EURURO Vol. 72 No. 3

380 476

1. Introduction
A significant effort is being placed on understanding
variations in the quality of surgical care as evidenced by
the growth of pay for performance programs and national
quality initiatives Efforts to measure quality have
expanded to most areas of surgical oncology with
specialty specific societies endorsing expert opinion
generated quality indicators (QI) measuring various
structural, process, and outcome elements of health care
delivery Despite this, a paucity of literature exists
capturing real-world data validating these indicators as
benchmarking tools that can accurately identify hospitals
providing poor care . Similarly, data capturing the
impact of hospital-level quality variation on patient
outcomes is significantly underreported for most QIs
. Consequently, concerns regarding the appropriate use
and choice of indicators have arisen given their impact on
financial and administrative resource consumption and
allocation as well as effect on patient autonomy and
physician reputation Hence, an urgent need to take
a robust data driven approach to validate putative QIs
is needed in order to prioritize the most valuable
measures.
To date, efforts to measure[9_TD$DIFF] the quality of renal cell
carcinoma (RCC)[10_TD$DIFF] surgical care remain in their infancy.
While putative expert generated RCC-specific QIs have
been proposed, no real-world data exists to validate these
metrics as benchmarking tools that can discriminate
provider performance in a manner that captures disparate
patient outcomes . This is in part due to the challenge
of adjusting for complex case-mix variation between
hospitals that must be completed in order to rigorously
benchmark performance . Comprehensive cancer
specific data initiatives, such as the National Cancer
Database in the USA, which capture granular patient and
tumor specific variables across large volumes of hospitals
have greatly facilitated case-mix adjusted quality bench-
marking assessments in surgical oncology, and as such
provide a platform to validate putative QIs in urologic
oncology
Due to the paucity of real-world data benchmarking
provider performance in RCC surgical care, our primary
objective was to determine whether nationwide variations
in quality exist on a hospital-level after adjusting for
differences in case-mix factors present within the National
Cancer Database (NCDB). Further, we investigated structur-
al elements (ie, hospital type, location, surgical volume)
associated with hospital-level quality. Lastly, we sought to
determine whether benchmarking hospital performance
using our developed case-mix adjusted QIs could discrimi-
nate provider performance in a manner that captures
disparate patient outcomes by assessing associations
between hospital-level quality and patient mortality. We
hypothesized that variations in RCC surgical quality exist,
with poor quality being associated with adverse patient
outcomes.
2. Materials and methods
2.1. Data
This cohort study utilized the NCDB, which prospectively collects
hospital-level data from Commission on Cancer Accredited Facilities in
the USA and Puerto Rico. The NCDB captures approximately 70% of newly
diagnosed cancer cases with over 30 million individual records
accumulated since its inception across over 1500 hospitals. Approval
by the University Health Network (Toronto, ON, USA) Research Ethics
Board was obtained.
2.2. QIs
Hospital-level performance was benchmarked according to five QIs.
Three process QIs were identified from a previously published modified
Delphi study including the proportion of patients with: (1) T1a
tumors undergoing partial nephrectomy (PN), (2) T1-2 tumors receiving
a minimally invasive (laparoscopic or robotic) approach for radical
nephrectomy (MIS), (3) a positive surgical margin following PN for T1
tumors (PM). Two outcome QIs: (1) length of hospital stay after radical
nephrectomy for T1-4 tumors (LOS), and (2) 30-d unplanned readmis-
sion proportion after radical nephrectomy for T1-4 tumors (RP), were
additionally chosen given their utility as quality benchmarking tools in
other realms of surgical oncology .
2.3. Study cohort
International Classification of Diseases for Oncology (third edition) and
site-specific surgery codes were employed to identify RCC patients who
underwent PN or radical nephrectomy between 2004 and 2014. Data
from earlier than 2004 was excluded due to incomplete patient
comorbidity information. For MIS, analysis was restricted to 2010 on-
ward as laparoscopy was not available prior. For the LOS and RP QIs,
patients with localized and metastatic disease were included whereas
metastatic patients were excluded for analysis of the MIS, PN, and PMR
QIs. Summaries of all inclusion and exclusion criteria, including
International Classification of Diseases and histology codes are available
in Supplementary Table 1.
2.4. Statistical analysis
Our statistical approaches closely followed those used in previous
analyses of NCDB data for quality comparisons in surgical oncology
. Interhospital variability in the QIs was investigated using
random intercept generalized linear models; PN, MIS, PM, and RP QIs
were modeled through logistic regression; LOS, transformed as natural
logarithm of 1 + LOS because of the skewed distribution of this QI, was
modeled through linear regression. We estimated an intraclass
correlation/between hospital variance proportion using the latent
variable method and calculated p-values for tests of the null of [34_TD$DIFF]no
between hospital variance component . QIs were adjusted for case-
mix using indirect standardization, where for each hospital we
calculated a standardized mortality ratio of[11_TD$DIFF] observed to expected
outcomes[12_TD$DIFF] The expected quality outcomes were calculated from
multivariable regression models (logistic for PN, MIS, PM, RP; linear for
the transformed LOS) fitted to the entire patient population without the
hospital-level random intercepts, given all relevant patient-level
demographic, comorbidity, disease progression, and tumor character-
istics recorded in NCDB, as listed in Supplementary Figures 1A–E. To
identify outlier hospitals, we used z-test statistics of the form Z = (O-E)/S,
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 3 7 9 – 3 8 6 380