- Email: [email protected]
Redefining and Contextualizing the Hospital Volume-Outcome Relationship for Robot-Assisted Radical Prostatectomy: Implications for Centralization of Care
Author's Accepted Manuscript Re-Defining and Contextualizing the Hospital Volume-Outcome Relationship for Robotic-Assisted Radical Prostatectomy: Implications for Centralization of Care Boris Gershman , Sarah K. Meier , Molly M. Jeffery , Daniel M. Moreira , Matthew K. Tollefson , Simon P. Kim , R. Jeffrey Karnes , Nilay D. Shah
PII: DOI: Reference:
S0022-5347(17)30184-2 10.1016/j.juro.2017.01.067 JURO 14341
To appear in:
The Journal of Urology
Please cite this article as: Gershman B, Meier SK, Jeffery MM, Moreira DM, Tollefson MK, Kim SP, Karnes RJ, Shah ND, Re-Defining and Contextualizing the Hospital Volume-Outcome Relationship for Robotic-Assisted Radical Prostatectomy: Implications for Centralization of Care, The Journal of Urology® (2017), doi: 10.1016/j.juro.2017.01.067. DISCLAIMER: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our subscribers we are providing this early version of the article. The paper will be copy edited and typeset, and proof will be reviewed before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to The Journal pertain.
Embargo Policy All article content is under embargo until uncorrected proof of the article becomes available online. We will provide journalists and editors with full-text copies of the articles in question prior to the embargo date so that stories can be adequately researched and written. The standard embargo time is 12:01 AM ET on that date. Questions regarding embargo should be directed to [email protected].
ACCEPTED MANUSCRIPT
1
Re-Defining and Contextualizing the Hospital Volume-Outcome Relationship for Robotic-
2
Assisted Radical Prostatectomy: Implications for Centralization of Care
3
Boris Gershman1, Sarah K. Meier2, Molly M. Jeffery2, Daniel M. Moreira3, Matthew K.
5
Tollefson4, Simon P. Kim5, R. Jeffrey Karnes4, and Nilay D. Shah2, 6
RI PT
4
6 1
Division of Urology, Rhode Island Hospital and The Miriam Hospital, Providence, RI
8
2
Mayo Clinic Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery,
9
Mayo Clinic, Rochester, MN
M AN U
SC
7
3
Department of Urology, University of Illinois at Chicago, Chicago, IL
11
4
Department of Urology, Mayo Clinic, Rochester, Minnesota
12
5
Department of Urology, University Hospitals Case Medical Center, Case Western Reserve
13
University School of Medicine, Urology Institute, Cleveland, OH, USA
14
6
OptumLabs, Cambridge, MA, USA
15
18 19 20 21 22
Nilay D. Shah, PhD
EP
17
Correspondence:
Mayo Clinic
200 First Street SW
AC C
16
TE D
10
Rochester, Minnesota 55905 Phone: (507) 266-2743 Fax: (507) 284-1731 E-mail: [email protected]
23
1
ACCEPTED MANUSCRIPT
24
Keywords: robotic surgery; radical prostatectomy; hospital volume; outcomes; complications
25
Manuscript word count: 2490; Abstract word count: 250
AC C
EP
TE D
M AN U
SC
RI PT
26
2
ACCEPTED MANUSCRIPT
ABSTRACT
28
Purpose: Robotic-assisted radical prostatectomy (RARP) has undergone rapid dissemination,
29
driven in part by market forces, to become the most frequently employed surgical approach in
30
the management of prostate cancer. Accordingly, a critical analysis of its volume-outcome
31
relationship has important health policy implications. We therefore evaluated the association of
32
hospital RARP volume with perioperative outcomes, and examined the distribution of hospital
33
RARP volumes to contextualize the volume-outcome relationship.
SC
RI PT
27
34
Materials and Methods: We identified 140,671 adult men who underwent RARP from 2009 to
36
2011 in the Nationwide Inpatient Sample. The associations of hospital volume with
37
perioperative outcomes and total hospital costs were evaluated using multivariable logistic
38
regression and generalized linear models.
TE D
39
M AN U
35
Results: In 2011, 70% of hospitals averaged one RARP per week or less, accounting for 28% of
41
RARPs. Compared to patients treated at the lowest quartile hospitals, those treated at the highest
42
quartile hospitals had significantly lower rates of intraoperative complications (0.6% vs
43
1.4%;Ptrend<0.001), postoperative complications (4.8% vs 13.9%;Ptrend<0.001), perioperative
44
blood transfusion (1.5% vs 4.0%; Ptrend<0.001), prolonged hospitalization (4.3% vs 13.8%;
45
Ptrend<0.001), and mean total hospital costs ($12,647 vs $15,394;Ptrend<0.001). When modeled as
46
a non-linear continuous variable, increasing hospital volume was independently associated with
47
improved rates of each perioperative endpoint up to approximately 100 RARPs per year, beyond
48
which there appeared to be marginal improvement.
AC C
EP
40
3
ACCEPTED MANUSCRIPT
49
Conclusions: Increasing hospital RARP volume was associated with improved perioperative
50
outcomes, up to approximately 100 RARP per year, beyond which there appeared to be marginal
51
improvement. A substantial proportion of RARPs are performed at low-volume hospitals.
AC C
EP
TE D
M AN U
SC
RI PT
52
4
ACCEPTED MANUSCRIPT
53 54
INTRODUCTION Radical prostatectomy (RP) is the most frequently utilized treatment modality for prostate cancer and carries a tremendous public health impact1, 2. While open RP (ORP) has long been
56
the gold-standard surgical approach, dissemination of robotic technology has resulted in a shift to
57
robotic-assisted radical prostatectomy (RARP) in recent years. By 2009, RARP accounted for
58
nearly 60% of all RPs with an annual volume of 90,000 cases3, 4, and in 2013 it had grown to
59
85%5. However, there is ongoing debate regarding the comparative effectiveness of ORP and
60
RARP, and some authors have raised concerns regarding the impact of RARP on costs and
61
outcomes3, 6-8.
SC
M AN U
62
RI PT
55
Notably, the rapid adoption of RARP occurred under unique circumstances, driven in part by direct-to-consumer marketing of varying accuracy4, 9 and increased market competition
64
among hospitals10, 11. Acquisition of robotic technology has been associated with a 114%
65
increase in radical prostatectomy (RP) volume, which has resulted in a paradoxical centralization
66
of RP11. By 2009, approximately one third of hospitals had acquired robotic technology and
67
performed 85% of RP cases12.
While higher surgeon and hospital volume have been associated with improved
EP
68
TE D
63
oncologic, functional, and perioperative outcomes for ORP13-17, there is little data examining the
70
volume-outcome relationship for RARP. This is particularly important given the recent
71
dissemination of RARP, driven more by market forces rather than traditional factors. We
72
therefore conducted a population-based analysis of the association of hospital volume with
73
perioperative outcomes following RARP, and examined the contemporary distribution of
74
hospital RARP volume to contextualize the public health impact of this volume-outcome
75
relationship.
AC C
69
76
5
ACCEPTED MANUSCRIPT
77
MATERIALS AND METHODS
78
Study Population
79
After receiving exempt status from the Mayo Clinic Institutional Review Board (IRB), we identified adult men who underwent RARP from the Nationwide Inpatient Sample (NIS)
81
from 2009-2011. The NIS is the largest publicly available, all-payer inpatient database in the
82
US. States participating in the NIS cover 97% of the US population18.
We identified men aged ≥18 years who underwent RARP for prostate cancer using ICD-9
SC
83
RI PT
80
procedure code 60.5 (radical prostatectomy) with modifier code 17.42 (robotic assisted
85
procedure) and ICD-9 diagnosis code 185 (prostate cancer). The study was limited to January
86
2009 through December 2011 based on the introduction of the robotic modifier code in October
87
2008 and a redesign of the NIS in 2012, which changed the sampling methodology. Men were
88
excluded if they lacked the appropriate ICD-9 codes.
M AN U
84
90 91
TE D
89
Patient and Hospital Characteristics
We abstracted the following patient and hospital characteristics: age at surgery, year of surgery, race, median zip code income, health insurance status, length of stay (LOS), admission
93
type (elective/emergent), hospital teaching status, hospital region, hospital location (rural/urban),
94
and annual hospital volume. Secondary diagnosis codes were used to calculate the Elixhauser
95
comorbidity index for each patient19. Surgeon volume is not available in the NIS. Observations
96
with missing values were dropped except for race, which was assigned to an “unknown”
97
category.
AC C
EP
92
98 99
Outcomes
6
ACCEPTED MANUSCRIPT
100
Intraoperative complications were defined as surgical laceration of an organ, nerve, or blood vessel using ICD-9 code 998.213. Postoperative complications were identified using
102
administrative codes and categorized as cardiac, respiratory, bowel, hemorrhage, infectious,
103
wound, deep venous thrombosis (DVT)/pulmonary embolism (PE), genitourinary, and
104
miscellaneous as described previously8, 20. Prolonged length of stay (pLOS) was defined as a
105
hospitalization duration >90th percentile. Perioperative transfusion was identified using ICD-9
106
codes 99.02 and 99.0413. In-hospital mortality was not evaluated given the rarity of events.
