Influence of facility caseload on subsequent survival of men with localized prostate cancer undergoing radical prostatectomy

Abstract

Background:

Several studies have investigated the relationship between experience measured in caseload and oncological outcomes, economics, and access to care for prostate cancer care. Oncological outcomes were limited to biochemical failure after radical prostatectomy (RP). Questions remain regarding the more definitive measures of outcome and their relationship to caseload.

Methods:

National Cancer Database was used to investigate the outcomes of RP in the United States. Using overall survival (OS) as the primary outcome, the relationship between facility annual caseload (FAC) for all prostate cancer encounters and facility annual surgical caseload (FASC) for those requiring RP were examined using Cox proportional hazards model. Four volume groups (VG) were defined as VG1: <50^th, VG2: 50^th-74^th, VG3: 75^th-89^th and VG4: top 10 percentile of caseload. Using FAC/FASC, 11/8, 17/18, 25/26, and 47/49 percent of patients were treated at VG 1 through 4, respectively.

Results:

Between 2004 and 2014, 488,389 patients underwent RP. At a median follow-up of 60.75 months, the median OS was not reached. There was a significant OS benefit as caseload increased. For FAC, the adjusted OS difference between VG1 and VG4 at 90^th percentile survivorship reached 13.2 months, HR 1.30 (95% CI: 1.23–1.36, p<0.0001). For FASC this was 11.3 months, HR 1.26 (95% CI: 1.192–1.321, p<0.0001).

Conclusion:

There is a statistically significant OS advantage to performing RP at a facility with a high annual caseload. Caseload measured in all prostate cancer encounters is a better predictor of favorable outcome compared to the number of surgeries performed at a facility.

Precis:

There is a statistically significant OS advantage to performing RP at facilities with a high annual caseload.

Introduction

The role of provider experience in determining outcomes of radical prostatectomy has previously been investigated and reported ^1–6. A previous outcomes and economics analysis reported that a radical prostatectomy in the hands of a highly-experienced surgeon would result in societal savings by reducing the costs of downstream management of treatment failure which could offset, at least in part, the costs of referral to a higher volume surgeon⁷. A follow up report also examined the travel distance needed to move from a below median volume surgeon to a top 15^th percentile surgeon and demonstrated that about 50% of radical prostatectomies by below median surgeons could be redirected to top 15 percentile providers within a 100 mile driving distance, with a median driving distance of 22.5 miles⁸.

Although the association between experience and outcomes appears well established there are still many unanswered questions as to why this phenomenon exists. Firstly, the outcomes investigated were limited to recurrence free survival, hospital stay, transfusion rates, complication rates or other short-term endpoints. The impact on oncologic outcome of overall survival has not been established. It also remains unclear whether this phenomenon is explained by patient demographics or disease characteristics, or adherence to guidelines for oncological care. Finally, it remains unclear whether this is related to experience by the individual surgeon or reflects the cumulative experience of the facility where the care is provided.

This study aims to conduct an in-depth examination of the determinants of outcomes for radical prostatectomy using the overall survival as the primary outcome and facility volume and facility surgical volume as predictors of outcome.

Methods

Data Source

The National Cancer Database (NCDB) represents a broad collection of prostate cancer cases treated at the Commission on Cancer (CoC) accredited facilities across the united states. This database represents treatment outcomes for radical prostatectomy at more than 1229 healthcare facilities over a 10 year span and includes detailed information about patient, disease, and healthcare facility characteristics, in addition to outcomes of care.

Variables

A list of variables in the dataset is available through NCDB portal (https://www.facs.org/quality-programs/cancer/ncdb). The primary outcome measure was overall survival defined as time from the date of radical prostatectomy to the date of death of all-causes, or censored on the date of last follow-up. Secondary outcomes measures were defined as operative and post-operative outcomes. Operative outcomes included Gleason’s score, surgical margin status, node status, and composite risk status based on NCCN guidelines. Post-operative outcomes included radiation therapy, hormonal therapy, and PSA recurrence.

In assessing the role of caseload, it is important to distinguish between the total prostate cancer cases treated (irrespective of modality or stage) and the total number of radical prostatectomy surgeries performed. Thus, two different facility caseload variables were defined in this analysis: 1) the average facility annual caseload (FAC) calculated using the number of prostate cancer patients treated at a facility each year regardless of stage or modality of therapy and 2) the average facility annual surgical caseload (FASC) calculated using the number of radical prostatectomies performed at a facility each year. FAC and FASC variables were used to assign facilities to one of 4 Volume Groups (VG): 1) below 50^th percentile, 2) 50–74^th percentile, 3) 75–90^th percentile, and 4) top 10 percentile. Based on the results pathology report after radical prostatectomy and using current NCCN practice guidelines and whether adjuvant treatment (radiation and or androgen deprivation) was given the appropriateness of post-operative treatment was determined and assigned as 1) Adjuvant Treatment Indicated and Given 2) Adjuvant Treatment Indicated but Not Given 3) Adjuvant Treatment Not Indicated but Given 4) Adjuvant Treatment Not Indicated and Not Given.

