Content uploaded by Anil Kapoor
Author content
All content in this area was uploaded by Anil Kapoor on Apr 01, 2017
Content may be subject to copyright.
Page 1 of 4
© Journal of Xiangya Medicine. All rights reserved. J Xiangya Med 2017;2:17
jxym.amegroups.com
Prostate cancer (PC) is the most frequently diagnosed male
malignancy and the 2nd or 3rd leading cause of cancer deaths
in men in the developed countries. The disease progresses
from locally invasive carcinoma to metastatic prostate
cancer (mPC). While PC metastasizes to the liver and lung,
bone is the most frequent site of PC metastasis. Distant
metastasis likely marks the point of no return progress
towards the worst prognosis. Owning to the landmark
discovery that metastatic PC requires androgen receptor
(AR) signaling by Charles Huggins in 1941, androgen
deprivation therapy (ADT) remains the standard of care
for mPC patients. Although the treatment provides initial
benefits in the majority of patients with mPC, metastatic
castration-resistant prostate cancer (mCRPC) inevitably
arises. Prior to 2011, docetaxel was the only second line-
therapy (1), and prolonged median overall survival (OS) in
patients with mCRPC by 3 months. Since then, the second
generation anti-androgens (abiraterone and enzalutamide),
radium-223, cabazitaxel, and Sipuleucel-T have become
available in the clinic. Although these therapies are not
curative, they extend OS in patients with mCRPC (2). As
these drugs have different mechanisms of action, they could
be used in a variety of combinations either sequentially or
simultaneously to maximize benets to patients with mPC
or mCRPC. For example, ADT plus docetaxel is superior
to either alone (3) and is becoming the new standard of
care for patients with mPC with good performance status.
Clearly, improving our knowledge on the course of mCRPC
will contribute to the development of rational treatment
plans with the currently available medicines and thereby
improves patient management. In this regard, identifying
parameters to accurately predict survival of patients with
mCRPC is an area of active research; there are 113 and
40 published studies related to the topic of mCRPC and
prognostic biomarkers or prognostic models in PubMed
(https://www.ncbi.nlm.nih.gov/pubmed/advanced) up to
Dec 3, 2016. In order to yield a robust predictive model, it
will be essential for a team with combined expertise in clinic
and machine learning to analyze comprehensive sets of
clinical data.
This effort was recently reported (4). Using the Dialogue
for Reverse Engineering Assessments and Methods
(DREAM) challenge platform through Project Data
Sphere, Guinney and colleagues analyzed the impact of
more than 150 clinical baseline variables on OS of 2,336
mCRPC patients with an array of state-of-the-art machine
learning tools including an ensemble of penalized Cox
regression (ePCR) model, and formulated a powerful
model for predicting OS of mCRPC patients (4). More
importantly, the study was a part of an even much larger
Progress towards accurate prediction of overall survival in men
with metastatic castration-resistant prostate cancer
Wenjuan Mei1,2,3,4, Anil Kapoor2,5, Pierre Major6, Bobby Shayegan2,5, Damu Tang1,2,3
1Division of Nephrology, Department of Medicine, McMaster University, Hamilton, Ontario, Canada; 2Father Sean O’Sullivan Research
Institute, Hamilton, Ontario, Canada; 3The Hamilton Center for Kidney Research, St. Joseph’s Hospital, Hamilton, Ontario, Canada; 4Division
of Nephrology, The First Afliated Hospital of Nanchang University, Nanchang 330006, China; 5Department of Surgery, McMaster University,
Hamilton, Ontario, Canada; 6Division of Medical Oncology, Juravinski Cancer Center, Hamilton, Ontario, Canada
Correspondence to: Damu Tang. T3310, St. Joseph’s Hospital, 50 Charlton Ave East, Hamilton, Ontario L8N 4A6, Canada. Email: damut@mcmaster.ca.
Provenance: This is a Guest Commentary commissioned by Section Editor Longfei Liu, MD, PhD (Department of Urology, Xiangya Hospital,
Central South University, Changsha, China) and Assistant Editor Yongbao Wei, MD (Department of Urology, Fujian Provincial Hospital, the
teaching hospital of Fujian Medical University, Fuzhou, China).
Comment on: Guinney J, Wang T, Laajala TD, et al. Prediction of overall survival for patients with metastatic castration-resistant prostate cancer:
development of a prognostic model through a crowdsourced challenge with open clinical trial data. Lancet Oncol 2017;18:132-42.