107
Total hospital costs were estimated using hospital-specific cost-to-charge ratios and adjusted to
108
2011 dollars, using the Agency for Healthcare Research and Quality (AHRQ) Component Price
109
Index for hospital care.
M AN U
SC
RI PT
101
110 111
Statistical Analysis
Annual hospital volume was both categorized into volume quartiles and modeled as a
113
continuous non-linear variable using restricted cubic splines with 3 knots at the 10th, 50th, and
114
90th percentiles of volume21, 22. Restricted cubic splines allow modelling of a non-linear
115
relationship between a continuous covariate and outcome using piecewise cubic polynomial
116
functions joined at “knots,” with linear functions at the tails21, 22. Volume quartile cutoffs and
117
spline knots were determined using pooled data across all study years; hospitals were then
118
categorized for each year separately according to annual volume. Baseline characteristics were
119
summarized using medians/interquartile ranges (IQR) and frequency counts/percentages, and
120
compared across hospital volume quartiles using the adjusted Wald test for means and Pearson’s
121
chi square for frequency counts/percentages. Perioperative outcomes were compared across
122
hospital volume quartiles using a nonparametric test for trend23. Multivariable logistic
AC C
EP
TE D
112
7
ACCEPTED MANUSCRIPT
regression was performed to further evaluate the associations of hospital volume quartile with
124
perioperative outcomes. Total hospital costs were evaluated using a generalized linear model,
125
specifying the gamma family and log link. Models were adjusted for patient and hospital
126
features including age at surgery (18-54, 55-64, 65-69, 70-74, ≥75 years), year of surgery, race,
127
Elixhauser comorbidity index (0-1, 2-3, 4+), median zip code income, admission type, insurance
128
status, hospital teaching status, hospital region, and hospital location. Separate multivariable
129
logistic regression models were constructed modeling hospital volume as a non-linear continuous
130
variable using restricted cubic splines with 3 knots. Predicted rates of perioperative outcomes
131
and total hospital costs were estimated using marginal effects at the most prevalent covariate
132
categories (age 55-64 years, year 2011, white race, Elixhauser comorbidity index 0-1, median zip
133
code income ≥$63,000, elective surgery, teaching hospital status, south hospital region, urban
134
hospital location, and private insurance).
SC
M AN U
Analyses were performed using Stata version 14 (StataCorp, College Station, TX), and
TE D
135
RI PT
123
all tests were two-sided with p-values ≤0.05 considered significant. Except where noted, counts
137
of hospitals and patients are scaled to the national population size, using provided weights for
138
discharges or hospitals, as appropriate. In accordance with the NIS Data Use Agreement, results
139
with 10 or fewer observations were censored.
AC C
140
EP
136
8
ACCEPTED MANUSCRIPT
141 142
RESULTS A total of 140,671 patients were discharged after undergoing RARP from January 2009 to December 2011 at 2472 hospitals. The total volume of RARP did not significantly differ
144
across study years (p=0.55 by adjusted Wald test). Hospital volume quartiles were defined as
145
very low (≤12), low (13-30), medium (31-66), and high (67-820). Baseline patient
146
characteristics stratified by hospital volume quartile are presented in Table 1. Mean age at
147
surgery was 61.5 (SD 7.2) years and 68% of men were Caucasian. There were significant
148
differences across hospital volume quartile with regard to several patient and hospital
149
characteristics. Specifically, higher volume quartile hospitals were more likely to be teaching
150
(p<0.0001) and urban hospitals (p=0.001), while patients treated at higher-volume hospitals were
151
more likely to have private insurance (p=0.004) and fewer comorbidities (p=0.0001), with a
152
trend toward higher median zip code income (p=0.07). These differences suggest that
153
regionalization may be associated with health disparities.
SC
M AN U
TE D
154
RI PT
143
During the study period, there was evidence of de-centralization of RARP volume. The proportion of RARP cases performed at high-volume quartile hospitals decreased from 78% in
156
2009 to 63% in 2011, driven by increasing volume at low- and medium-volume quartile
157
hospitals (Figure 1; p<0.001). Moreover, there was an increase in the proportion of hospitals that
158
were lower volume – by 2011, 70% of hospitals averaged one RARP per week or less, up from
159
59% in 2009.
AC C
160
EP
155
To more accurately capture subtle temporal changes in the distribution of RARP, we
161
plotted the cumulative distribution of RARP across hospital volume as a continuous variable,
162
stratified by year (Figure 2). Here, too, there was evidence of a leftward shift from higher to
9
ACCEPTED MANUSCRIPT
163
lower volume hospitals over the study period. For example, the proportion of RARPs performed
164
at hospitals averaging ≤1 RARP per week increased from 15% in 2009 to 28% in 2011.
165
We conducted an exploratory analysis to examine how total RP volume may contextualize changes in RARP volume over time. Over the study period, there was no
167
statistically significant change in total RP volume (p=0.46 by adjusted Wald test). We then
168
estimated the proportion of RARPs performed within each quartile of total RP volume
169
(Supplementary Figure 1). During the study period, RARP appeared to shift from hospitals in
170
the top quartile of total RP volume to those in the second highest quartile of total RP volume,
171
with very little change in RARP volume among the lowest two quartiles of total RP volume.
SC
M AN U
172
RI PT
166
Next, we examined the association of hospital volume with perioperative outcomes using a traditional quartile-based categorization of volume (Table 2). Increasing hospital volume
174
quartile was significantly associated with improved perioperative outcomes and lower total
175
hospital costs. Specifically, compared to patients treated at very low-volume quartile hospitals,
176
those treated at high-volume quartile hospitals had lower rates of intraoperative complications
177
(0.6% vs 1.4%; Ptrend<0.001), postoperative complications (4.8% vs 13.9%; Ptrend<0.001),
178
perioperative blood transfusion (1.5% vs 4.0%; Ptrend<0.001), and prolonged hospitalization
179
(4.3% vs 13.8%; Ptrend<0.001), as well as lower mean total hospital costs ($12,647 vs $15,394;
180
Ptrend<0.001).
EP
AC C
181
TE D
173
On multivariable analysis (Table 3) adjusted for patient and hospital features, treatment at
182
a high-volume quartile hospital remained independently associated with a significantly reduced
183
risk of intraoperative complications (OR 0.44; p=0.03), postoperative complications (OR 0.32;
184
p<0.001), perioperative blood transfusion (OR 0.49; p=0.003), prolonged hospitalization (OR
185
0.28; p<0.001), and lower total hospital costs (-$2344; p<0.001) compared to very low-volume
10
ACCEPTED MANUSCRIPT
186
hospitals. Moreover, there was a statistically significant overall trend toward improved
187
perioperative outcomes with increasing hospital volume quartile (p≤0.01 for each).
188
To assess for non-linear relationships between hospital volume and outcomes, we modeled hospital volume as a continuous non-linear variable using restricted cubic splines
190
(Supplementary Table 1). Predicted rates of perioperative outcomes and total hospital costs were
191
then estimated, adjusted for patient and hospital features using marginal effects at the most
192
prevalent covariate levels. The resulting relationships of hospital volume with perioperative
193
outcomes and total hospital costs are illustrated in Figure 3. Notably, there was a relatively
194
consistent relationship wherein perioperative outcomes improved until approximately 100
195
RARPs per year, with marginal improvement at higher volumes. However, even in 2011, 54%
196
of RARPs were performed at hospitals below this annual volume.
M AN U
SC
RI PT
189
AC C
EP
TE D
197
11
ACCEPTED MANUSCRIPT
198 199
DISCUSSION In this study, we examined the association of hospital RARP volume with perioperative outcomes and total hospital costs, modeling hospital volume using both a conventional, quartile-
201
based categorization and as a non-linear continuous variable. Increasing hospital volume
202
quartile was associated with improved perioperative outcomes – that is, significantly lower rates
203
of perioperative complications, blood transfusion, and prolonged hospitalization, even when
204
adjusting for patient and hospital features. Notably, a more accurate illustration of the hospital
205
volume-outcome relationship emerged when hospital volume was modeled as a non-linear
206
continuous variable. Specifically, although perioperative outcomes improved with increasing
207
hospital volume, the association was most pronounced until approximately 100 RARPs per year,
208
beyond which the marginal benefit appeared to diminish.
M AN U
SC
RI PT
200
In order to contextualize the described volume-outcome relationship from a health policy
210
perspective, we also examined the cumulative distribution of hospital RARP volume. In contrast
211
to prior reports that noted a centralization of total RP volume12, 24, driven in part by year-over-
212
year volume increases following acquisition of robotic technology11, we observed a de-
213
centralization of RARP volume from 2009 to 2011. During that time period, the proportion of
214
hospitals averaging ≤1 RARP per week increased from 59% to 70%, with the corresponding
215
proportion of RARPs performed at such hospitals increasing from 15% to 28%. This may reflect
216
continued dissemination of robotic technology with a shift toward lower volume facilities. More
217
importantly, a substantial proportion of RARPs are performed at facilities that lie along the
218
“steep” portion of the volume-outcome curves, whereby improvements in perioperative
219
outcomes may be expected with increases in volume. This has important policy implications,
220
particularly given that acquisition of robotic technology has been driven more by direct-to-
AC C
EP
TE D
209
12
ACCEPTED MANUSCRIPT
221
consumer marketing4, 9 and increased market competition among hospitals10, 11 than traditional
222
technologic adoption based on clinical outcomes.
