyses but can quickly make a publication-quality figure unreadable.

For example:

ggplot(
  data,
  aes(
    x = ADY,
    y = PCHG,
    group = USUBJID
  )
) +
  geom_line(
    alpha = 0.45
  ) +
  geom_text(
    data = subset(
      data,
      ADY == max(ADY)
    ),
    aes(label = USUBJID),
    hjust = -0.1,
    size = 3
  )

Patient labels are generally better suited to exploratory graphics or interactive displays than to crowded CSR figures.

Controlling the Y-Axis

The y-axis should be selected so that meaningful tumor-response behavior is visible.

For example:

scale_y_continuous(
  limits = c(-100, 100),
  breaks = seq(-100, 100, 20)
)

However, fixed limits can hide important observations if patients fall outside the chosen range.

Therefore, axis limits should be selected deliberately and validated against the underlying data.

Should the Y-Axis Be Symmetric?

Not necessarily.

Oncology datasets may contain substantially more tumor growth than shrinkage, or vice versa.

The important requirement is that the axis be interpretable and not visually distort the trajectories.

Handling Extreme Values

A small number of patients may have very large increases in tumor burden.

For example:

One extreme value can compress all other trajectories toward zero if the full range is displayed.

Possible solutions include:

• Using an explicitly justified axis range
• Using a secondary exploratory plot
• Displaying an inset
• Using an axis break only when scientifically appropriate

Waterfall and Spider Plots Together

A particularly effective oncology efficacy package often includes both plots.

These figures answer complementary questions about:

• Depth of response
• Frequency of response
• Durability of response
• Time to progression
• Heterogeneity of treatment response

Common Mistakes

1. Confusing a spider plot with a waterfall plot. A spider plot preserves the longitudinal time dimension.
2. Calling every trajectory below −30% a partial response. The formal RECIST response category requires the complete assessment.
3. Treating +20% from baseline as the RECIST PD criterion. RECIST progression is assessed relative to nadir and has additional requirements.
4. Connecting measurements across missing assessments without consideration. This can imply observations that were never collected.
5. Ignoring patients who discontinue early. Early discontinuation can explain why trajectories terminate.
6. Using inconsistent time scales. Actual days, nominal visits, and months should not be mixed without explanation.
7. Overplotting hundreds of patients. Large studies may require transparency, faceting, or interactive displays.
8. Using the figure as an inferential analysis. Spider plots are descriptive and should complement, not replace, formal statistical analyses.
9. Failing to validate the plotted values. Every plotted patient trajectory should be traceable to the analysis dataset.

What Should Be Included in a Clinical Study Report?

A CSR spider plot should have a clear title and footnote.

A useful figure description might state:

Spider Plot Specifications

Before programming the figure, the statistical programmer should have a clear specification.

Spider Plot Quality Control

A spider plot should undergo both programming QC and clinical review.

At minimum, verify:

• Every patient has the correct baseline.
• Every percentage change is calculated correctly.
• Time values are correct.
• Patients are not duplicated.
• Duplicate assessments are handled appropriately.
• Missing assessments are handled correctly.
• Out-of-range values are investigated.
• Treatment assignment is correct.
• Reference lines are correctly positioned.
• The axis labels are accurate.
• The analysis population matches the SAP.

Manual Validation Example

Suppose Patient 001 has:

The plotted trajectory should therefore pass through:

If the graphic does not contain these points, the programming logic should be investigated.

Interpreting a Dense Spider Plot

When many lines are displayed, do not attempt to interpret every patient individually.

Instead, examine broad patterns:

• How many trajectories move downward?
• How many remain near baseline?
• How many rise substantially?
• How frequently is −30% crossed?
• Are responses durable?
• Do many trajectories rebound after reaching a nadir?
• Are there many early progressors?

Spider Plots and Heterogeneity

One of the most valuable properties of a spider plot is its ability to show heterogeneity.

Suppose the median percentage change is −25%.

That number could arise from:

• Most patients clustering around −25%
• Half the patients having deep responses and half having progression
• A mixture of durable and transient responses

These scenarios can have very different clinical implications despite similar summary statistics.

The spider plot makes that heterogeneity visible.

Spider Plots and Time-to-Event Endpoints

Spider plots should not be confused with Kaplan-Meier plots.

A Kaplan-Meier analysis addresses a time-to-event endpoint such as:

• Progression-free survival
• Overall survival
• Duration of response

A spider plot instead displays longitudinal tumor-burden measurements.

The two approaches can complement each other.

Spider Plot vs. Swimmer Plot

A swimmer plot is another commonly used oncology visualization.

The three figures can therefore serve different purposes:

Creating a Spider Plot From Long-Format Data

Long-format data are generally convenient for creating longitudinal plots.

For example:

USUBJID   ADY   PCHG
001       0      0
001       28   -15
001       56   -30
001       84   -40
002       0      0
002       28    -8
002       56   -20
002       84   -12
003       0      0
003       28    10
003       56    25
003       84    35

The plotting system groups observations by patient identifier.

ggplot(
  data,
  aes(
    x = ADY,
    y = PCHG,
    group = USUBJID
  )
) +
  geom_line()

Adding Points to the Trajectories

Points can be added at each assessment.

ggplot(
  data,
  aes(
    x = ADY,
    y = PCHG,
    group = USUBJID
  )
) +
  geom_line(
    alpha = 0.5
  ) +
  geom_point(
    size = 1.5
  )

This can make the actual assessment schedule more visible.

