Page 1
Floor Discussion 3:45 PM
Design of Statistical Graphics for Clinical Data Richard M. Heiberger, Temple University
3:25 PM
Graphical Analyses of Clinical Trial Safety Data Haijun Ma, Amgen Inc.; Kefei Zhou, Amgen Inc.; Hong A. Xia, Amgen Inc.; Matthew Austin, Amgen Inc.; George Li, Amgen Inc.; Michael O'Connell, InsightfulCorporation
3:05 PM
Statistical Graphics for the Analysis of Safety and Efficacy Data from Clinical Trials William Bushnell, GlaxoSmithKline
2:45 PM
Using Graphics To Discover and Explore Julia Wang, Johnson & Johnson PRD
2:25 PM
Statistical Graphics for Analysis and Reporting of Clinical Trials Michael O'Connell, Insightful Corporation
2:05 PM
Chair(s): Stephen Kaluzny, Insightful Corporation
Organizer(s): Michael O'Connell, Insightful Corporation
Section on Statistical Graphics, Section on Risk Analysis, WNAR
Statistical Graphics for Analysis of Drug Safety and Efficacy
Todays AgendaTodays Session
Page 2
www.splus.mathsoft.com 2Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 2
Todays Session
Michael O’Connell+ Statistical Graphics Principles – Functional Areas, Software
Julia Wang + Graphs for Exploratory Data Analysis
William Bushnell+ Graphs for Safety - Labs, AEs, patient profile, …
Haijun Ma + Graphs for Safety - Labs, AEs, patient profile, …
Rich Heiberger+ Graphs / Design for Vitals, Labs, AEs, …
Page 3
Copyright ©1999-2007Insightful Corporation. All Rights Reserved.
Statistical Graphics for Analysis and Reporting of Clinical Data
Michael O’Connell
July, 2007
Page 4
www.splus.mathsoft.com 4Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 4
My Outline
Business Problem+ Clinical Drug Development – Use of Graphics
Statistical Graphics + Graphics Principles, Elements, Types and Patterns
Graphics in Functional Areas+ Design, Review, Report
Software+ Graphics breakdown and mashup
Page 5
First inHuman
Proof of Concept Confirmation Marketing
NDA FilingLSSS ?
IND Filing
End of Phase 2 Mtg
Drug Development
Healthy VolunteerSafety ADME
Diseased PatientDosingM & S
Real WorldOutcomes
Epidemiology
Large PopulationSafety
Efficacy
Clinical Drug Development
Candidate Selection
Phase I Phase II Phase III Phase IVPre-Clinical
Compound EvaluationPharmaceutic /
Chemical Development
FDA
–Pha
rma I
nter
actio
nsPh
arm
a Mile
ston
es
Need to go faster, better/safer and cheaper !! [$1B/yr drug = $3M/day]
Page 6
Monitors
Data Mgt
Clinical
Statistics
Statistics
Programming
Publishing
Medical Writing
Clinical Trial Environment – Use Cases, Actors
Instream Unblinded
Statistics
Clinical
Management
ProtocolSAP
DataCleaning
Safety InstreamClinicalReview
CSRNDA
Labeling
JournalsScientific Meetings
Trial Design EDA / Review Report: Submission, Publication
Design
Page 7
www.splus.mathsoft.com 7Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 7
Statistical Graphics drive faster and better decisions
Speed the analysis and reporting process+ Faster registrational documents
Improve the quality of decisions+ Rapid and clear information from data
Increase likelihood of discerning safety and efficacy signals+ Comparative graphical analysis on all safety and efficacy data
Provide consistent form of communication across organization, cultures and functional areas
Improve interactions between sponsors and FDA+ FDA wants transparency and clear analysis / presentation
Statistical Graphics Value
Page 8
www.splus.mathsoft.com 8Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 8
Tufte’s Principles
+ Show all the data when possible
• Multivariate data and metadata+ Induce the viewer to think about the substance rather than the
graphic design – maximize the data-to-ink ratio+ Encourage the eye to compare different pieces of data – leverage
reader’s investment by showing multiple plots of same type
+ Reveal the data at several levels of detail, from a broad overview to the fine structure
+ Serve a clear purpose: description, exploration, tabulation+ Be closely integrated with the statistical and verbal descriptions of a