223
There exists an extensive body of literature supporting improved oncologic, functional, and perioperative outcomes for ORP performed at high-volume hospitals or by high-volume
225
surgeons17. However, there is little data on the volume-outcome relationship specifically for
226
RARP. Prior to October 2008, when the robotic-modifier code was introduced, it was not
227
possible to distinguish laparoscopic prostatectomy from RARP, and accordingly, few studies
228
have examined this question. Yu et al25 provided one of the first such reports in examining the
229
volume-outcome relationship for RARP using data from the last quarter of 2008 in the NIS. The
230
authors noted reduced complication rates, shorter LOS, and lower median costs, although the
231
study may be subject to bias related to analysis of data immediately after introduction of the new
232
robotic modifier code. Sammon et al26 compared perioperative outcomes and costs for men
233
undergoing ORP and RARP in 2009 using the NIS. The authors noted superior outcomes for
234
RARP overall and for high-volume hospitals regardless of surgical technique. Our study
235
provides further support for the hospital volume-perioperative outcome relationship for RARP.
236
Moreover, examination of volume as a non-linear continuous variable provides important
237
insights into the true underlying volume-outcome relationship.
SC
M AN U
TE D
EP
This is also the first study, to our knowledge, to examine the cumulative distribution of
AC C
238
RI PT
224
239
hospital RARP volume. We believe this is critical to contextualizing the population-level impact
240
of the described volume-outcome relationship. In the present study, we observed de-
241
centralization of RARP from 2009-2011, with a marked shift toward lower volume hospitals
242
with anticipated worse perioperative outcomes. A recent study27 also noted a decrease in the
13
ACCEPTED MANUSCRIPT
243
median hospital RARP caseload, suggesting a shift from high-volume to low-volume hospitals
244
with an associated increase in complication rates from 2009-2011.
245
Centralization of major surgery has been a debated topic and has potential disadvantages such as increased travel distance12, which must be weighed against the potential benefits in terms
247
of outcomes. As such, US health policy has not mandated centralization in any form for urologic
248
surgery28. However, lessons can be learned from the United Kingdom, where in 2002 the
249
National Health Services (NHS) recommended that RP only be performed in hospitals with
250
annual volumes of at least 50 cases, effectively mandating centralization of cancer services29.
251
Although such a wide-scale reorganization posed substantial logistical challenges and
252
necessitated initiation of quality-assurance programs, recent data suggest overall improvements
253
in quality of care30.
SC
M AN U
254
RI PT
246
This study has several limitations. It is a population-based analysis and subject to the potential biases of claims-based ascertainment, although an extensive body of literature supports
256
the methodology applied in this report. In addition, due to limitations of the inpatient dataset, we
257
were unable to examine post-discharge adverse events, 90-day attributable costs, oncologic
258
outcomes, or functional outcomes. Furthermore, due to a change in NIS sampling methodology
259
in 2012, we were unable to examine trends over a longer time period. Also, total RP volume
260
may affect the association of RARP volume with perioperative outcomes and deserves further
261
investigation. Lastly, we were unable to adjust for surgeon volume and disease characteristics,
262
such as tumor grade or stage, which are not captured in the NIS.
EP
AC C
263
TE D
255
14
ACCEPTED MANUSCRIPT
264 265
CONCLUSIONS We observed improved perioperative outcomes with increasing hospital volume up to approximately 100 RARPs per year, beyond which there appeared to be diminishing marginal
267
returns. Notably, there has been a recent de-centralization of hospital RARP volume, with nearly
268
one third of cases in 2011 performed at hospitals averaging ≤1 RARP per week. While further
269
studies are necessary to identify additional determinants of perioperative outcomes and
270
hospitalization costs, these results have important implications for health policy.
SC
RI PT
266
AC C
EP
TE D
M AN U
271
15
ACCEPTED MANUSCRIPT
272
REFERENCES
273
1.
274
primary treatment of localized prostate cancer. J Clin Oncol, 28: 1117, 2010
275
2.
276
Natl Compr Canc Netw, 12: 686, 2014
277
3.
278
assisted and open radical prostatectomy in the postdissemination era. J Clin Oncol, 32: 1419,
279
2014
280
4.
281
promotion of robotic prostatectomy exhibits varying quality of information. Health Aff
282
(Millwood), 31: 760, 2012
283
5.
284
Prostatectomy: A Contemporary Analysis of an All-payer Discharge Database. Eur Urol, 70:
285
837, 2016
286
6.
287
laparoscopic versus open retropubic radical prostatectomy among a nationwide random sample
288
of medicare-age men. J Clin Oncol, 30: 513, 2012
289
7.
290
open radical prostatectomy. Jama, 302: 1557, 2009
291
8.
292
prostatectomy attributable to robotic surgery. Eur Urol, 64: 11, 2013
293
9.
294
information is available? Urology, 77: 941, 2011
Cooperberg, M. R., Broering, J. M., Carroll, P. R.: Time trends and local variation in
RI PT
Mohler, J. L., Kantoff, P. W., Armstrong, A. J. et al.: Prostate cancer, version 2.2014. J
SC
Gandaglia, G., Sammon, J. D., Chang, S. L. et al.: Comparative effectiveness of robot-
M AN U
Mirkin, J. N., Lowrance, W. T., Feifer, A. H. et al.: Direct-to-consumer Internet
TE D
Leow, J. J., Chang, S. L., Meyer, C. P. et al.: Robot-assisted Versus Open Radical
EP
Barry, M. J., Gallagher, P. M., Skinner, J. S. et al.: Adverse effects of robotic-assisted
AC C
Hu, J. C., Gu, X., Lipsitz, S. R. et al.: Comparative effectiveness of minimally invasive vs
Kim, S. P., Shah, N. D., Karnes, R. J. et al.: Hospitalization costs for radical
Steinberg, P. L., Ghavamian, R.: Searching robotic prostatectomy online: what
16
ACCEPTED MANUSCRIPT
295
10.
Barbash, G. I., Friedman, B., Glied, S. A. et al.: Factors associated with adoption of
296
robotic surgical technology in US hospitals and relationship to radical prostatectomy procedure
297
volume. Ann Surg, 259: 1, 2014
298
11.
299
technology and hospital prostatectomy volumes: increasing market share through the adoption of
300
technology. Cancer, 118: 371, 2012
301
12.
302
centralization, robotics, and access to urologic cancer care. Cancer, 118: 54, 2012
303
13.
304
contemporary population-based analysis. Urology, 79: 332, 2012
305
14.
306
prostatectomy. N Engl J Med, 346: 1138, 2002
307
15.
308
cancer control after radical prostatectomy. Journal of the National Cancer Institute, 99: 1171,
309
2007
310
16.
311
radical prostatectomy as markers of surgical quality: analysis of heterogeneity between surgeons
312
at a single cancer center. Eur Urol, 59: 317, 2011
313
17.
314
outcome relationship for radical prostatectomy. Eur Urol, 64: 786, 2013
315
18.
316
19.
317
administrative data. Med Care, 36: 8, 1998
RI PT
Neuner, J. M., See, W. A., Pezzin, L. E. et al.: The association of robotic surgical
SC
Stitzenberg, K. B., Wong, Y. N., Nielsen, M. E. et al.: Trends in radical prostatectomy:
M AN U
Schmitges, J., Sun, M., Abdollah, F. et al.: Blood transfusions in radical prostatectomy: a
Begg, C. B., Riedel, E. R., Bach, P. B. et al.: Variations in morbidity after radical
TE D
Vickers, A. J., Bianco, F. J., Serio, A. M. et al.: The surgical learning curve for prostate
AC C
EP
Vickers, A., Savage, C., Bianco, F. et al.: Cancer control and functional outcomes after
Trinh, Q. D., Bjartell, A., Freedland, S. J. et al.: A systematic review of the volume-
McCannon, J., Berwick, D. M.: A new frontier in patient safety. JAMA, 305: 2221, 2011 Elixhauser, A., Steiner, C., Harris, D. R. et al.: Comorbidity measures for use with
17
ACCEPTED MANUSCRIPT
318
20.
Hu, J. C., Wang, Q., Pashos, C. L. et al.: Utilization and outcomes of minimally invasive
319
radical prostatectomy. J Clin Oncol, 26: 2278, 2008
320
21.
321
functions in public health research. Stat Med, 29: 1037, 2010
322
22.
323
Logistic Regression, and Survival Analysis: New York: Spriner, 2001
324
23.
Cuzick, J.: A Wilcoxon-type test for trend. Stat Med, 4: 87, 1985
325
24.
Anderson, C. B., Penson, D. F., Ni, S. et al.: Centralization of radical prostatectomy in
326
the United States. J Urol, 189: 500, 2013
327
25.
328
outcomes of robot-assisted laparoscopic radical prostatectomy. J Urol, 187: 1632, 2012
329
26.
330
prostatectomy: the differential effect of regionalization, procedure volume and operative
331
approach. J Urol, 189: 1289, 2013
332
27.
333
Diffusion: Complications Following Robot-Assisted Radical Prostatectomy. BJU Int, 2015
334
28.
