EURURO Vol. 72 No. 4

519 656

Platinum Priority – Editorial
Referring to the article published on pp. 509–518 of this issue
Prediction is ‘‘Still’’ Difficult when it is About the Past
Farhad Kosari R. Jeffrey Karnes
a Center for Individualized Medicine, Mayo Clinic, Rochester, MN, USA; b Department of Molecular Medicine, Mayo Clinic, Rochester, MN, USA; c Department
of Urology, Mayo Clinic, Rochester, MN, USA
Our article title is a play on the words of physicist Niels
Bohr. In earnest, how does one design and validate an assay
on archived prostate cancer tissue to identify those with
metastatic potential? In this issue of European Urology,
Walker et al publish their findings for a 70-transcript
signature to identify cases with a metastatic biology. This
adds to the growing field of molecular modeling of
advanced prostate cancer, and validation is anxiously
awaited. Years ago, our group reported on a reverse
transcriptase–polymerase chain reaction assay of gene
expression for archived specimens, with an area under
the receiver operating characteristic (ROC) curve (AUC) of
approximately 0.8; the model was based on a case-control
design involving surgical patients with higher-risk features.
As featured in that article and based on a Cox model
comprising >10 000 surgical patients, the ability to assign
the risk of metastasis for all-comers using routine
clinicopathologic parameters alone was 0.8, but the AUC
drops to 0.69 if limited to higher-grade disease, and if
narrowed further to patients with extracapsular extension
and/or regional lymph node involvement at surgery, then
the AUC approaches the probability of a coin flip (0.5; data
not shown). The latter group certainly represents the
patients who are at risk of developing metastasis in the
future. There might also exist subgroups of differential gene
expression with prognostic capability based on TMPRSS2
fusion status, although the latter lacks independent ability
. Regardless, it was a case-control design that led to the
development of a 22-gene expression marker to predict the
risk of metastasis at 5 yr after surgery . This genomic
classifier is now known as Decipher and has been
independently validated in numerous cohorts involving
biopsy specimens and thousands of men, including a case-
cohort design in an at-risk population It has also been
shown that the assay provides independent prognostic
information over what is provided by the CAPRA-S score
. More recently, it was shown that Decipher also provides
an important 10-yr disease-specific mortality prediction for
an at-risk population Thus, do we need another assay?
Certainly there is room for improvement.
The current investigators used gap statistics to find
structures in the microarray data and then regression
modeling (partial least squares) to develop their model. In
addition, in a somewhat novel approach, an interim
validation was carried out by processing five sections from
the same tumor in a subset of discovery-1 samples and
replication of 20 cases in discovery-2 samples to ensure
robust expression and address potential tumor heteroge-
neity.
Nonetheless, there are some potential caveats for the
performance of this 70-signature assay. The goal of
developing ‘‘an assay that could identify metastatic-
subgroup tumours’’ may not be fully correct or justified.
As shown in Figure 1B, C1 (the metastatic subgroup) has
approximately 50% ‘‘primary + de novo metastatic’’ and
lymph node ‘‘metastatic disease’’. Focusing on C1, ‘‘the
metastatic subgroup’’ will be inherently problematic
because of specificity issues, and this may partly explain
the suboptimal specificity in the validation sets. In fact, a
study using Decipher for pairwise comparison of primary
and metastatic lymph node disease found only approxi-
mately 70% concordance between them .
The assay delivers a signature as a dichotomous result
rather than a continuous risk, and it might be difficult to
‘‘interpret’’ such a result since metastatic risk is rarely a
simple yes or no. Furthermore, does the assay add to the
CAPRA-S score for use in clinical decision-making? Even
though the Cox analysis suggests the assay is significant
ava ilable at
journal homepage:
DOI of original article: .
* Corresponding author. Department of Urology, Mayo Clinic Rochester, MN 55905, USA.
E-mail address: (R.J. Karnes).
0302-2838/# 2017 European Association of Urology. Published by Elsevier B.V. All rights reserved.