PharmaSUG 2016 - Paper DG03 - Waterfall plot: two different approaches, one beautiful graph

Oct 1, 2026

Uploaded by OmBimo

This paper discusses two SAS-based approaches for presenting oncologic data using waterfall plots. One approach leverages the Statistical Graphics (SG) procedure SGPANEL within the ODS Graphics system available with SAS 9.2 and later. The other uses Graph Template Language (GTL) directly via PROC TEMPLATE and PROC SGRENDER. The paper compares the procedures for generating visually effective waterfall plots, starting with basic plots and progressively adding custom features for stratified complex trial information.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Abstract

Introduction

Snap Shots of Input Dataset

Generation of the Basic Plot

By PROC SGPANEL

By GTL

PharmaSUG 2016-Paper DG03 Waterfall plot: two different approaches, one beautiful graph Ting Ma , Pharmacyclics LLC, Sunnyvale, CA ABSTRACT This paper discusses two approaches to presenting oncologic data by use of a waterfall plot. One approach focuses on the statistical graphics (SG) family of procedure (SGPANEL) and the other utilizes the Graph Template Language (GTL) . Both approaches are within the ODS Graphics system available with SAS 9.2 and later release. This paper will compare the specific SAS procedures to generate the visually effective waterfall plot. Basic plots will be built asthe first step and then customized features will be added. These procedures should provide guidance to present stratified information from a complex clinical trial. INTRODUCTION Waterfall plots are gaining increasing popularity in oncology studies because it effectively demonstrates the best improvement in tumor size in relationship to the response to a drug seen in each subject. In such a plot, each patient in the trial is presented by a vertical bar on the plot and each bar represents the maximum change in the measurement of tumor. In trials with two arms, waterfall plots are often used to compare the outcome of active drug versus placebo. Various techniques are available in SAS to produce waterfall plots. Here we describe two techniques. One utilizes SAS/GRAPH Statistical Graphics (SG) procedures, PROC SGPANEL. This procedure supports a simple and concise syntax for creating an attractive graph via the power of GTL behind the scenes. This procedure involves minimum coding, yet generates consistent quality of graph for reporting and publication. The other approach uses the GTL syntax directly, via PROC TEMPLATE and PROC SGRENDER. This system is powerful and flexible, ideal for creating complicated custom graphs with some learning curve. However, after a template is created, subsequent execution of the procedure can be easily carried out. SNAP SHOTS OF INPUT DATASET To get the effect of waterfall, the variable to be plotted on the vertical bar (PCHG in this dataset) has to be sorted by the descending order so that the largest value will show up on the left most bar and the smallest value on the rightmost bar. The input dataset should contain one observation per subject (Figure1) . The dataset below is a fraction of a complete dataset used to generate figures in this paper. But the number of variables and the attributes of these variables need no change when we build individual plots in this paper. Figure 1 : A snap shot of the first 10 boservations of input dataset GENERATION OF THE BASIC PLOT Figure 2: Basic water plot generated by PROC SGPANEL and GTL • BY PROC SGPANEL ods graphics on / width=1280 height=720 ; ods rtf body='C:\Users \tma \Desktop\test \WaterFall \ f_a1 .rtf'; proc sgpanel data=final; *========================*; *== START OF SECTION A ==*; *========================*; panelby armn /columns=2 novarname; vbar order /response=pchg GROUP=bestrircn BARWIDTH=0 . 6 legendlabel='Best Response' name='resp' nooutline ; 1 colaxis display=none; rowaxis values=(-100 to 170 by 50) label="Maximum % Change from baseline"; title "Maximum Percent Reduction from Baseline for Spleen Volume by Best Response" font="Courier"; format armn armf . bestrircn bestf . ; 2 label bestrircn = "Best Response "; run; ods graphics off; *=======================*; *== END OF SECTION A ==*; *=======================*; 1 : The VBAR statement plots the variable ORDER. Subjects are grouped by variable BESTRIRCN. 2 : The two formats here convert the numeric information in the arm and response variable to characters. • BY GTL ods graphics on /imagename="f_b1" imagefmt=emf width=1280 height=720; ods rtf file="C: \Users \tma \Desktop\test \WaterFall \ f_b1 .rtf" proc template; define statgraph waterfall; 3 dynamic _ORDER _PCHG _ARM; dynamic _panelnumber_ ; begingraph / designwidth=960 de

FAQs

What problem do waterfall plots solve in oncology studies?

They demonstrate the best improvement in tumor size relative to a drug response for each subject, and in two-arm trials they help compare active drug outcomes versus placebo.

Which two SAS techniques does the paper compare for creating waterfall plots?

The paper compares PROC SGPANEL using SAS/GRAPH Statistical Graphics with a GTL-based template rendered via PROC TEMPLATE and PROC SGRENDER.

How does the paper ensure the waterfall ordering is visually correct?

It sorts the plotted variable (PCHG) in descending order so the largest value appears on the leftmost bar and the smallest value on the rightmost bar.

Oct 1, 2026

Uploaded by OmBimo