335
considerations. J Am Coll Surg, 190: 341, 2000
336
29.
337
improving outcomes in urological cancers, 2002
338
30.
339
Prostate Cancer Surgery During Reorganisation of Cancer Services. Eur Urol, 2015
RI PT
Desquilbet, L., Mariotti, F.: Dose-response analyses using restricted cubic spline
M AN U
SC
Harrell, F. E., Jr.: Regression Modeling Strategies: With Applications to Linear Models,
Yu, H. Y., Hevelone, N. D., Lipsitz, S. R. et al.: Hospital volume, utilization, costs and
TE D
Sammon, J. D., Karakiewicz, P. I., Sun, M. et al.: Robot-assisted versus open radical
EP
Sammon, J. D., Abdollah, F., Klett, D. E. et al.: The Diminishing Returns of Robotic
AC C
Birkmeyer, J. D.: Should we regionalize major surgery? Potential benefits and policy
National Institute for Clinical Excellence (NICE). Guidance on cancer services:
Cathcart, P., Sridhara, A., Ramachandran, N. et al.: Achieving Quality Assurance of
340
18
ACCEPTED MANUSCRIPT
341
ACKNOWLEDGEMENTS
342
None.
AC C
EP
TE D
M AN U
SC
RI PT
343
19
ACCEPTED MANUSCRIPT
344
CONFLICT OF INTEREST
345
The authors have no conflicts of interest to disclose.
AC C
EP
TE D
M AN U
SC
RI PT
346
20
ACCEPTED MANUSCRIPT
Table 1: Baseline characteristics of patients who underwent RARP from January 2009 through December 2011 at 2472 hospitals (N=140,671). Numbers represent mean (SD) or %.
No. patients2 No. hospitals2 Annual hospital volume, median (range) Age at surgery, mean (yrs), [SD] Year of Surgery (%) 2009 2010 2011 Race (%) Caucasian African-American Hispanic Other Unknown Elixhauser Comorbidity Index (%) 0-1 2-3 4+ Median Zip Code Income (%) $1-38999 $39000-47999 $48000-62999 $63000+ Admission Type (%) Elective Non-Elective Hospital Teaching Status (%) Teaching Non-teaching Hospital Region (%)
140,671 2,472
Volume Quartile Medium
12,407 581 20 (13, 30)
28,457 618 44 (31, 66)
61.5 [7.2]
61.6 [7.1]
61.6 [7.3]
33.6 30.3 36.1
31.6 33.8 34.5
16.7 44.6 38.7
68.2 9.6 4.8 5.4 11.9
64.1 13.0 10.4 6.3 6.1
p-value1 High
95,873 617 100 (67, 820)
SC
Low
(x, 820)
Very Low 3,934 655 x (x,12)
RI PT
Characteristic
Total
-
61.5 [7.3]
0.943
24.9 30.7 44.4
38.4 28.2 33.4
0.03
64.8 12.3 5.4 3.2 14.3
67.9 9.8 5.1 4.5 12.7
68.9 9.1 4.4 5.9 11.6
69.5 26.8 3.7
70.3 26.5 3.2
70.2 27.4 2.4
74.0 24.0 2.0
0.0001
18.0 22.4 26.7 32.8
22.9 26.4 26.1 24.6
23.3 23.9 26.6 26.1
16.1 21.5 27.1 35.3
17.7 22.4 26.7 33.3
0.07
95.9 4.1
88.5 11.5
94.3 5.8
97.9 2.1
95.9 4.1
0.03
70.5 29.5
37.9 62.1
53.3 46.7
56.9 43.1
78.2 21.8
<0.0001
TE D
M AN U
61.5 [7.1]
AC C
EP
72.8 25.0 2.2
0.38
1
ACCEPTED MANUSCRIPT
23.1 16.2 40.7 20.0
17.6 31.1 36.9 14.4
14.0 25.6 34.8 25.5
20.2 23.0 32.4 24.4
0.59
2.1 97.9
7.9 92.1
6.9 93.1
3.7 96.3
0.7 99.3
0.001
31.7 1.8 63.0 3.5
34.7 4.1 53.9 7.3
35.0 1.8 59.3 3.8
31.1 1.8 63.3 3.8
31.4 1.7 63.7 3.2
0.004
RI PT
18.8 24.0 33.5 23.6
SC
Northeast Midwest South West Hospital Location (%) Rural Urban Insurance Category (%) Medicare Medicaid Private Other
Pearson’s chi-square test
2
Weighted figures given. The unweighted hospital count was 507; the unweighted discharge
count was 28,454 across the three years. 3
Adjusted Wald test
M AN U
1
AC C
EP
TE D
x=not depicted due to cell suppression policy
2
ACCEPTED MANUSCRIPT
Table 2: Perioperative outcomes stratified by hospital volume quartile.
1737 100 323 X 48 254 1262 514 1542
13.9 X 3.7
Ptrend
RI PT
High
1.3
1.1
0.6
8.8 0.5 2.1
8.0 0.6 1.4
4.8 0.3 0.8
0.2 1.1 6.0 2.2 6.6
0.2 0.8 3.4 1.5 4.3
<0.001 0.03 <0.001 X 0.20 0.20 <0.001 <0.001 <0.001
$13,674
$12,647
<0.001
X X 1.4 10.8 4.0 13.8
0.8 6.6 2.5 8.8
$15,394
$13,787
<0.001
SC
222
Volume Quartile Low Medium
Very Low 1.4
M AN U
Intraoperative complication (%) Postoperative complication (%) Any Cardiac Respiratory Vascular Wound/Bleeding Genitourinary Miscellaneous Blood transfusion (%) Prolonged hospitalization (%) Total costs (mean)
# of events*
TE D
Outcome
Unweighted count of events
AC C
*
EP
X=Not depicted due to cell suppression policy
3
ACCEPTED MANUSCRIPT
Table 3. Multivariable associations of hospital volume quartile with perioperative outcomes and adjusted total hospital costs. Model
teaching status, hospital region, hospital location, and insurance status.
ref
1.00
ref
Blood Transfusion
1.00
ref
Prolonged Length of Stay
1.00
ref
–
–
Total Costs2
SC
1.00
OR (95% CI) 1.07 (0.50, 2.30) 0.59 (0.44,0.80) 0.74 (0.43,1.27) 0.61 (0.43,0.87) -1553 (-2956, -151)
Volume Quartile Medium High p-value1 OR pOR pp-value (95% CI) value (95% CI) value 0.91 0.44 0.86 0.81 0.03 <0.0001 (0.43, 1.9) (0.21, 0.94) 0.54 0.32 0.001 <0.001 <0.001 <0.0001 (0.41,0.72) (0.24,0.43) 0.67 0.49 0.27 0.11 0.003 0.01 (0.42,1.09) (0.30,0.78) 0.47 0.28 0.006 <0.001 <0.001 <0.0001 (0.34,0.66) (0.21,0.39) -1835 -2344 0.03 0.005 <0.001 0.002 (-3101, -568) (-3633, -1054)
M AN U
Intraoperative Complication Postoperative Complication
Low
TE D
Outcome
Very Low OR p-value (95% CI)
RI PT
adjusted for age at surgery, year of surgery, race, Elixhauser comorbidity index, median zip code income, admission type, hospital
p-value for test of quartile coefficients jointly zero
2
Marginal effects computed at age 55-64, year 2011, white race, Elixhauser comorbidity index 0-1, median zip code income
AC C
EP
1
≥$63,000, elective surgery, teaching hospital, south hospital region, urban hospital location, and private insurance.
4
ACCEPTED MANUSCRIPT
FIGURE LEGEND
Figure 1
Distribution of RARP annual caseload across hospital volume quartiles stratified
Figure 2
RI PT
by year of surgery.
Cumulative distribution of RARP caseload across hospital volume as a continuous variable, stratified by year of surgery.
Associations of hospital volume (modeled as a non-linear continuous variable)
SC
Figure 3
AC C
EP
TE D
M AN U
with perioperative outcomes and total hospital costs.
ACCEPTED MANUSCRIPT
Figure 1: Distribution of RARP annual caseload across hospital volume quartile, stratified by
AC C
EP
TE D
M AN U
SC
RI PT
year of surgery.
ACCEPTED MANUSCRIPT
Figure 2: Cumulative distribution of RARP caseload across hospital volume as a continuous variable, stratified by year of surgery. Hospitals with annual volume < 15 not represented due
AC C
EP
TE D
M AN U
SC
RI PT
NIS data suppression policy.
ACCEPTED MANUSCRIPT
Figure 3: Associations of hospital volume (modeled as a non-linear continuous variable) with perioperative outcomes and total hospital costs. Event rates estimated using marginal effects as
RI PT
described in Methods. Shaded area represents 95% confidence interval. Prob=probability.
