Skip to main content
v2026.11,986 entries · CC-BY 4.0

Waterfall Plots: Reading and Making the Oncology Waterfall Plot

How to read a waterfall plot of best percentage tumour change, with RECIST lines, a worked example and code.

Written and maintained by CASRAI Editorial Board

Published

On this page

A waterfall plot is a bar chart in which each bar is one patient (or one subject) and the bar’s height is that patient’s best percentage change in tumour size from baseline, with the bars sorted from the largest decrease on the left to the largest increase on the right. In oncology trials it is the standard way to show the spread of tumour response across a cohort at a glance. This guide covers the clinical waterfall plot, then the unrelated finance-style waterfall chart that shares the name.

What is a waterfall plot?

In clinical research a waterfall plot displays one bar per participant. Bars below zero mean the sum of target-lesion diameters shrank; bars above zero mean it grew. Because bars are ranked, the plot reads like a cascade, which gives it the name. Bars are usually coloured or labelled by best overall response category (for example partial response, stable disease, progressive disease). A published example describes its figure as a “waterfall plot of best percentage change in sum of diameters and best overall response”, with tumour response assessed by RECIST v1.1 at 6-week intervals (PMC4646607).

It is a descriptive figure. It does not show time, duration of response, or survival; for those see how to read a Kaplan-Meier curve. For the trial design context, see what makes oncology clinical trials different.

Formula: the quantity on the y-axis

For each patient i the plotted value is:

Δᵢ = (Sᵢ − S₀ᵢ) / S₀ᵢ × 100

  • Δᵢ: percentage change for patient i (the bar height).
  • S₀ᵢ: baseline sum of target-lesion diameters (mm).
  • Sᵢ: the on-study sum used for the “best” change. In statistical analysis plans this is the largest decrease, or the smallest increase if the sum never fell.

How to read an oncology waterfall plot

Read left to right: the leftmost bars are the deepest shrinkage, the rightmost the greatest growth. Two horizontal reference lines are conventional, and they come from RECIST 1.1 (EORTC, recist.eortc.org, read 2026-11-08; Eisenhauer et al., Eur J Cancer 2009;45(2):228-247):

Line RECIST 1.1 definition Reference point
−30% Partial response: at least a 30% decrease in the sum of diameters Baseline sum
+20% Progressive disease: at least a 20% increase in the sum of diameters, with a minimum absolute increase of 5 mm, or new lesions Smallest sum on study (nadir)

The reference points differ. A bar crossing +20% on a baseline-referenced axis is not by itself a progressive-disease call, because RECIST measures progression from the smallest sum on study and also requires the 5 mm absolute increase. New lesions also count as progression but do not change a bar’s height, so a plot should flag them (for example with a marker) rather than imply the bar tells the whole story.

Worked example with real arithmetic

Illustrative composite data, invented to show the calculation; these are not results from any real trial. Seven patients, baseline and best on-study sum of diameters in mm:

Patient Baseline S₀ Best S Δ = (S − S₀) / S₀ × 100 Rank (left to right)
P1 62 31 −31/62 = −50.0% 1
P3 90 54 −36/90 = −40.0% 2
P6 40 28 −12/40 = −30.0% 3
P7 52 39 −13/52 = −25.0% 4
P2 48 40 −8/48 = −16.7% 5
P4 35 38 +3/35 = +8.6% 6
P5 70 84 +14/70 = +20.0% 7

Three bars (P1, P3, P6) reach or pass the −30% line. That is a statement about the best change in size only. RECIST response also needs the other criteria (non-target lesions, no new lesions, and confirmation where the protocol requires it), so you should not call 3/7 a response rate from this figure alone.

Interpretation: reading guidance

What you see What it supports
Bar at or below −30% Size change meets the partial-response size threshold, pending the other RECIST criteria
Bar between −30% and 0 Shrinkage that did not reach the threshold
Bar above 0 Best change was an increase
Long run of bars below zero Broad shrinkage across the cohort, without any statement about duration

Assumptions and common mistakes

  • Best versus last change. The plot uses the best (largest decrease) change, which is more flattering than the final measurement. Say so in the legend.
  • Missing patients. Patients with no post-baseline assessment cannot have a bar. Analysis plans typically exclude them, so report how many and why; a plot of only evaluable patients overstates activity. This relates to intention-to-treat thinking.
  • Unconfirmed responses. In PMC4646607, an unconfirmed decrease of 30% or more was classed as stable disease, so bar height and the response label can disagree by design.
  • Reading it as survival. A deep bar says nothing about how long the effect lasted.
  • Pre-specification. Define the ordering, the “best” rule and the reference lines in the statistical analysis plan before seeing data.