In addition to Volume Groups (VG), patient demographics, disease characteristics and facility characteristics categories were used to model predictors of outcomes.

Patient Selection

The study population included patients in NCDB with a diagnosis of localized prostate cancer who were treated primarily by radical prostatectomy. Patients that received radiation or systemic therapy prior to surgery were excluded from analysis. Patients who received radiation therapy or systemic therapy both before and after radical prostatectomy were also excluded from analysis. Among the remaining patients, those with metastatic disease were excluded. Finally, patients with time from diagnosis (biopsy) to radical prostatectomy that was missing, 0 or 1 day were also excluded (see Figure 1).

Figure 1:

Patient Selection Criteria.

Hypothesis

It was hypothesized that experience represented by annual caseload is a predictor of oncological outcome. It was further hypothesized that facility annual caseload (FAC) defined to include all cases with a diagnosis of prostate cancer regardless of stage or treatment modality (N=1,280,019) was a stronger predictor of outcome that facility annual surgical caseload (FASC) which was defined to include only prostate cancer patients undergoing radical prostatectomy as described in patient selection (N=592,587).

Such a distinction was previously reported in a SEER Medicare analysis for urothelial carcinoma⁹.

Statistical Analysis

Descriptive analysis was used to examine the differences of patient characteristics and disease characteristics and facility characteristics categories by the volumes groups (VG) based on FAC and FASC. Multivariable Cox proportional hazards regression model was used to analyze overall survival (OS) by volume groups (VG) when adjusting for other prognostic factors. The potential correlation of overall survival among patients treated in the same facility was taken into account using the robust sandwich estimates of Lin and Wei with the Cox regression model¹⁰. Departures from the proportional hazards assumption for the model were examined graphically using hazard plots and revealed no violations. Sensitivity analysis was performed to detect the patterns of the associations between VG and OS using the univariable Cox model. Multivariable model was re-examined by excluding 90-day mortality, by whether treatment was provided at more than 1 center, by cumulative years of diagnosis, and in subsets of patients by Gleason’s score and NCCN risk group.

All analyses were conducted using SAS statistical software (version 9.4; SAS Institute Inc, Cary, NC). All tests were 2-sided at a significance level of .05.

Results

Between 2004 and 2014, 488,389 patients who underwent radical prostatectomy as the primary therapy for localized prostate cancer were identified. Median overall age for the patients was 61 (23–90) and interquartile range of 56 to 66. Over 80% of patients were of white race and 63.5% had private insurance. Forty-four and 40 percent of patients were treated at Academic and Comprehensive Community Cancer Programs, respectively. Eighty four percent of patients had a Charlson-Deyo comorbidities score of 0 and only less than 2% had a comorbidity score of two or more. Table 1 summarizes patient characteristics (for more details refer to Supplementary Table 1).

Table 1:

Patient characteristics.

Variable	N	%
Number of Patients	488,389
Median Age at Diagnosis	61
Interquartile range	56,66
Range	(23–90)
Race/ethnicity
Non-Hispanic White	393,343	80.5%
Non-Hispanic Black	56,046	11.5%
Asian	7,440	1.5%
Hispanic	17,908	3.7%
Native American	882	0.2%
Other	12,770	2.6%
Primary Payor
Not Insured	6,923	1.4%
Private Insurance	315,812	64.7%
Medicaid	7,544	1.5%
Medicare	141,743	29.0%
Other Government	6,374	1.3%
Insurance Status Unknown	9,993	2.0%
Charlson-Deyo Score
0	411,213	84.2%
1	68,701	14.1%
2	8,475	1.7%
Great Circle Distance
Median	14
Facility Type
Community Cancer Program	26,845	5.5%
Comprehensive Community Cancer Program	194,661	39.9%
Academic/Research Program	212,054	43.4%
Integrated Network Cancer Program	54,298	11.1%
Other or unknown	531	0.1%

Patients were treated at 1,245 facilities across the United States. Forty five percent of these facilities were Comprehensive Community Cancer Programs and 32% were Community Cancer Programs. Table 2 summarizes the facility characteristics.

Table 2.

Facility characteristics and distribution of radical prostatectomies based on volume group.