AC C
EP
TE D
M AN U
SC
Intraoperative complications:
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Postoperative complications:
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Perioperative blood transfusion:
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Prolonged hospitalization:
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Total hospital costs:
ACCEPTED MANUSCRIPT
ABBREVIATIONS AHRQ: Agency for Healthcare Research and Quality
IQR: interquartile range MIRP: minimally-invasive radical prostatectomy NIS: Nationwide Inpatient Sample
SC
ORP: open radical prostatectomy
M AN U
pLOS: prolonged length of stay PSA: prostate-specific antigen
RARP: robotic-assisted radical prostatectomy
AC C
EP
TE D
RP: radical prostatectomy
RI PT
ICD-9: International Classification of Diseases Ninth Revision
PII: DOI: Reference:
S0022-5347(17)30184-2 10.1016/j.juro.2017.01.067 JURO 14341
To appear in:
The Journal of Urology
Please cite this article as: Gershman B, Meier SK, Jeffery MM, Moreira DM, Tollefson MK, Kim SP, Karnes RJ, Shah ND, Re-Defining and Contextualizing the Hospital Volume-Outcome Relationship for Robotic-Assisted Radical Prostatectomy: Implications for Centralization of Care, The Journal of Urology® (2017), doi: 10.1016/j.juro.2017.01.067. DISCLAIMER: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our subscribers we are providing this early version of the article. The paper will be copy edited and typeset, and proof will be reviewed before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to The Journal pertain.
Embargo Policy All article content is under embargo until uncorrected proof of the article becomes available online. We will provide journalists and editors with full-text copies of the articles in question prior to the embargo date so that stories can be adequately researched and written. The standard embargo time is 12:01 AM ET on that date. Questions regarding embargo should be directed to [email protected].
ACCEPTED MANUSCRIPT
1
Re-Defining and Contextualizing the Hospital Volume-Outcome Relationship for Robotic-
2
Assisted Radical Prostatectomy: Implications for Centralization of Care
3
Boris Gershman1, Sarah K. Meier2, Molly M. Jeffery2, Daniel M. Moreira3, Matthew K.
5
Tollefson4, Simon P. Kim5, R. Jeffrey Karnes4, and Nilay D. Shah2, 6
RI PT
4
6 1
Division of Urology, Rhode Island Hospital and The Miriam Hospital, Providence, RI
8
2
Mayo Clinic Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery,
9
Mayo Clinic, Rochester, MN
M AN U
SC
7
3
Department of Urology, University of Illinois at Chicago, Chicago, IL
11
4
Department of Urology, Mayo Clinic, Rochester, Minnesota
12
5
Department of Urology, University Hospitals Case Medical Center, Case Western Reserve
13
University School of Medicine, Urology Institute, Cleveland, OH, USA
14
6
OptumLabs, Cambridge, MA, USA
15
18 19 20 21 22
Nilay D. Shah, PhD
EP
17
Correspondence:
Mayo Clinic
200 First Street SW
AC C
16
TE D
10
Rochester, Minnesota 55905 Phone: (507) 266-2743 Fax: (507) 284-1731 E-mail: [email protected]
23
1
ACCEPTED MANUSCRIPT
24
Keywords: robotic surgery; radical prostatectomy; hospital volume; outcomes; complications
25
Manuscript word count: 2490; Abstract word count: 250
AC C
EP
TE D
M AN U
SC
RI PT
26
2
ACCEPTED MANUSCRIPT
ABSTRACT
28
Purpose: Robotic-assisted radical prostatectomy (RARP) has undergone rapid dissemination,
29
driven in part by market forces, to become the most frequently employed surgical approach in
30
the management of prostate cancer. Accordingly, a critical analysis of its volume-outcome
31
relationship has important health policy implications. We therefore evaluated the association of
32
hospital RARP volume with perioperative outcomes, and examined the distribution of hospital
33
RARP volumes to contextualize the volume-outcome relationship.
SC
RI PT
27
34
Materials and Methods: We identified 140,671 adult men who underwent RARP from 2009 to
36
2011 in the Nationwide Inpatient Sample. The associations of hospital volume with
37
perioperative outcomes and total hospital costs were evaluated using multivariable logistic
38
regression and generalized linear models.
TE D
39
M AN U
35
Results: In 2011, 70% of hospitals averaged one RARP per week or less, accounting for 28% of
41
RARPs. Compared to patients treated at the lowest quartile hospitals, those treated at the highest
42
quartile hospitals had significantly lower rates of intraoperative complications (0.6% vs
43
1.4%;Ptrend<0.001), postoperative complications (4.8% vs 13.9%;Ptrend<0.001), perioperative
44
blood transfusion (1.5% vs 4.0%; Ptrend<0.001), prolonged hospitalization (4.3% vs 13.8%;
45
Ptrend<0.001), and mean total hospital costs ($12,647 vs $15,394;Ptrend<0.001). When modeled as
46
a non-linear continuous variable, increasing hospital volume was independently associated with
47
improved rates of each perioperative endpoint up to approximately 100 RARPs per year, beyond
48
which there appeared to be marginal improvement.
AC C
EP
40
3
ACCEPTED MANUSCRIPT
49
Conclusions: Increasing hospital RARP volume was associated with improved perioperative
50
outcomes, up to approximately 100 RARP per year, beyond which there appeared to be marginal
51
improvement. A substantial proportion of RARPs are performed at low-volume hospitals.
AC C
EP
TE D
M AN U
SC
RI PT
52
4
ACCEPTED MANUSCRIPT
53 54
INTRODUCTION Radical prostatectomy (RP) is the most frequently utilized treatment modality for prostate cancer and carries a tremendous public health impact1, 2. While open RP (ORP) has long been
56
the gold-standard surgical approach, dissemination of robotic technology has resulted in a shift to
57
robotic-assisted radical prostatectomy (RARP) in recent years. By 2009, RARP accounted for
58
nearly 60% of all RPs with an annual volume of 90,000 cases3, 4, and in 2013 it had grown to
59
85%5. However, there is ongoing debate regarding the comparative effectiveness of ORP and
60
RARP, and some authors have raised concerns regarding the impact of RARP on costs and
61
outcomes3, 6-8.
SC
M AN U
62
RI PT
55
Notably, the rapid adoption of RARP occurred under unique circumstances, driven in part by direct-to-consumer marketing of varying accuracy4, 9 and increased market competition
64
among hospitals10, 11. Acquisition of robotic technology has been associated with a 114%
65
increase in radical prostatectomy (RP) volume, which has resulted in a paradoxical centralization
66
of RP11. By 2009, approximately one third of hospitals had acquired robotic technology and
67
performed 85% of RP cases12.
While higher surgeon and hospital volume have been associated with improved
EP
68
TE D
63
oncologic, functional, and perioperative outcomes for ORP13-17, there is little data examining the
70
volume-outcome relationship for RARP. This is particularly important given the recent
71
dissemination of RARP, driven more by market forces rather than traditional factors. We
72
therefore conducted a population-based analysis of the association of hospital volume with
73
perioperative outcomes following RARP, and examined the contemporary distribution of
74
hospital RARP volume to contextualize the public health impact of this volume-outcome
75
relationship.
AC C
69
76
5
ACCEPTED MANUSCRIPT
77
MATERIALS AND METHODS
78
Study Population
79
After receiving exempt status from the Mayo Clinic Institutional Review Board (IRB), we identified adult men who underwent RARP from the Nationwide Inpatient Sample (NIS)
81
from 2009-2011. The NIS is the largest publicly available, all-payer inpatient database in the
82
US. States participating in the NIS cover 97% of the US population18.
We identified men aged ≥18 years who underwent RARP for prostate cancer using ICD-9
SC
83
RI PT
80
procedure code 60.5 (radical prostatectomy) with modifier code 17.42 (robotic assisted
85
procedure) and ICD-9 diagnosis code 185 (prostate cancer). The study was limited to January
86
2009 through December 2011 based on the introduction of the robotic modifier code in October
87
2008 and a redesign of the NIS in 2012, which changed the sampling methodology. Men were
88
excluded if they lacked the appropriate ICD-9 codes.
M AN U
84
90 91
TE D
89
Patient and Hospital Characteristics
We abstracted the following patient and hospital characteristics: age at surgery, year of surgery, race, median zip code income, health insurance status, length of stay (LOS), admission
93
type (elective/emergent), hospital teaching status, hospital region, hospital location (rural/urban),
94
and annual hospital volume. Secondary diagnosis codes were used to calculate the Elixhauser
95
comorbidity index for each patient19. Surgeon volume is not available in the NIS. Observations
96
with missing values were dropped except for race, which was assigned to an “unknown”
97
category.
AC C
EP
92
98 99
Outcomes
6
ACCEPTED MANUSCRIPT
100
Intraoperative complications were defined as surgical laceration of an organ, nerve, or blood vessel using ICD-9 code 998.213. Postoperative complications were identified using
102
administrative codes and categorized as cardiac, respiratory, bowel, hemorrhage, infectious,
103
wound, deep venous thrombosis (DVT)/pulmonary embolism (PE), genitourinary, and
104
miscellaneous as described previously8, 20. Prolonged length of stay (pLOS) was defined as a
105
hospitalization duration >90th percentile. Perioperative transfusion was identified using ICD-9
106
codes 99.02 and 99.0413. In-hospital mortality was not evaluated given the rarity of events.
107
Total hospital costs were estimated using hospital-specific cost-to-charge ratios and adjusted to
108
2011 dollars, using the Agency for Healthcare Research and Quality (AHRQ) Component Price
109
Index for hospital care.