Waterfall chart (finance) versus waterfall plot (oncology)

The same word covers a different chart. A finance-style waterfall chart, such as Plotly’s go.Waterfall, shows a running total: each bar is a “relative” step added to or subtracted from the previous one, and “total” bars reset or summarise (plotly.com/python/waterfall-charts, read 2026-11-08). Bars are not sorted and not per-subject. If you want a bridge from a starting value to an ending value, use that; if you want per-patient tumour change, use the plot described here.

How to make a waterfall plot in Python, R or Excel

Python (matplotlib)

import matplotlib.pyplot as plt
base = {'P1':62,'P2':48,'P3':90,'P4':35,'P5':70,'P6':40,'P7':52}
best = {'P1':31,'P2':40,'P3':54,'P4':38,'P5':84,'P6':28,'P7':39}
d = sorted(((k,(best[k]-base[k])/base[k]*100) for k in base), key=lambda t:t[1])
fig, ax = plt.subplots()
ax.bar([k for k,_ in d],[v for _,v in d])
ax.axhline(-30, ls=':'); ax.axhline(20, ls=':')
ax.set_ylabel('Best change from baseline (%)')
plt.show()

R (ggplot2)

library(ggplot2)
df <- data.frame(id=c('P1','P2','P3','P4','P5','P6','P7'),
  base=c(62,48,90,35,70,40,52), best=c(31,40,54,38,84,28,39))
df$chg <- (df$best - df$base)/df$base*100
ggplot(df, aes(reorder(id, chg), chg)) + geom_col() +
  geom_hline(yintercept=c(-30,20), linetype='dotted') +
  labs(x=NULL, y='Best change from baseline (%)')

Excel

Put baseline in column B and best sum in column C, compute =(C2-B2)/B2*100 in column D, sort the table by column D ascending, then insert a clustered column chart of column D. Add the two reference lines as extra series with constant values −30 and 20.

How to report a waterfall plot

Put the definition in the figure note. An illustrative APA-style caption: Figure 1. Best percentage change from baseline in the sum of target-lesion diameters per patient (n = 7 evaluable). Dotted lines mark −30% and +20%. Patients without a post-baseline assessment are not shown. State the response criteria version (RECIST 1.1), the assessment interval and how many patients were excluded.

FAQ

What is a waterfall plot used for?

To show each patient’s best percentage change in tumour size, ranked, so the spread of response across a cohort is visible in one figure.

What do the −30% and +20% lines mean?

They mark RECIST 1.1 thresholds for partial response (30% decrease from baseline) and progressive disease (20% increase from the smallest sum, with at least 5 mm absolute increase).

Is a waterfall plot the same as a waterfall chart?

No. A waterfall chart in finance and Plotly shows cumulative steps; the clinical plot shows one ranked bar per subject. Related figures: forest plots, and general research figure practice.

Can I use a waterfall plot to claim a response rate?

Not on its own. Response needs the full RECIST assessment, including confirmation where the protocol requires it.

Sources

  • EORTC, RECIST 1.1 pages, recist.eortc.org, read 2026-11-08.
  • Eisenhauer EA et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 2009;45(2):228-247.
  • Waterfall plot of best percentage change and best overall response, PMC4646607, read 2026-11-08.
  • Plotly, Waterfall charts in Python, plotly.com, read 2026-11-08.

Follow CASRAI

Research-administration guidance, standards updates and independent tool reviews.

Ask CASRAI · free to try

Ask about Waterfall Plots: Reading and Making the Oncology Waterfall Plot

2 free questions a day, no account.

An AI assistant for research: funder rules, compliance, methods and publishing. It links the sources behind its answers and labels answers that come from a web search.

Answers draw on CASRAI's guides and dictionary plus the federal and funder documents we index: Federal Register, Grants.gov, Regulations.gov and UKRI.

Works on this site and inside Claude, Cursor and the AI tools you already use.

Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays open to read, with no account and no card.

Referenced across the research world

University of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logoUniversity of Cambridge logoColumbia University logoCrossref logoUniversity of Edinburgh logoHarvard University logoUniversity of Oxford logoPrinceton University logoStanford School of Medicine logoUniversity College London logoORCID logo
  • University of Cambridge logo
  • Columbia University logo
  • Crossref logo
  • University of Edinburgh logo
  • Harvard University logo
  • University of Oxford logo
  • Princeton University logo
  • Stanford School of Medicine logo
  • University College London logo
  • ORCID logo

View CASRAI adoption →