data set+ Use gray scale and color sparingly
Excellence in statistical graphics consists of complex ideas communicated with clarity, precision and efficiency
Statistical Graphics Principles
Page 9
www.splus.mathsoft.com 9Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 9
Variable Grouping+ Symbols, colors for groups – compare pieces of data
+ Treatment v Placebo
Trellis Plots and Conditioning + Panels for subjects, sub-groups – principle of small multiples
+ Adverse Event groups
Matrix Plots + Plots for individual variables together – multivariate data
+ Lab panels
Metadata – Margins and Axes+ Use of ticks, color-regions, margins around plots – multivariate metadata
+ Safety and Efficacy combination graphics
Brushing, Drilldown+ Data browsing – levels of detail
+ Population Lab -> Subject Lab -> AE -> demographics
Statistical Graphics – Functional Elements
Page 10
www.splus.mathsoft.com 10Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 10
Summary stats – univariate categorical + Dotplot+ Barchart [low data-to-ink ratio]+ Pie chart [poor perception]
Distributions – univariate continuous+ Boxplot+ Empirical CDF+ Density Plot+ Histogram
Relations and trends+ Line plot+ Scatter plot+ Scatter plot matrix+ Surface, contour and image (heatmap) plots
Basic Graph Types
+ Forest Plot
+ Kaplan Meier Plot
Page 11
www.splus.mathsoft.com 11Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 11
Figures convey trial designs and expected trial behavior+ Data analyst produces for clinician and management+ Multiple trial options are compared – operations (e.g. accrual) and
potential outcomes for variety of scenarios
Examples+ Protocol + SAP
Color and Format + Needs to be clear and easy to interpret and compare designs+ Statistical (e.g. # events) and calendar information need to be clearly
communicated
Illustrate today with survival trials
Trial Design Graphic
Page 12
Trial Design Graphic: Sequential HR DesignLine Plot | Group Design
0.4
0.6
0.8
1.0
0 100 200 300 400 500 600
Number of Events
haza
rd ra
tiodsnA11Fixed
Futility boundary(Pocock)
Efficacy boundary(O’Brien-Fleming)
Superiority of Treatment
Inferiority of Treatment
H0: HR > 1H1: HR < 0.7
Page 13
Trial Design Graphic: Sequential HR Design Line Plot | Group Design
Calendar Time
Num
ber o
f Sub
ject
s
0 10 20 30 40 50
020
040
060
0NAccrual= 700 Accrual Rate= 25.6 Accrual Time= 27.3 Study Time= 48.3
AccruedEventsAt Risk
HR= 0.7HR= 1
Accrual
Events
At Risk
(Months)
Interim Analyses
Page 14
5 15 25 35 45 55 65 75 85 95 105 115 125 135 145
0.0
0.1
0.2
0.3
Sample Size
Distribution of Efficacy Boundary Crossing
Z (E
ffici
ent S
core
)
-4
-2
0
2
4
6
8
10 Fixed-Sample Size
Design Parameters:Simulation runs = 100000α = 0.05Power = 0.8
Simulation Results:P(cross upper boundary) = 0.93712P(cross lower boundary) = 0.06288
Mean sample size = 54.62555Median sample size = 50
Superiority
Inferiority
Futility
Trial Design Graphic: Adaptive DesignDesign Line Plot | Matrix
Courtesy Kye Gilder, Biogen
Page 15
www.splus.mathsoft.com 15Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 15
Figure not necessarily self-contained+ Data analyst produces for detailed review with clinician+ Data analyst produces for the data analyst
Examples+ Clinical data review by/with clinician+ Viewing all data – data cleaning
+ Residual diagnostics - identification of functional form of relationship, assessing model assumptions
Color and Format + Few pre-defined rules + Just make it clear and easy to interpret
Exploratory / Review Graphic
Page 16
www.splus.mathsoft.com 16Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 16
Adverse Events - Interpretive GoalsWhich adverse events are elevated in treatment vs. placebo?
How rapid is onset in treatment vs. placebo?
Population level analysis is important
Labs - Interpretive GoalsWhich subjects have elevated (liver) labs?
Are there subjects with elevation on multiple labs?
