Results
We present the application of MI meta-analysis based on two cases
previously published in literature.
Case 1. Detection of endometrial cancer using
endovaginal ultrasonography (US).
For this case, we used data presented in Deeks 22,
originally published in Smith-Bindman et al 23. The
dataset is the result of a systematic review process on 35 papers
presenting the diagnostic performance of endovaginal US in the detection
of endometrial cancer. Evidence synthesis tables on test sensitivity and
specificity are provided in Deeks 22.
Figure 1 displays the meta-analytic summary plots based on US studies.
It includes the summary ROC curve, individual study estimate, and
summary point estimate of the “traditional” measures of performance of
endovaginal US in the detection of endometrial cancer. It is difficult
to interpret, how “good” the test is, and in particular how much
uncertainty the test reduced in each study where US was evaluated. For
example, a US test diagnosing endometrial cancer is 40% specific and
94% sensitive. The combination of these values is difficult to
interpret and may lead to inappropriate assessment (e.g. one could
favour the test due to its high sensitivity, ignoring its low
specificity). In terms of MI however, the test reduces diagnostic
uncertainty by 10%, which is marginal and thus the test is clearly not
very useful.
Figure 2 demonstrates meta-analysis of MI. We can clearly see that the
US results provided only 0.05 (0.04 to 0.07) bits of information
(recall, that the maximum amount information in the binary diagnostic
case is 1). Although this gives us an estimate about overall diagnostic
performance of US for diagnosis of endometrial cancer, what we really
want to know is the amount of diagnostic uncertainty the US can possibly
reduce (on scale 0 to 100%). This can be expressed by calculating RMI.
Figure 3a shows the performance of US expressed in terms of RMI. The
information presented is much clearer: a decision-maker has much better
understanding on how much diagnostic uncertainty was reduced in each
study. The pooled estimate of the reduction in diagnostic uncertainty is
13% for pre-test probability of disease 14%. That is, US can reduce
the uncertainty related to endometrial cancer by 13%. Figure 3b
presents the sample size of each study.