Received: 13 December 2016; Accepted: 11 January 2017; Published: 13 February 2017.
doi: 10.21037/jxym.2017.02.10
View this article at: http://dx.doi.org/10.21037/jxym.2017.02.10
Commentary
Journal of Xiangya Medicine, 2017Page 2 of 4
© Journal of Xiangya Medicine. All rights reserved. J Xiangya Med 2017;2:17
jxym.amegroups.com
and comprehensive effort: sharing and analyzing the
clinical data by 163 experts consisting of 50 independent
teams worldwide (4). The clinical data used were compiled
from the control arms of five large randomized phase III
clinical trials through the effort of Project Data Sphere
(n=2,336, Table 1). Except ENTHUSE M1, patients in the
comparator arms from the rest of clinical trials were treated
with docetaxel (Table 1). More than 150 clinical variables
provided in individual clinical trials were centrally curated
by the organizers of the DREAM challenge to yield a core
table; of which data from three clinical trials (n=1,600) were
distributed to 50 teams for training analysis, and data from
ENTHUSE 33 and ENTHUSE M1 were used to score a
winning ePCR model and validate the model (Figure 1) (4).
The model performances well for stratifying low and
high risk patients in both ENTHUSE 33 (n=313) and
ENTHUSE M1 (n=226) cohorts (Figure 1) with predictive
accuracy determined by iAUC (integrated time-dependent
area under the curve) of respectively 0.791 and 0.768 (4).
Both risk groups have a significant difference in OS as
analyzed by the Kaplan-Meier method: hazard ratio (HR)
3.32, 95% confidence intervals (CI): 2.39–4.62, P<0.0001
for ENTHUSE 33 and HR 2.86, 95% CI: 2–4.12,
P<0.0001 for the control arm of ENTHUSE M1 (4).
Unlike patients in the control arms of other four clinical
trials, patients from the comparator arm of ENTHUSE M1
were receiving only placebo (Table 1) (4). Validation of the
ePCR model in the latter cohort confirmed its predictive
value as disease- rather than treatment-dependent.
The ePCR model not only won the challenge amount
50 competitive teams but also outperformed a prognostic
model recently published by Halabi and colleagues (5) in
stratifying low and high risk patients in both ENTHUSE
33 and ENTHUSE M1 (4). The Halabi model holds
superiority to similar models published prior to their
research (5); ePCR is thus likely the best prognostic
platform currently available in the public domain to predict
Table 1 Patientsa used by the DREAM challenge platform
CohortbNumber Treatment received Application LivereKidneye
ASCENT2 476 Doc + predncTraining Normal Normal
MAINSAIL 526 Doc + predn + placebo Training AdequatefAdequateg
VENICE 598 Doc + predn + placebo Training Adequate Adequate
ENTHUSE 33 470 Doc + placebo Test and scoredAdequatehAdequatei
ENTHUSE M1 266 Placebo Validation Adequate Adequate
a, mCRPC patients; b, all cohorts were from the control arms of the indicated phase III clinical trials; c, Doc/docetaxel and predn/prednisone
(prednisone or prednisolone for VENICE); d, a subgroup (n=157) was used for testing individual models and a second subgroup (n=313)
was for predicting OS (see Figure 1 for details); e, liver and kidney function at the time of patient recruitment; f, total bilirubin <1xULN (up
limit of normal), AST and/or ALT (alanine aminotransferase) <1.5×ULN; g, creatinine clearance >50 mL/minute; h, patients with ALT or AST
>2.5×ULN in the presence of liver metastasis were excluded; i, patients with 24 hours creatinine clearance <50 mL/min were excluded.
DREAM, Dialogue for Reverse Engineering Assessments and Methods.
ASCENT2
MAINSAIL
VENICE
n=1600
ENTHUSE 33
n=470
standardization of
clinical variables
ASCENT2
MAINSAIL
VENICE
*
training by
50 teams
1 ... ... 50
ENTHUSE 33 *
n=470
n=126
n=126
n=126
total n=157
round 1
round 2
round 3
3 rounds of submission and test
1 ... ... 50
n=333
prediction of survival
scoring the best model
split
ePCR
ENTHUSE M1
n=266
curation of
clinical
virables
ENTHUSE M1*
validation
Figure 1 Illustration of the research flow reported by Guinney
and colleagues. The control arms of the indicated clinical trials are
shown. *, indicates the sets of clinical variables generated through
central standardization. 1 … … 50 are for teams 1–50. A sub-group
(n=157) from ENTHUSE 33* was randomly (100 random split)
divided into three overlap subgroups (n=126) for teams to test their
models in the three rounds of submission and test; three (n=126)
subgroups collectively cover all 157 patients.
Journal of Xiangya Medicine, 2017 Page 3 of 4
© Journal of Xiangya Medicine. All rights reserved. J Xiangya Med 2017;2:17
jxym.amegroups.com
OS of mCRPC patients. However, this comes with no
surprise, as the DREAM challenge also outmatched the
Halabi group in terms of patient resources, team size (50
teams and 163 individuals) and collective expertise, as well as
state-of-the-art machine learning and statistical modelling
methods (4). While the Halabi model incorporates 22
clinical variables using the adaptive least absolute shrinkage
and selection operator (LASSO) penalized Cox regression,
ePCR analyzed 150 clinical variables using an advanced
Cox regression model: an ePCR (4,5). Nonetheless, the
Halabi model was able to stratify very similar groups of
low or high risk patients from ENTHUSE 33 compared
to the ePCR model (4), and is a much simpler model than
ePCR. Importantly, the Halabi model is available on line
as a reference tool for physicians to evaluate their mCRPC
patients (https://www.cancer.duke.edu/Nomogram/
firstlinechemotherapy.html.); the model thus has its
applications. Likewise, Guinney and colleagues will likely
make their model available in the same manner in the near
future. Nonetheless, it is clear that the ePCR model is more
comprehensive compared to the Halabi model. The study
thus has set a new standard of data sharing and analyzing in
model-building. This is particularly beneficial for clinical
trials not only because of the massive effort that has been
spent on these trials but also due to immediate benets that
it can bring to patients.