Subject Level analysis is important
Adverse Events
Page 17
Adverse Events with Severity and Time
Data Courtesy Mat Soukup, FDA
Page 18
Adverse Event Review Graphic
ReviewLine Plot | Group, Trellis
Page 19
PT SOC TRT.A/214 TRT.B/428Abdominal distension Gastrointestinal disorder 5 0Abdominal pain Gastrointestinal disorder 11 11Abdominal pain lower Gastrointestinal disorder 2 0Abdominal pain upper Gastrointestinal disorder 5 2Acne Skin and subcutaneous tis 2 7 Ageusia Nervous system disorders 2 1 Alanine aminotransferase Investigations 8 22Alopecia Skin and subcutaneous tis 4 4Anaemia Blood and lymphatic syste 5 1Anorexia Metabolism and nutrition 9 10Anxiety Psychiatric disorders 4 3Arthralgia Musculoskeletal and conne 9 8Aspartate aminotransfera Investigations 32 3Asthenia General disorders and adm 7 2Back pain Musculoskeletal and conne 5 3Blood alkaline phosphata Investigations 3 1Blood creatinine increas Investigations 3 0Burning sensation Nervous system disorders 2 2Cancer pain Neoplasms benign, maligna 3 2Chest pain General disorders and adm 1 1Chills General disorders and adm 7 0... ... ... ... ... ...
Data Courtesy Mike Durante, GSK
Adverse Events Counts
Page 20
ReviewScatter / pRisk Plot | Brush, Drill
Adverse Event Review Graphic: B&B Model
B&B Model: θij ~ πi I{0} + (1–πi) N(μθi, σθi2)
Page 21
Insightful Clinical Review
Page 22
Insightful Clinical Review
Page 23
Insightful Clinical Review
Page 24
Insightful Clinical Review
Page 25
www.splus.mathsoft.com 25Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 25
Figure must be self-contained+ In-text figures should have an explanation in the caption
Figure must be documented+ Log file+ Time stamp+ Source data and output file references
Examples+ NDA Submission+ Scientific Publication+ Clinical Study Report and Presentation
Color / Format+ Must be interpretable when copied in black and white+ Compatible with Microsoft Word + Resizable
Report Graphic
Page 26
DepressionDyspnoea.exertionalHepatic.function.abn
Pain.in.jawPyrexia
Skin.hypopigmentatioAnaemia
ParaesthesiaHair.texture.abnorma
HypotrichosisPharyngolaryngeal.pa
CystitisDysphonia
HyperbilirubinaemiaDry.mouth
Visual.acuity.reducePain.in.extremity
HypokalaemiaUrinary.tract.infect
DysgeusiaHeadache
Aspartate.aminotransHypertension
Fatigue
0 5 10 15 20 25 30
Protocol: 38-316 Page 1 of 1Candidate: Oncology 38-316-001
AE Dotplot: Percent by PT
Run: M O'ConnellTime: March 21, 2007; 10.06pm ETData Source: H:/Oncology/P38316/Report1/AEterm.sas7bdatOutput: H:/Oncology/P38316/Report1/AEterm.pdf
Percent
Trt.ATrt.B
Adverse Event Report GraphicReportDot Plot | Group
Page 27
EpistaxisGastrooesophageal.reflux.disease
Pain.in.extremityMouth.ulceration
DysphoniaUrinary.tract.infection
Oedema.peripheralSkin.hypopigmentation
ConstipationParaesthesia
ChillsHeadache
Confusional.stateDysgeusia
Aspartate.aminotransferase.increasedPain.in.jaw
Skin.depigmentationHyperbilirubinaemia
Abdominal.distensionHypertension
Dry.mouthNausea
HypokalaemiaFatigue
2 4 6 8
Protocol: 38-316 Page 1 of 1Candidate: Oncology 38-316-001
Dotplot of Variable ImportanceRandom Forest Inside-Out Model
Run: M O'ConnellTime: March 27, 9.04pmData Source: H:/Oncology/P3816/Report/AEterm.sas7bdat
Variable Importance
Pre
ferre
d Te
rmAdverse Event Report Graphic
ReportDot Plot | Group
Trtij = f (PTj) + eij
i = 1, …, 642 subjects, j = 1, …, 111 AEs
Page 28
www.splus.mathsoft.com 28Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 28
Easy creation of (information-rich) statistical graphics for clinical data analysis
+ Dot plot, box plot, line plot, etc…
Easy re-purposing of graphs across studies and functional areas+ Graph standards – layering on (CDISC) data standards
