Graphical Approaches to the Analysis of Safety Data from Clinical Trials Ohad Amit, Richard M ...

39 Pages·2009·1.43 MB·English

1
Graphical Approaches to the Analysis of
Safety Data from Clinical Trials
Ohad Amit, Richard M. Heiberger, Peter W. Lane
Patient safety has always been a primary focus in the development of new
pharmaceutical products.
Safety issues in clinical trials are usually reported in tables.
Formal analysis of safety data is much less developed than for efficacy data.
Safety data provides an ideal opportunity to use graphical methods
•
Present concise summaries
•
Communicate main messages
•
Back up with tables as needed
Graphs can be used in
•
an exploratory setting to help identify emerging safety signals.
•
a confirmatory setting as a tool to elucidate known safety issues.
Graphical Approaches to the Analysis of Safety Data from Clinical Trials Richard M. Heiberger 2
I spent a research leave year at GSK where I joined a company-wide team
investigating graphical issues for display of clinical trial information.
We developed several graphical displays for routine safety data collected during
a clinical trial, covering a broad range of graphical techniques. Many of the
displays we devised are now included in the GSK software library. Our results,
coauthored with Ohad Amit and Peter W. Lane of GSK, appeared in 2008 in
Pharmaceutical Statistics.
The displays focus on key safety endpoints in clinical trials
•
the QT interval from electrocardiograms
•
laboratory measurements for detecting hepatotoxicity
•
adverse events of special interest.
We discuss in detail the statistical and graphical principles underlying the pro-
duction and interpretation of the displays.
Graphical Approaches to the Analysis of Safety Data from Clinical Trials Richard M. Heiberger 3
We illustrate eleven specific graphical designs, many of which display the data
along with statistics derived from them. Each will be discussed in detail.
•
Two are simple, comparing distributions with boxplots or cumulative plots.
4
3
Maximum LFT (ULN)2
1
0FFoorr ABSILATTOA, TAL CALKTCPLH is, a>n1d.5 A ULDLArNuT;g, w tAhhe e( NCAre=Sl i2nUA0iLcT9aNL)l iviCse otrhn Fecue Unrncp tpLioeenrv LeTele ivDsseAt r>lu Lo2gKf UPBNLH o(NNrm;=a4l0 R5)ange BILTOT
•
Five more display data and summaries over time, comparing information
from two groups in terms of distribution (with boxplots), cumulative inci-
dence, hazard, or simply means with error bars.
0.25 DDrruugg AB 0.008 DDrruugg AB
Cumulative Proportion with Event 0.00.050.100.150.20 Hazard Rate000...0000000.2460
DDrr_Suu_ggu _bAB_je_c_ts_ 24a_140t_87 R__is_k_230319 13905050 Days o12177n068 0Study 121665240 2 0020 A vaet rraisgkeDD d nrruuuorgg.in oABgf isnuteb0rjevca42tl31s00203240004031190060310800Da8y0s 21s88in00c1e0 r0a21n87d00o1m2i0za21t77io00n140216600160195301801183200
•
The other four are multi-panel displays: one-dimensional and two-dimensional
arrays of scatterplots, a trellis of individual profiles, and a paired dotplot
displaying risk together with relative risk.
Most Frequent On−Therapy A dverse Events Sorted by R elative Risk
Maximum (/ULN) 01234FFw0oohrre ABre1SIL UATALTOLN,2A TADD T,is rrLtuu hKtggh3e PABe HC ((UNN,C p==4aL24pn 00ieds95r0 )) A1L.Le5Av U1eTlLA, oNtLhfK; e2NP oCHrlmiBn3aaicsla eRl liaC4nnoegn (e0c/UerLnN 1L)eAvSe2Al iTs 23 ULN4;0 1BILT2OT3 4 ALKPH (/ULN)BILTOT (/ULN)ALAT (/ULN)0123012301230AS1AT 2(/ULDDN3rr)uugg AB ((0NN==A24001L95A))T 2(/ULN3) 0BIL1TOT2 (/UL3N) CUHGPRAPOSUENTRRRIRCI NERO SOAEEPRSBSIYPSWOR TATTPERIARRTHNIGUBHOAAAFDCHYCCERGFOTPATCTTYBECLHI ERDM THV DCRAIIBAHDDADENNRTYIVISEIEDEOOLAONTENINIMMIHFFKFSORSZI AANMCYUCUNSNRASNAOEEARAAYPZIOHDRMASRLNEKGOCITUEL CCTRILTRAEAVIRRAREUPNEIL JHNFSHMIETUTTPPPPELPLWIGUCNDANEHSEARLIIIIIMIIGGANARAAXSTTTNNOOSRHNUCEUEEESSAAIIIIIIIIIIIIIIGGHNNNNRNNYAAEAAAEAASSAAASSEAEXYL (( (((((((((((((((((((((((((((((((((((((())(((((((((())(((((((((((((())))))))))))))))(()))))))))))))))))))))))))))))))))))))))))))) ))
FFwoohrre ABreSIL UATLTON, TA isCL KCthPLeH iUs, pa1pn.5ed r U ALLLeNAv;eTl, othf eN oCrlminaicla Rl aCnognecern Level is 2 ULN; 0 10 20 30 .125 .512481632
Percent Relative Risk with 95% CI
TREATMENT A (N=216•) TREATMENT B (N=431) •
Graphical Approaches to the Analysis of Safety Data from Clinical Trials Richard M. Heiberger 4
Graphical displays for safety data
We illustrate each of our graphs using real data from a drug development pro-
gram: specifically, safety data from a large Phase III pivotal trial. This was a
two-arm randomized trial comparing a fixed dose of the test drug (Drug B be-
low) to placebo (Drug A). Following a four-week placebo run-in, patients were
randomized to receive either placebo or test drug for 24 weeks. The safety data
collected in this trial is typical of the type of routine safety data collection that
is part of almost all Phase II and III trials of pharmaceutical products.
While the graphs were developed in the context of a randomized parallel-group
trial we have provided commentary on their utility in other settings. This
real dataset has been used to demonstrate key concepts that we identified as
important in the evaluation of safety, but the displays were not tailored to draw
specific conclusions regarding the safety profile of the test drug.
Graphical Approaches to the Analysis of Safety Data from Clinical Trials Richard M. Heiberger 5
QT Interval
QTc measure of heart rhythm has become pivotal
Prolonged QTc interval is treated as a surrogate for serious arrhythmia
Change from baseline considered most relevant
Change associated with drug:
• →
> 30 msec clinical concern
• →
> 60 msec serious clinical concern
Entire distribution is of interest, as well as tail behaviour
We propose three graphs to help evaluate QTc.
Graphical Approaches to the Analysis of Safety Data from Clinical Trials Richard M. Heiberger 6
100
90
80
t
n
70
e
c
r
e 60
P

