Graphics Reference
In-Depth Information
Figure
.
.
he let-hand panel shows the percentage of subjects in each treatment arm who
experienced an adverse event. he right-hand panel shows the relative risk and asymptotic
%
confidence intervals (Files: h
/stdt/code/aerelrisk.s, hh/jsm
/code/aerelrsk.jsm
.s)
his could be reversed, to put the largest increases at the bottom if preferred, or the
ordercould bedefined bythe actual riskin one of the treatment arms rather than the
relative risk.
Together, the two panels display four variables for each type of adverse event: the
incidences of A and B in the let-hand panel, and the relative risk and its confidence
interval in the right-hand panel. he two panels are linked, sharing the same row
names on the y-axis identifying the types of adverse events experienced.
he two treatments are differentiated by color, and by symbol in case the display
is viewed in black and white.
he relative risks are displayed on a log scale, as the asymptotic confidence in-
tervals are symmetrical on this scale. A vertical reference line is drawn to indicate
equality of risks, to facilitate appreciation of the size of each relative risk with refer-
ence to its confidence interval. he x-axis labeling may need adjustment to ensure
legibility: in this example, the
.
tick label was suppressed to avoid overwriting the
.
tick label.
Although this example displays the relative risk in the right-hand panel, any rele-
vant measure of treatment effect for proportions can be used. For example, in dose-
ranging studies, the odds ratio from a logistic regression model with dose as a co-
variate may be substituted for the relative risk.
hesizeofthedisplayedconfidenceintervalcouldbevaried.hereisclearlyamul-
tiplicity issue here, if there is interest in assessing the statistical significance of differ-
ences of the relative risk for so many types of events. However, the main aim of this
display is to highlight potential signals by providing an estimate of treatment effect
and the precision about that estimate. he criterion for inclusion of specific adverse