Use of same graphs in exploratory and validated areas + Windows/UNIX exploratory and UNIX compliant
Consistent use of graphs across organization + Graphic language for effective communication
Key Use CasesSoftware Requirements
Page 29
Monitors
Data Mgt
Apply Patternto Data
Clinical
Statistics
Apply Patternto Bound Data
Statistics
Programming
Run Bound Pattern
for NDA
Publishing
Apply CSS Style to
Bound Pattern
Graphics System Mashup
Instream Unblinded
Design Review Report
PersistedPersistedGraph Types, Graph Types,
Patterns, Styles & Patterns, Styles & MetadataMetadata
(WEBDAV)(WEBDAV)
SS--PLUS PLUS Script FilesScript Files
(SSC)(SSC)
Clinical Graphics Store
Statistics
Specify GraphFrom Type Palette
Create/Add Graph Pattern to Store
Select Graph Pattern from Store
Add Type to Palette
Page 30
www.splus.mathsoft.com 30Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 30
Summary stats – univariate categorical + Dotplot+ Barchart [low data-to-ink ratio]+ Pie chart [poor perception]
Distributions – univariate continuous+ Boxplot+ Empirical CDF+ Density Plot+ Histogram
Relations and trends+ Line plot + Scatter plot+ Scatter plot matrix+ Surface, contour and image plots
Graph Types and Patterns
+ Forest Plot+ My AE Dotplot with Intervals
+ My Boxplot
+ Kaplan Meier Plot+ My Patient Profile Liver Plot+ My Liver Lab Shift Plot+ My Liver Lab Scatter Plot Matrix
Graph Types Graph Patterns
Page 31
Prostinol Phase III - Insightful Fuji SystemProstinol Phase III - Insightful Fuji System
File Edit View Insert Format Graph Data Window Help
New Graph...
Data Annotation History Metadata
Find Graphs... Saved GraphsImport Graphs...
Arial 12pt.
Saved Graphs
My most recent
All
LFT_ScatterPlotMatrix
Sort by:
Show:
AE_by_SOC_DotPlot
Base_2_End_LinePlot
Heart_Rate_LinePlot
Hemo_from_Base_LinePlot
Time_2_Response_KM
Discontinue_KM
Diaried_Change_BarPlot
Open Send to...
New Graph – Graph TypeNew Graph – Graph Type
Cancel
Line PlotKaplan MeierBar ChartScatter Plot Empirical CDFDot PlotHistogramPie ChartQQ PlotBox PlotForest Plot
My Recent All Graphs
Base Dot PlotDot Plot with IntervalAE Incidence Double Dot PlotAE Relative Risk Dot PlotAE Incidence and Relative Risk Dot Plot
Dot Plot with Interval
Data Source >
Graph Type
View by:
< Previous
Start with Graph Type
Start with Data Source
Selected Theme:CSR
Themes...
Dot Plot
Dot Plot with Interval(Preview)
Software Interface
Page 32
vanHeerden (n = 9)
Trazzera (n = 18)
Tennenberg (n = 15)
Ramsay (n = 27)
Maki (n = 24)
Heard (n = 27)
Hannan (n = 17)
George (n = 12)
Combined (n = 61)
Collins (n = 8)
Ciresi (n = 17)
Bach(b) (n = 8)
Bach(a) (n = NaN)
-4 -3 -2 -1 0
logOR
nam
es
Page 1 of 1
Software Interface
Page 33
DiarrhoeaDyspepsia
NauseaStomatitis
Gastrointestinal painVomiting
Haemorrhoidal haemorrhageLarge intestine perforation
ProctitisFlatulence
Abdominal painAbdominal pain lower
ToothacheGastric disorder
Gastrooesophageal reflux diseaseAbdominal pain upper
GlossitisMouth ulceration
ConstipationAbdominal distension
Dry mouth
-0.5 0.0 0.5 1.0
|||
|||
||||
|||||
|||||
|
log-10 Empirical RR Bayes Posterior Mean of Log Relative Risk with 99% BCI
Dot Plot Review
My AE Dot Plot with Interval Pattern
Page 34
0.0
0.1
0.2
0.3
0.4
-2 0 2 4
Placebo 0.25mg
0.5mg
0.0
0.1
0.2
0.3
0.4
-2 0 2 4
1.0mg
Page 1 of 1
x
Freq
uenc
yHistogram | Trellis Report
My Histogram for Internal Power Point Pattern
Page 35
0.00.10.20.30.4
-2 0 2 4
Placebo0.00.10.20.30.4
0.25mg0.00.10.20.30.4
0.5mg0.00.10.20.30.4
1.0mg
Page 1 of 1
x
Freq
uenc
y
Histogram | Trellis Report
My Histogram for Internal Power Point Pattern
Page 36
Combined Graphical and Tabular Reports
Treatment Minus Placebo Stain at Week 4
Body Plot Review
Page 37