M AN U
SC
RI PT
101
110 111
Statistical Analysis
Annual hospital volume was both categorized into volume quartiles and modeled as a
113
continuous non-linear variable using restricted cubic splines with 3 knots at the 10th, 50th, and
114
90th percentiles of volume21, 22. Restricted cubic splines allow modelling of a non-linear
115
relationship between a continuous covariate and outcome using piecewise cubic polynomial
116
functions joined at “knots,” with linear functions at the tails21, 22. Volume quartile cutoffs and
117
spline knots were determined using pooled data across all study years; hospitals were then
118
categorized for each year separately according to annual volume. Baseline characteristics were
119
summarized using medians/interquartile ranges (IQR) and frequency counts/percentages, and
120
compared across hospital volume quartiles using the adjusted Wald test for means and Pearson’s
121
chi square for frequency counts/percentages. Perioperative outcomes were compared across
122
hospital volume quartiles using a nonparametric test for trend23. Multivariable logistic
AC C
EP
TE D
112
7
ACCEPTED MANUSCRIPT
regression was performed to further evaluate the associations of hospital volume quartile with
124
perioperative outcomes. Total hospital costs were evaluated using a generalized linear model,
125
specifying the gamma family and log link. Models were adjusted for patient and hospital
126
features including age at surgery (18-54, 55-64, 65-69, 70-74, ≥75 years), year of surgery, race,
127
Elixhauser comorbidity index (0-1, 2-3, 4+), median zip code income, admission type, insurance
128
status, hospital teaching status, hospital region, and hospital location. Separate multivariable
129
logistic regression models were constructed modeling hospital volume as a non-linear continuous
130
variable using restricted cubic splines with 3 knots. Predicted rates of perioperative outcomes
131
and total hospital costs were estimated using marginal effects at the most prevalent covariate
132
categories (age 55-64 years, year 2011, white race, Elixhauser comorbidity index 0-1, median zip
133
code income ≥$63,000, elective surgery, teaching hospital status, south hospital region, urban
134
hospital location, and private insurance).
SC
M AN U
Analyses were performed using Stata version 14 (StataCorp, College Station, TX), and
TE D
135
RI PT
123
all tests were two-sided with p-values ≤0.05 considered significant. Except where noted, counts
137
of hospitals and patients are scaled to the national population size, using provided weights for
138
discharges or hospitals, as appropriate. In accordance with the NIS Data Use Agreement, results
139
with 10 or fewer observations were censored.
AC C
140
EP
136
8
ACCEPTED MANUSCRIPT
141 142
RESULTS A total of 140,671 patients were discharged after undergoing RARP from January 2009 to December 2011 at 2472 hospitals. The total volume of RARP did not significantly differ
144
across study years (p=0.55 by adjusted Wald test). Hospital volume quartiles were defined as
145
very low (≤12), low (13-30), medium (31-66), and high (67-820). Baseline patient
146
characteristics stratified by hospital volume quartile are presented in Table 1. Mean age at
147
surgery was 61.5 (SD 7.2) years and 68% of men were Caucasian. There were significant
148
differences across hospital volume quartile with regard to several patient and hospital
149
characteristics. Specifically, higher volume quartile hospitals were more likely to be teaching
150
(p<0.0001) and urban hospitals (p=0.001), while patients treated at higher-volume hospitals were
151
more likely to have private insurance (p=0.004) and fewer comorbidities (p=0.0001), with a
152
trend toward higher median zip code income (p=0.07). These differences suggest that
153
regionalization may be associated with health disparities.
SC
M AN U
TE D
154
RI PT
143
During the study period, there was evidence of de-centralization of RARP volume. The proportion of RARP cases performed at high-volume quartile hospitals decreased from 78% in
156
2009 to 63% in 2011, driven by increasing volume at low- and medium-volume quartile
157
hospitals (Figure 1; p<0.001). Moreover, there was an increase in the proportion of hospitals that
158
were lower volume – by 2011, 70% of hospitals averaged one RARP per week or less, up from
159
59% in 2009.
AC C
160
EP
155
To more accurately capture subtle temporal changes in the distribution of RARP, we
161
plotted the cumulative distribution of RARP across hospital volume as a continuous variable,
162
stratified by year (Figure 2). Here, too, there was evidence of a leftward shift from higher to
9
ACCEPTED MANUSCRIPT
163
lower volume hospitals over the study period. For example, the proportion of RARPs performed
164
at hospitals averaging ≤1 RARP per week increased from 15% in 2009 to 28% in 2011.
165
We conducted an exploratory analysis to examine how total RP volume may contextualize changes in RARP volume over time. Over the study period, there was no
167
statistically significant change in total RP volume (p=0.46 by adjusted Wald test). We then
168
estimated the proportion of RARPs performed within each quartile of total RP volume
169
(Supplementary Figure 1). During the study period, RARP appeared to shift from hospitals in
170
the top quartile of total RP volume to those in the second highest quartile of total RP volume,
171
with very little change in RARP volume among the lowest two quartiles of total RP volume.
SC
M AN U
172
RI PT
166
Next, we examined the association of hospital volume with perioperative outcomes using a traditional quartile-based categorization of volume (Table 2). Increasing hospital volume
174
quartile was significantly associated with improved perioperative outcomes and lower total
175
hospital costs. Specifically, compared to patients treated at very low-volume quartile hospitals,
176
those treated at high-volume quartile hospitals had lower rates of intraoperative complications
177
(0.6% vs 1.4%; Ptrend<0.001), postoperative complications (4.8% vs 13.9%; Ptrend<0.001),
178
perioperative blood transfusion (1.5% vs 4.0%; Ptrend<0.001), and prolonged hospitalization
179
(4.3% vs 13.8%; Ptrend<0.001), as well as lower mean total hospital costs ($12,647 vs $15,394;
180
Ptrend<0.001).
EP
AC C
181
TE D
173
On multivariable analysis (Table 3) adjusted for patient and hospital features, treatment at
182
a high-volume quartile hospital remained independently associated with a significantly reduced
183
risk of intraoperative complications (OR 0.44; p=0.03), postoperative complications (OR 0.32;
184
p<0.001), perioperative blood transfusion (OR 0.49; p=0.003), prolonged hospitalization (OR
185
0.28; p<0.001), and lower total hospital costs (-$2344; p<0.001) compared to very low-volume
10
ACCEPTED MANUSCRIPT
186
hospitals. Moreover, there was a statistically significant overall trend toward improved
187
perioperative outcomes with increasing hospital volume quartile (p≤0.01 for each).
188
To assess for non-linear relationships between hospital volume and outcomes, we modeled hospital volume as a continuous non-linear variable using restricted cubic splines
190
(Supplementary Table 1). Predicted rates of perioperative outcomes and total hospital costs were
191
then estimated, adjusted for patient and hospital features using marginal effects at the most
192
prevalent covariate levels. The resulting relationships of hospital volume with perioperative
193
outcomes and total hospital costs are illustrated in Figure 3. Notably, there was a relatively
194
consistent relationship wherein perioperative outcomes improved until approximately 100
195
RARPs per year, with marginal improvement at higher volumes. However, even in 2011, 54%
196
of RARPs were performed at hospitals below this annual volume.
M AN U
SC
RI PT
189
AC C
EP
TE D
197
11
ACCEPTED MANUSCRIPT
198 199
DISCUSSION In this study, we examined the association of hospital RARP volume with perioperative outcomes and total hospital costs, modeling hospital volume using both a conventional, quartile-
201
based categorization and as a non-linear continuous variable. Increasing hospital volume
202
quartile was associated with improved perioperative outcomes – that is, significantly lower rates
203
of perioperative complications, blood transfusion, and prolonged hospitalization, even when
204
adjusting for patient and hospital features. Notably, a more accurate illustration of the hospital
205
volume-outcome relationship emerged when hospital volume was modeled as a non-linear
206
continuous variable. Specifically, although perioperative outcomes improved with increasing
207
hospital volume, the association was most pronounced until approximately 100 RARPs per year,
208
beyond which the marginal benefit appeared to diminish.
M AN U
SC
RI PT
200
In order to contextualize the described volume-outcome relationship from a health policy
210
perspective, we also examined the cumulative distribution of hospital RARP volume. In contrast
211
to prior reports that noted a centralization of total RP volume12, 24, driven in part by year-over-
212
year volume increases following acquisition of robotic technology11, we observed a de-
213
centralization of RARP volume from 2009 to 2011. During that time period, the proportion of
214
hospitals averaging ≤1 RARP per week increased from 59% to 70%, with the corresponding
215
proportion of RARPs performed at such hospitals increasing from 15% to 28%. This may reflect
216
continued dissemination of robotic technology with a shift toward lower volume facilities. More
217
importantly, a substantial proportion of RARPs are performed at facilities that lie along the
218
“steep” portion of the volume-outcome curves, whereby improvements in perioperative
219
outcomes may be expected with increases in volume. This has important policy implications,
220
particularly given that acquisition of robotic technology has been driven more by direct-to-
AC C
EP
TE D
209
12
ACCEPTED MANUSCRIPT
221
consumer marketing4, 9 and increased market competition among hospitals10, 11 than traditional
222
technologic adoption based on clinical outcomes.