Should Points Always Be Shown?

Not necessarily.

For a study with relatively few patients, points can be useful.

For a study with hundreds of patients, points can create excessive visual clutter.

The choice should therefore depend on the number of patients and assessment frequency.

Displaying Confirmed Response

Some studies may want to distinguish trajectories belonging to patients who achieved a formal response.

This can be done using an analysis variable indicating the best overall response.

ggplot(
  data,
  aes(
    x = ADY,
    y = PCHG,
    group = USUBJID,
    color = BOR
  )
) +
  geom_line(
    alpha = 0.55
  ) +
  geom_hline(
    yintercept = -30,
    linetype = "dashed"
  ) +
  geom_hline(
    yintercept = 0
  )

However, this approach should be used carefully because the formal response category is derived from the complete response assessment rather than from the visual trajectory alone.

Spider Plot Interpretation Example

Imagine a study in which most trajectories initially decline.

At first glance, the treatment appears highly active.

However, after Week 12, many trajectories begin moving upward.

This suggests that:

• Initial tumor shrinkage is common.
• Some responses may be transient.
• Tumor regrowth may occur later during treatment.
• The timing of progression deserves additional investigation.

The waterfall plot alone might not reveal this pattern.

A Third Actual Spider Plot: Heterogeneous Response

The following figure illustrates a more realistic heterogeneous response distribution.

What the Spider Plot Does Not Tell You

The figure alone does not establish:

• Whether the treatment is statistically superior to control
• Whether observed responses are clinically meaningful
• Whether progression was treatment-related
• Whether a patient meets formal RECIST criteria
• Whether the treatment improves survival

Those questions require additional clinical and statistical analyses.

Protocol and SAP Considerations

The analysis plan should specify the derivation of the plotted tumor-burden measure.

Important considerations include:

• Definition of baseline
• Definition of target lesions
• Handling of non-target lesions
• Handling of new lesions
• Assessment timing
• Handling of unscheduled assessments
• Handling of missing assessments
• Handling of post-progression assessments
• Analysis population
• Handling of patients without post-baseline assessments

Post-Progression Measurements

One particularly important question is whether measurements obtained after formal progression should remain on the spider plot.

There is no universal answer independent of the analysis objective.

For a descriptive longitudinal tumor-burden figure, post-progression measurements may be informative.

For a figure specifically intended to display tumor response through the formal response-assessment period, the SAP may specify a different rule.

Clinical Interpretation Requires Context

A downward trajectory generally indicates tumor shrinkage, but tumor response should be interpreted within the disease and treatment context.

For example, some immunotherapy-treated patients may exhibit atypical response patterns, including delayed responses or apparent initial progression.

Such patterns should not automatically be labeled as treatment benefit without the appropriate clinical and protocol-defined assessment.

Spider Plots in Immuno-Oncology

Spider plots are particularly interesting in immuno-oncology because tumor trajectories can be heterogeneous.

Possible patterns include:

• Conventional tumor shrinkage
• Stable disease
• Delayed shrinkage
• Transient enlargement
• Mixed response
• Progression followed by subsequent response

However, specialized response criteria or protocol-specific rules may be required depending on the trial.

The Most Important RECIST Caveat

A spider plot should never be used as a visual shortcut for assigning formal RECIST categories.

For example, a patient whose trajectory reaches −32% visually satisfies the 30% tumor-shrinkage magnitude for target lesions.

But formal partial response requires the full RECIST assessment.

Similarly, a trajectory that rises from −40% to −15% has increased substantially from nadir, but formal progressive disease requires consideration of the appropriate RECIST criteria.

Waterfall + Spider + Swimmer

For a comprehensive oncology efficacy presentation, the three displays can work together particularly well.

Together, these figures provide complementary views of the same underlying patient-level clinical experience.

Practical Figure Checklist

Common Reporting Language

A concise figure footnote might state:

Key Differences From a Waterfall Plot

Summary of the Four Example Trajectories

The Most Important Concept

The most important feature of a spider plot is that it preserves the longitudinal history of each patient.

A single best-response number can hide clinically important information.

For example, two patients can both have a best response of −50%, while one maintains that response for a year and the other progresses shortly afterward.

The waterfall plot treats the two patients similarly.

The spider plot does not.

It shows the entire trajectory.

References

Eisenhauer, E.A., Therasse, P., Bogaerts, J., et al. (2009). New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1). European Journal of Cancer, 45(2), 228–247.
Therasse, P., Arbuck, S.G., Eisenhauer, E.A., et al. (2000). New guidelines to evaluate the response to treatment in solid tumors. Journal of the National Cancer Institute, 92(3), 205–216.
Seymour, L., Bogaerts, J., Perrone, A., et al. (2017). iRECIST: Guidelines for response criteria for use in trials testing immunotherapeutics. The Lancet Oncology, 18(3), e143–e152.
Nishino, M., Tirumani, S.H., Ramaiya, N.H., et al. (2014). Reviving response assessment in oncology: RECIST 1.1 and beyond. American Journal of Roentgenology, 201(1), 1–10.