Insightful Clinical Review
Page 38
Insightful Clinical Review
Page 39
Futures: Interactive Clinical Review on the iPhone
Page 40
Interactive Clinical Review on the iPhone
Page 41
Interactive Clinical Review on the iPhone
Page 42
Interactive Clinical Review on the iPhone
Page 43
www.splus.mathsoft.com 43Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 43
Point-click specification of statistical graphics (stats, programming)+ Easy to create informative statistical graphics (point-click)+ Comprehensive palette of graph types: dot plot, box plot, line plot, …
Create / save a graph pattern for re-use on bound or new data+ Instream clinical data review / cleaning (data refresh)+ Analysis of different trial/endpoint by different person/TA (CDISC helps)
System creation of script to be run in validated environment+ Oracle / SAS and submission reporting
Graph styling and presentation with bound data (publishing)+ Graph in to company power point style e.g. b&w <-> color+ Graph in to journal style + No change to scientific graph content
Create / add new graph type (stats)+ New statistical graphics added by statistics for use throughout
Permissioned re-purposing of complex statistical graphics
Key Use CasesKey Clinical Graphics Use Cases – Mashups
Page 44
Insightful Services Oriented Architecture
XML Graph
Definition
S-PLUS Enterprise Server
PersistedGraph Patterns,
Styles & Metadata(WEBDAV)
S-PLUS Script File
(SSC)
Clinical Data
SAS Oracle CDISC EDC Safety
Graph View File
(SPJ)
SPJWMFPDFLOG
Libraries XSL GOM Libraries
SAS / S-PLUSMacros
SAS Engine
TOM
Create/Edit GraphPublish PatternRetrieve Pattern
S-PLUS Production Job Submit
SAS Production Job Submit
Graph View - UI Hooks -
S-PLUS Web / .Net ClientWeb Job Client
ODSLOGWMFPDF
CSS
SPJWMFPDF
XML Protocol
Define.xml
Insightful Clinical Graphics
Page 45
www.splus.mathsoft.com 45Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 45
Statistical Graphics are at the core in the analysis and interpretation of clinical / safety data
+ Comparative analysis of treatment effects+ Exploratory – understand the data, cleaning and outliers+ Review – all components of population and patient level data+ Submission – clinical study reports + Presentation – scientific and marketing applications
Consistency in statistical graphics principles/standards is key+ S-PLUS TrellisTM and S-PLUS GraphletsTM provide consistency across
graph types and environments + Same graphs in reports and exploratory data browsing+ Graphical language across organizations and industry
This is widely recognized by industry and the FDA
Vision for Statistical Graphics
Page 46
www.splus.mathsoft.com 46Copyright ©1999-2007Insightful Corporation. All Rights Reserved. 46
Statistical Graphics are widely used to anchor the messaging from clinical studies
+ Group and Trellis / Metadata / Drill-down
Statistical Graphics are mashed up within pharma companies for rapid design, analysis, review, submission, publication and presentation
+ Using the clinical graphics taxonomy – types and patterns
A standard Statistical Graphics palette with rich APIs for repurposing is used across industry and the FDA
+ Simple interactions between industry and the FDA
In the not too distant future…… I see a world where
Safety and the critical path are making this reality
Vision for Statistical Graphics
Page 47
Michael O’Connell moconnell@insightful .com
Today’s presentation will be available on our website soonhttp://www.insightful.com/news_events
Insightful’s 10th User Conference in Atlantic CityInsightful Impact 2007
October 10-11, 2007, Tropicana Hotel, Atlantic Cityhttp://www.insightful.com/news_events/impact2007/default.asp
References and Contacts