223
There exists an extensive body of literature supporting improved oncologic, functional, and perioperative outcomes for ORP performed at high-volume hospitals or by high-volume
225
surgeons17. However, there is little data on the volume-outcome relationship specifically for
226
RARP. Prior to October 2008, when the robotic-modifier code was introduced, it was not
227
possible to distinguish laparoscopic prostatectomy from RARP, and accordingly, few studies
228
have examined this question. Yu et al25 provided one of the first such reports in examining the
229
volume-outcome relationship for RARP using data from the last quarter of 2008 in the NIS. The
230
authors noted reduced complication rates, shorter LOS, and lower median costs, although the
231
study may be subject to bias related to analysis of data immediately after introduction of the new
232
robotic modifier code. Sammon et al26 compared perioperative outcomes and costs for men
233
undergoing ORP and RARP in 2009 using the NIS. The authors noted superior outcomes for
234
RARP overall and for high-volume hospitals regardless of surgical technique. Our study
235
provides further support for the hospital volume-perioperative outcome relationship for RARP.
236
Moreover, examination of volume as a non-linear continuous variable provides important
237
insights into the true underlying volume-outcome relationship.
SC
M AN U
TE D
EP
This is also the first study, to our knowledge, to examine the cumulative distribution of
AC C
238
RI PT
224
239
hospital RARP volume. We believe this is critical to contextualizing the population-level impact
240
of the described volume-outcome relationship. In the present study, we observed de-
241
centralization of RARP from 2009-2011, with a marked shift toward lower volume hospitals
242
with anticipated worse perioperative outcomes. A recent study27 also noted a decrease in the
13
ACCEPTED MANUSCRIPT
243
median hospital RARP caseload, suggesting a shift from high-volume to low-volume hospitals
244
with an associated increase in complication rates from 2009-2011.
245
Centralization of major surgery has been a debated topic and has potential disadvantages such as increased travel distance12, which must be weighed against the potential benefits in terms
247
of outcomes. As such, US health policy has not mandated centralization in any form for urologic
248
surgery28. However, lessons can be learned from the United Kingdom, where in 2002 the
249
National Health Services (NHS) recommended that RP only be performed in hospitals with
250
annual volumes of at least 50 cases, effectively mandating centralization of cancer services29.
251
Although such a wide-scale reorganization posed substantial logistical challenges and
252
necessitated initiation of quality-assurance programs, recent data suggest overall improvements
253
in quality of care30.
SC
M AN U
254
RI PT
246
This study has several limitations. It is a population-based analysis and subject to the potential biases of claims-based ascertainment, although an extensive body of literature supports
256
the methodology applied in this report. In addition, due to limitations of the inpatient dataset, we
257
were unable to examine post-discharge adverse events, 90-day attributable costs, oncologic
258
outcomes, or functional outcomes. Furthermore, due to a change in NIS sampling methodology
259
in 2012, we were unable to examine trends over a longer time period. Also, total RP volume
260
may affect the association of RARP volume with perioperative outcomes and deserves further
261
investigation. Lastly, we were unable to adjust for surgeon volume and disease characteristics,
262
such as tumor grade or stage, which are not captured in the NIS.
EP
AC C
263
TE D
255
14
ACCEPTED MANUSCRIPT
264 265
CONCLUSIONS We observed improved perioperative outcomes with increasing hospital volume up to approximately 100 RARPs per year, beyond which there appeared to be diminishing marginal
267
returns. Notably, there has been a recent de-centralization of hospital RARP volume, with nearly
268
one third of cases in 2011 performed at hospitals averaging ≤1 RARP per week. While further
269
studies are necessary to identify additional determinants of perioperative outcomes and
270
hospitalization costs, these results have important implications for health policy.
SC
RI PT
266
AC C
EP
TE D
M AN U
271
15
ACCEPTED MANUSCRIPT
272
REFERENCES
273
1.
274
primary treatment of localized prostate cancer. J Clin Oncol, 28: 1117, 2010
275
2.
276
Natl Compr Canc Netw, 12: 686, 2014
277
3.
278
assisted and open radical prostatectomy in the postdissemination era. J Clin Oncol, 32: 1419,
279
2014
280
4.
281
promotion of robotic prostatectomy exhibits varying quality of information. Health Aff
282
(Millwood), 31: 760, 2012
283
5.
284
Prostatectomy: A Contemporary Analysis of an All-payer Discharge Database. Eur Urol, 70:
285
837, 2016
286
6.
287
laparoscopic versus open retropubic radical prostatectomy among a nationwide random sample
288
of medicare-age men. J Clin Oncol, 30: 513, 2012
289
7.
290
open radical prostatectomy. Jama, 302: 1557, 2009
291
8.
292
prostatectomy attributable to robotic surgery. Eur Urol, 64: 11, 2013
293
9.
294
information is available? Urology, 77: 941, 2011
Cooperberg, M. R., Broering, J. M., Carroll, P. R.: Time trends and local variation in
RI PT
Mohler, J. L., Kantoff, P. W., Armstrong, A. J. et al.: Prostate cancer, version 2.2014. J
SC
Gandaglia, G., Sammon, J. D., Chang, S. L. et al.: Comparative effectiveness of robot-
M AN U
Mirkin, J. N., Lowrance, W. T., Feifer, A. H. et al.: Direct-to-consumer Internet
TE D
Leow, J. J., Chang, S. L., Meyer, C. P. et al.: Robot-assisted Versus Open Radical
EP
Barry, M. J., Gallagher, P. M., Skinner, J. S. et al.: Adverse effects of robotic-assisted
AC C
Hu, J. C., Gu, X., Lipsitz, S. R. et al.: Comparative effectiveness of minimally invasive vs
Kim, S. P., Shah, N. D., Karnes, R. J. et al.: Hospitalization costs for radical
Steinberg, P. L., Ghavamian, R.: Searching robotic prostatectomy online: what
16
ACCEPTED MANUSCRIPT
295
10.
Barbash, G. I., Friedman, B., Glied, S. A. et al.: Factors associated with adoption of
296
robotic surgical technology in US hospitals and relationship to radical prostatectomy procedure
297
volume. Ann Surg, 259: 1, 2014
298
11.
299
technology and hospital prostatectomy volumes: increasing market share through the adoption of
300
technology. Cancer, 118: 371, 2012
301
12.
302
centralization, robotics, and access to urologic cancer care. Cancer, 118: 54, 2012
303
13.
304
contemporary population-based analysis. Urology, 79: 332, 2012
305
14.
306
prostatectomy. N Engl J Med, 346: 1138, 2002
307
15.
308
cancer control after radical prostatectomy. Journal of the National Cancer Institute, 99: 1171,
309
2007
310
16.
311
radical prostatectomy as markers of surgical quality: analysis of heterogeneity between surgeons
312
at a single cancer center. Eur Urol, 59: 317, 2011
313
17.
314
outcome relationship for radical prostatectomy. Eur Urol, 64: 786, 2013
315
18.
316
19.
317
administrative data. Med Care, 36: 8, 1998
RI PT
Neuner, J. M., See, W. A., Pezzin, L. E. et al.: The association of robotic surgical
SC
Stitzenberg, K. B., Wong, Y. N., Nielsen, M. E. et al.: Trends in radical prostatectomy:
M AN U
Schmitges, J., Sun, M., Abdollah, F. et al.: Blood transfusions in radical prostatectomy: a
Begg, C. B., Riedel, E. R., Bach, P. B. et al.: Variations in morbidity after radical
TE D
Vickers, A. J., Bianco, F. J., Serio, A. M. et al.: The surgical learning curve for prostate
AC C
EP
Vickers, A., Savage, C., Bianco, F. et al.: Cancer control and functional outcomes after
Trinh, Q. D., Bjartell, A., Freedland, S. J. et al.: A systematic review of the volume-
McCannon, J., Berwick, D. M.: A new frontier in patient safety. JAMA, 305: 2221, 2011 Elixhauser, A., Steiner, C., Harris, D. R. et al.: Comorbidity measures for use with
17
ACCEPTED MANUSCRIPT
318
20.
Hu, J. C., Wang, Q., Pashos, C. L. et al.: Utilization and outcomes of minimally invasive
319
radical prostatectomy. J Clin Oncol, 26: 2278, 2008
320
21.
321
functions in public health research. Stat Med, 29: 1037, 2010
322
22.
323
Logistic Regression, and Survival Analysis: New York: Spriner, 2001
324
23.
Cuzick, J.: A Wilcoxon-type test for trend. Stat Med, 4: 87, 1985
325
24.
Anderson, C. B., Penson, D. F., Ni, S. et al.: Centralization of radical prostatectomy in
326
the United States. J Urol, 189: 500, 2013
327
25.
328
outcomes of robot-assisted laparoscopic radical prostatectomy. J Urol, 187: 1632, 2012
329
26.
330
prostatectomy: the differential effect of regionalization, procedure volume and operative
331
approach. J Urol, 189: 1289, 2013
332
27.
333
Diffusion: Complications Following Robot-Assisted Radical Prostatectomy. BJU Int, 2015
334
28.
335
considerations. J Am Coll Surg, 190: 341, 2000
336
29.
337
improving outcomes in urological cancers, 2002
338
30.