e
v 50
i
t
a
ul 40
m
u
30
C
20
10
Drug A (N=215)
Drug B (N=429)
0
-30 -20 -10 0 10 20 30 40 50 60 70 80 90
Change in QTc interval (msec)
Note: Increase <30msec ’Normal’, 30-60msec ’Concern’, >60msec ’High’
Empirical distribution function for maximum change in QTc
Figure 1:
Graphical Approaches to the Analysis of Safety Data from Clinical Trials Richard M. Heiberger 7
The y-variable is the maximum change for each patient over the 24-week treat-
ment period. Each point represents the percentage of patients on the y-axis
with a change in QTc less than or equal to the corresponding value on the
x-axis. Reference lines have been provided at 30 and 60 msec as well as at 0
msec to help the interpretation.
The distribution functions for the two treatment groups are drawn with fairly
thin lines to allow assessment of small differences at the upper end; different line-
styles are used as well as different colours for each treatment so that differences
are obvious when the graph is printed in black and white. As can easily be
seen in the graph, in this particular dataset there is little difference in how the
distributions behave at the upper tail, though the distribution for the test drug
is slightly less concentrated around the median than that for placebo. The
proportion of patients with a maximum change less than zero is about one-
ninth, as expected in a trial with eight visits after baseline and no effect on
QTc.
Graphical Approaches to the Analysis of Safety Data from Clinical Trials Richard M. Heiberger 8
Boxplot of change from baseline in QTc by time and treatment
Figure 2:
Graphical Approaches to the Analysis of Safety Data from Clinical Trials Richard M. Heiberger 9
Figure 2 is a boxplot that displays the distribution of the changes in QTc at
each time-point during treatment. The distribution of the maximum change
over the entire treatment period is displayed in a “margin” at the right-hand
side of the graph. Alternative or additional information can be displayed in
the margin, such as derived variables like the last-observation-carried-forward
(LOCF). In addition to providing a visual summary of variability and central
tendencies, this type of graphical display explicitly identifies the extreme values
in the distribution. Reference lines are drawn as before at 30 and 60 msec to
aid in the interpretation.
The concept of a graphical margin as used in Figure 2 is a powerful way to
add extra summary information in the context of a graph, just as a tabular
margin adds value to a table. It is of particular value in many graphs that show
effects over time to add a lower margin as here to show the number of subjects
involved in the summaries displayed at each time-point. Note that the numbers
of subjects may differ among the graphs we present, because of different patterns
of missing observations and different definitions of populations for the variables
involved.
Graphical Approaches to the Analysis of Safety Data from Clinical Trials Richard M. Heiberger 10
3
2
1
) 0
c
e
s
m
(
-1
e
g
n
a
h
C -2

n
a
e
M
-3
-4
-5
-6
0 2 4 8 12 16 20 24 LOCF
Drug A Drug B
Week
S__u_b_je_c_ts_ _a_t _V_is_i_t
Drug A 216 210206 199 191 184 176 169 164 214
Drug B 431 423384 362 337 315 311 299 293 429
Note: Vertical lines are 95% confidence limit ranges, LOCF is last observation carried forward
Mean (95% CI) change from baseline in QTc by time and treatment
Figure 3: