EURURO Vol. 73 No. 1

13 148

radiotherapy when compared to surgery had significantly
higher risk of PCa-specific mortality. Thus, while clinically
meaningful, BCR should not be used to compare the
oncologic efficacy of PCa treatments, and this review
focuses on survival outcomes.
3.1.1. Randomized survival data
Radical prostatectomy was the only treatment shown in a
randomized controlled trial to improve overall and cancer-
specific survival for patients with localized PCa when
compared to watchful waiting In the Scandinavian
Prostate Cancer Group Trial #4 (SPCG-4), Bill-Axelson et al
randomized 695 men with early clinically detected PCa
to radical prostatectomy or watchful waiting. In addition to
a survival benefit, surgery reduced the risk of metastasis
. These benefits were not confirmed in a similar study
(Prostate Cancer Intervention versus Observation Trial,
PIVOT) although this study was limited by the
inclusion of proportionallymoremenwith low-risk disease
and more men with significant comorbidities and shorter
follow-up (median 10 yr). In the SPCG-4 trial, the benefit of
surgery has continued to increase as ongoing follow-up has
accrued.
Two older randomized trials compared survival out-
comes following radical prostatectomy and radiotherapy.
The first was conducted by the Uro-Oncology research
group in the era before prostate-specific antigen (PSA)
screening and the second by the Japanese Study Group for
Locally Advanced Prostate Cancer more recently . Both
demonstrated improved outcomes in surgically treated
patients; however, owing to methodologic limitations
(including insufficient follow-up) and the evolution of
medical practice (including stage migration due to the
introduction of PSA screening), neither of these studies has
influenced current clinical practice.
Recently, the ProtecT trial reported survival outcomes
among 1643 patients randomized to active monitoring,
radical prostatectomy, and radiotherapy . The investi-
gators found no significant difference in their primary
outcome of PCa-specific mortality (p = 0.48), with eight
attributable deaths in the monitoring group, five in the
surgery group, and four in the radiotherapy group
. Overall mortality rates were also comparable
(p = 0.87). Limitations in applying these data to clinical
practice have previously been reported . Most
notably, there is a lack of statistical power, a fact recognized
years before manuscript publication , because of
significant overestimation of predicted mortality rates at
the time of the study design. In addition, there is over-
representation of patients with low-risk disease among the
study cohort . Because of these limitations, it is unlikely
that meaningful comparisons of mortality for patients
treated with surgery and radiotherapy will ever be made
from this cohort. Nonetheless, ProtecT identified a signifi-
cant reduction in both clinical progression and metastatic
disease among men receiving definitive therapy.
Among 89 patients with localized or locally advanced PCa
randomized to surgery or radiotherapy (external-
beam radiotherapy [EBRT] + brachytherapy boost + ADT),
Lennernas et al recently reported no difference in
overall or cancer-specific mortality, although the authors
correctly concluded that they were underpowered to assess
survival outcomes.
3.1.2. Observational survival data
Considering the limitations among available randomized
trials, a recent meta-analysis of observational studies
compared overall and PCa-specific mortality for patients
treated with surgery and radiotherapy . Utilizing pooled
results for 95 791 patients for the outcome of overall
mortality and 118 830 patients for PCa-specific mortality,
patients treated with radiotherapy had a significantly
higher risk of death (overall mortality, hazard ratio [HR]
1.63, 95% confidence interval [CI] 1.54–1.73; PCa-specific
mortality, HR 2.08, 95% CI 1.76–2.47). These findings were
robust to subgroup and sensitivity analyses, including PCa
risk categorization, study accrual period, radiotherapy
modality (EBRT or brachytherapy), duration of follow-up,
and geographic region of study . It is notable that a
survival benefit was found even among patients with low-
risk disease, probably reflecting a combination of the Will
Rogers phenomenon and residual confounding.
While observational data cannot account for unmea-
sured confounding in the manner of a randomized
controlled trial, as others have highlighted , the
studies included were deemed at low to moderate risk of
bias using the Newcastle-Ottawa Scale, a validated measure
recommended by the Cochrane Collaborative to evaluate
observational studies . By contrast, another meta-
analysis that downplayed differences in survival between
surgery and radiotherapy did not use a validated
measure for bias assessment. Initially, the authors
attempted to use the GRADE criteria Rather than
relying on this validated measure, they subsequently
constructed a subjective reliability scale based on their
‘‘perceptions about the relative importance of each factor’’
. The authors deemed single-institution studies to be of
higher ‘‘reliability’’ than multi-institutional reports, and
penalized studies reporting on populations >12 000 patients,
even though these studies have greater external validity.
Despite this, the authors demonstrated that radical prosta-
tectomy was associated with improved overall and cancer-
specific survival compared to radiotherapy.
There are many ways to account for selection biases,
principally confounding by indication, in observational
studies including regression techniques, propensity score
approaches (including matching and weighting), and
instrumental variable techniques. Many of the studies in
the meta-analysis included all identifiably relevant patient
and tumor characteristics in regression or propensity score–
matched analyses While these approaches account for
observed confounders, instrumental variable analyses may
also account for unmeasured confounding. Using such an
approach for patients with predicted life expectancy of 10
yr, Sun et al found improved survival among those
treated with surgery compared to radiotherapy (HR 0.66,
95% CI 0.56–0.79). While it has been shown that instru-
mental variable analyses provide less biased estimates of
E U R O P E A N U R O L O G Y 7 3 ( 2 0 1 8 ) 1 1 – 2 0 13