339
Prostate Cancer Surgery During Reorganisation of Cancer Services. Eur Urol, 2015
RI PT
Desquilbet, L., Mariotti, F.: Dose-response analyses using restricted cubic spline
M AN U
SC
Harrell, F. E., Jr.: Regression Modeling Strategies: With Applications to Linear Models,
Yu, H. Y., Hevelone, N. D., Lipsitz, S. R. et al.: Hospital volume, utilization, costs and
TE D
Sammon, J. D., Karakiewicz, P. I., Sun, M. et al.: Robot-assisted versus open radical
EP
Sammon, J. D., Abdollah, F., Klett, D. E. et al.: The Diminishing Returns of Robotic
AC C
Birkmeyer, J. D.: Should we regionalize major surgery? Potential benefits and policy
National Institute for Clinical Excellence (NICE). Guidance on cancer services:
Cathcart, P., Sridhara, A., Ramachandran, N. et al.: Achieving Quality Assurance of
340
18
ACCEPTED MANUSCRIPT
341
ACKNOWLEDGEMENTS
342
None.
AC C
EP
TE D
M AN U
SC
RI PT
343
19
ACCEPTED MANUSCRIPT
344
CONFLICT OF INTEREST
345
The authors have no conflicts of interest to disclose.
AC C
EP
TE D
M AN U
SC
RI PT
346
20
ACCEPTED MANUSCRIPT
Table 1: Baseline characteristics of patients who underwent RARP from January 2009 through December 2011 at 2472 hospitals (N=140,671). Numbers represent mean (SD) or %.
No. patients2 No. hospitals2 Annual hospital volume, median (range) Age at surgery, mean (yrs), [SD] Year of Surgery (%) 2009 2010 2011 Race (%) Caucasian African-American Hispanic Other Unknown Elixhauser Comorbidity Index (%) 0-1 2-3 4+ Median Zip Code Income (%) $1-38999 $39000-47999 $48000-62999 $63000+ Admission Type (%) Elective Non-Elective Hospital Teaching Status (%) Teaching Non-teaching Hospital Region (%)
140,671 2,472
Volume Quartile Medium
12,407 581 20 (13, 30)
28,457 618 44 (31, 66)
61.5 [7.2]
61.6 [7.1]
61.6 [7.3]
33.6 30.3 36.1
31.6 33.8 34.5
16.7 44.6 38.7
68.2 9.6 4.8 5.4 11.9
64.1 13.0 10.4 6.3 6.1
p-value1 High
95,873 617 100 (67, 820)
SC
Low
(x, 820)
Very Low 3,934 655 x (x,12)
RI PT
Characteristic
Total
-
61.5 [7.3]
0.943
24.9 30.7 44.4
38.4 28.2 33.4
0.03
64.8 12.3 5.4 3.2 14.3
67.9 9.8 5.1 4.5 12.7
68.9 9.1 4.4 5.9 11.6
69.5 26.8 3.7
70.3 26.5 3.2
70.2 27.4 2.4
74.0 24.0 2.0
0.0001
18.0 22.4 26.7 32.8
22.9 26.4 26.1 24.6
23.3 23.9 26.6 26.1
16.1 21.5 27.1 35.3
17.7 22.4 26.7 33.3
0.07
95.9 4.1
88.5 11.5
94.3 5.8
97.9 2.1
95.9 4.1
0.03
70.5 29.5
37.9 62.1
53.3 46.7
56.9 43.1
78.2 21.8
<0.0001
TE D
M AN U
61.5 [7.1]
AC C
EP
72.8 25.0 2.2
0.38
1
ACCEPTED MANUSCRIPT
23.1 16.2 40.7 20.0
17.6 31.1 36.9 14.4
14.0 25.6 34.8 25.5
20.2 23.0 32.4 24.4
0.59
2.1 97.9
7.9 92.1
6.9 93.1
3.7 96.3
0.7 99.3
0.001
31.7 1.8 63.0 3.5
34.7 4.1 53.9 7.3
35.0 1.8 59.3 3.8
31.1 1.8 63.3 3.8
31.4 1.7 63.7 3.2
0.004
RI PT
18.8 24.0 33.5 23.6
SC
Northeast Midwest South West Hospital Location (%) Rural Urban Insurance Category (%) Medicare Medicaid Private Other
Pearson’s chi-square test
2
Weighted figures given. The unweighted hospital count was 507; the unweighted discharge
count was 28,454 across the three years. 3
Adjusted Wald test
M AN U
1
AC C
EP
TE D
x=not depicted due to cell suppression policy
2
ACCEPTED MANUSCRIPT
Table 2: Perioperative outcomes stratified by hospital volume quartile.
1737 100 323 X 48 254 1262 514 1542
13.9 X 3.7
Ptrend
RI PT
High
1.3
1.1
0.6
8.8 0.5 2.1
8.0 0.6 1.4
4.8 0.3 0.8
0.2 1.1 6.0 2.2 6.6
0.2 0.8 3.4 1.5 4.3
<0.001 0.03 <0.001 X 0.20 0.20 <0.001 <0.001 <0.001
$13,674
$12,647
<0.001
X X 1.4 10.8 4.0 13.8
0.8 6.6 2.5 8.8
$15,394
$13,787
<0.001
SC
222
Volume Quartile Low Medium
Very Low 1.4
M AN U
Intraoperative complication (%) Postoperative complication (%) Any Cardiac Respiratory Vascular Wound/Bleeding Genitourinary Miscellaneous Blood transfusion (%) Prolonged hospitalization (%) Total costs (mean)
# of events*
TE D
Outcome
Unweighted count of events
AC C
*
EP
X=Not depicted due to cell suppression policy
3
ACCEPTED MANUSCRIPT
Table 3. Multivariable associations of hospital volume quartile with perioperative outcomes and adjusted total hospital costs. Model
teaching status, hospital region, hospital location, and insurance status.
ref
1.00
ref
Blood Transfusion
1.00
ref
Prolonged Length of Stay
1.00
ref
–
–
Total Costs2
SC
1.00
OR (95% CI) 1.07 (0.50, 2.30) 0.59 (0.44,0.80) 0.74 (0.43,1.27) 0.61 (0.43,0.87) -1553 (-2956, -151)
Volume Quartile Medium High p-value1 OR pOR pp-value (95% CI) value (95% CI) value 0.91 0.44 0.86 0.81 0.03 <0.0001 (0.43, 1.9) (0.21, 0.94) 0.54 0.32 0.001 <0.001 <0.001 <0.0001 (0.41,0.72) (0.24,0.43) 0.67 0.49 0.27 0.11 0.003 0.01 (0.42,1.09) (0.30,0.78) 0.47 0.28 0.006 <0.001 <0.001 <0.0001 (0.34,0.66) (0.21,0.39) -1835 -2344 0.03 0.005 <0.001 0.002 (-3101, -568) (-3633, -1054)
M AN U
Intraoperative Complication Postoperative Complication
Low
TE D
Outcome
Very Low OR p-value (95% CI)
RI PT
adjusted for age at surgery, year of surgery, race, Elixhauser comorbidity index, median zip code income, admission type, hospital
p-value for test of quartile coefficients jointly zero
2
Marginal effects computed at age 55-64, year 2011, white race, Elixhauser comorbidity index 0-1, median zip code income
AC C
EP
1
≥$63,000, elective surgery, teaching hospital, south hospital region, urban hospital location, and private insurance.
4
ACCEPTED MANUSCRIPT
FIGURE LEGEND
Figure 1
Distribution of RARP annual caseload across hospital volume quartiles stratified
Figure 2
RI PT
by year of surgery.
Cumulative distribution of RARP caseload across hospital volume as a continuous variable, stratified by year of surgery.
Associations of hospital volume (modeled as a non-linear continuous variable)
SC
Figure 3
AC C
EP
TE D
M AN U
with perioperative outcomes and total hospital costs.
ACCEPTED MANUSCRIPT
Figure 1: Distribution of RARP annual caseload across hospital volume quartile, stratified by
AC C
EP
TE D
M AN U
SC
RI PT
year of surgery.
ACCEPTED MANUSCRIPT
Figure 2: Cumulative distribution of RARP caseload across hospital volume as a continuous variable, stratified by year of surgery. Hospitals with annual volume < 15 not represented due
AC C
EP
TE D
M AN U
SC
RI PT
NIS data suppression policy.
ACCEPTED MANUSCRIPT
Figure 3: Associations of hospital volume (modeled as a non-linear continuous variable) with perioperative outcomes and total hospital costs. Event rates estimated using marginal effects as
RI PT
described in Methods. Shaded area represents 95% confidence interval. Prob=probability.
AC C
EP
TE D
M AN U
SC
Intraoperative complications:
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Postoperative complications:
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Perioperative blood transfusion:
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Prolonged hospitalization:
ACCEPTED MANUSCRIPT
AC C
EP
TE D
M AN U
SC
RI PT
Total hospital costs:
ACCEPTED MANUSCRIPT
ABBREVIATIONS AHRQ: Agency for Healthcare Research and Quality
IQR: interquartile range MIRP: minimally-invasive radical prostatectomy NIS: Nationwide Inpatient Sample
SC
ORP: open radical prostatectomy
M AN U
pLOS: prolonged length of stay PSA: prostate-specific antigen
RARP: robotic-assisted radical prostatectomy
AC C
EP
TE D
RP: radical prostatectomy
RI PT
ICD-9: International Classification of Diseases Ninth Revision