In addition to producing a robust prognostic model,
the data sharing effort also provides novel knowledge
in the prediction of OS for mCRPC patients. Aspartate
aminotransferase (AST) was indicated as an important
prognostic biomarker (4). However, most patients were
recruited to these clinical trials with normal or adequate
liver function (Table 1). The prognostic potential of
AST might be resulted from liver injury caused by liver
metastasis, as 1–14% of patients in all ve clinical trials used
in this study had liver metastasis (4). Hepatic metastasis is a
well-known cause for poor outcome. It will be interesting to
examine AST’s prognostic values in patients with only bone
metastasis or lack of liver metastasis. The same concern
also applies to kidney function. Patients were recruited also
with adequate renal function (Table 1); whether the renal
functional measurements (creatinine, creatinine clearance,
and calculated creatinine clearance) display prognostic
values should be further investigated. Will it be possible that
the prognostic values of these kidney criteria reflect their
interactions with other hematologic markers? As Guinney
and colleagues acknowledged that these interactions did not
reach a significant level (4), the network contributions to
the evaluation of OS in patients with mCRPC, as implied in
this report, should be explored in future.
Nonetheless, the concept of network of interaction
in predicting OS for mCRPC patients is intriguing. The
network involves biomarkers derived from the immune,
liver, and renal systems (4). While the prognostic
contributions of immunologic biomarkers are not surprising,
how renal functions impairment contributes to mCRPC
progression remains unclear. Will insufficient filtration
of some compounds contribute to mCRPC progression?
More attention has been paid to search for tumor-derived
prognostic factors. Guinney and colleagues raise an
interesting issue about the impact of the patient overall
health condition on the deadly progression of mCRPC.
Furthermore, work on the impact of the heterogeneities
in individual tumors versus individual host conditions in
mCRPC progression may reveal novel insights.
Answering the above questions will certainly requires
extensive and open data sharing and collaborative research
efforts. Furthermore, this type of big-data analyses needs to
incorporate molecular events. For example, prostate cancer
stem cells (PCSCs) are the driving force in PC evolution;
potential PCSC biomarkers should be considered. Will
the tumor samples in the clinical trials used in the study
by Guinney et al. be available for proling gene expression
changes using RNA sequencing and genomic alterations
through whole genome sequencing? If not, future studies
should be coordinated at the levels of RNA and DNA
alterations.
Acknowledgements
D Tang is supported by grants from Teresa Cascioli
Charitable Foundation Research Award in Women’s
Health, Canadian Breast Cancer Foundation, and Cancer
Research Society.
Footnote
Conicts of Interest: The authors have no conicts of interest
to declare.
References
1. Mukherji D, Omlin A, Pezaro C, et al. Metastatic
castration-resistant prostate cancer (CRPC): preclinical
and clinical evidence for the sequential use of novel
therapeutics. Cancer Metastasis Rev 2014;33:555-66.
Journal of Xiangya Medicine, 2017Page 4 of 4
© Journal of Xiangya Medicine. All rights reserved. J Xiangya Med 2017;2:17
jxym.amegroups.com
2. Agarwal N, Di Lorenzo G, Sonpavde G, et al. New agents
for prostate cancer. Ann Oncol 2014;25:1700-9.
3. Sweeney CJ, Chen YH, Carducci M, et al. Chemohormonal
Therapy in Metastatic Hormone-Sensitive Prostate Cancer.
N Engl J Med 2015;373:737-46.
4. Guinney J, Wang T, Laajala TD, et al. Prediction of
overall survival for patients with metastatic castration-
resistant prostate cancer: development of a prognostic
model through a crowdsourced challenge with open
clinical trial data. Lancet Oncol 2017;18:132-142.
5. Halabi S, Lin CY, Kelly WK, et al. Updated prognostic
model for predicting overall survival in rst-line
chemotherapy for patients with metastatic castration-
resistant prostate cancer. J Clin Oncol 2014;32:671-7.
doi: 10.21037/jxym.2017.02.10
Cite this article as: Mei W, Kapoor A, Major P, Shayegan B,
Tang D. Progress towards accurate prediction of overall survival
in men with metastatic castration-resistant prostate cancer. J
Xiangya Med 2017;2:17.