	Volume Group Based on Facility Annual Caseload (FAC)								All		Volume Group Based on Facility Annual Surgical Caseload (FASC)
Group Number	1		2		3		4				1		2		3		4
Caseload (Percentile)	<63 (<50)		63–115 (50–74)		116–207 (75–89)		>207 (>=90)				<22 (<50)		22–53 (50–74)		54–111 (75–89)		>111 (>=90)
Type of Facilities
Community Cancer Program	367	61%	22	7%	5	3%	5	4%	399	32%	352	57%	36	12%	6	3%	5	4%
Comprehensive Community Cancer Program	200	33%	232	73%	104	54%	29	22%	565	45%	222	36%	199	64%	109	57%	35	27%
Academic/Research Program	30	5%	58	18%	65	34%	73	55%	226	18%	32	5%	72	23%	50	26%	72	56%
Integrated Network Cancer Program	6	1%	5	2%	18	9%	26	20%	55	4%	8	1%	5	2%	25	13%	17	13%
Total N of facilities	603	100%	317	100%	192	100%	133	100%	1245	100%	614	100%	312	100%	190	100%	129	100%
Number of Patients Served
Community Cancer Program	22,143	43%	4,038	5%	559	0.5%	105	0.05%	26,845	6%	16,743	43%	8,067	9%	1,802	1%	233	0.1%
Comprehensive Community Cancer Program	25,528	49%	61,677	73%	69,201	56%	38,255	17%	194,661	40%	18,682	48%	54,643	64%	71,966	58%	49,370	21%
Academic/Research Program	4,065	8%	16,379	20%	43,494	35%	148,116	65%	212,054	43%	3,204	8%	21,327	25%	35,074	28%	152,449	64%
Integrated Network Cancer Program	312	1%	1,836	2%	9,858	8%	42,292	18%	54,298	11%	647	2%	1,617	2%	15,772	13%	36,262	15%
Total Number of Patients	52,048	11%	83,930	17%	123,112	25%	228,768	47%	487,858	100%	39,276	8%	85,654	18%	124,614	26%	238,314	49%

Using the prostate cancer annual caseload (FAC), there were 613 facilities in below median (Volume Group 1), 308 in 50–74^th percentile (Volume Group 2), 185 in 75–89^th percentile (Volume Group 3), and 123 in the top 10 percentile (Volume Group 4). Similarly, using prostate surgery (radical prostatectomy) annual caseload (FASC), there were 612 facilities below median, 304 in 50–74^th percentile, 182 in 75–89^th percentile, and 123 in the top 10 percentile.

Gleason’s score of 6 and 7 accounted for more than 87% and 90% of diagnoses on biopsy and radical prostatectomy specimens, respectively. Among all patients undergoing radical prostatectomy, comparison of biopsy and prostatectomy Gleason scores revealed a decline of 13% in Gleason score 6 and an increase of 15% in Gleason score 7. Gleason score 8 and 10 also showed a decline of 3.5% from 8.1% to 4.6 and 0.1% from 0.2% to 0.1% in biopsy and prostatectomy specimens, respectively. Gleason score 9 increased by 1.3% from 4.2% to 5.5% in biopsy and prostatectomy specimens.

Based on clinical staging 21.3% patients were considered low risk; however, this figure declined by 19% to 2.3% when risk levels were reassessed based on prostatectomy specimen. Conversely, the rates of intermediate, high and very high risk cancers increased by 12, 3 and 4 percent. The mean positive margin rate was 21.1% for all patients.

Results by Facility Volume Groups

Using facility annual caseload (FAC), number of radical prostatectomies of <63, 63 to 115, 116 to 207 and > 207 reflected Volume Groups 1 through 4, respectively. Using facility annual surgical caseload (FASC), these figures were <22, 22 to 53, 54–111 and > 111 for Volume Groups 1 through 4, respectively. The majority of surgeries in Volume Groups 1, 2 and 3 were done at Comprehensive Community Cancer Programs, while the majority of surgeries at facilities with highest volume were performed in Academic/Research Programs. Overall, 44% of surgeries were performed at Academic/Research Programs which constitute only 18% of facilities- Table 2.

The median number of nodes examined was 2 with a range of 0 to 90 across all groups; however, the interquartile range was slightly higher for Volume Group 4 (0– 7) vs (0–5) for Volume Groups 1, 2, and 3. The average rate was node positive disease was 2%- Table 3.

Table 3.

Disease characteristics based on volume groups.

	Volume Group based on Facility Annual Caseload (FAC)								All		Volume Group based on Facility Annual Surgical Caseload (FASC)
Group Number	1		2		3		4				1		2		3		4
Caseload (Percentile)	<63 (<50)		63–115 (50–74)		116–207 (75–89)		>207 (>=90)		Total		<22 (<50)		22–53 (50–74)		54–111 (75–89)		>111 (>=90)
N	52,089	11%	83,993	17%	123,206	25%	229,101	47%	488,389	100%	39,304	8%	85,724	18%	124,720	26%	238,641	49%
Surgical Margins
Positive	12,561	25%	20,224	24%	27,434	23%	43,764	19%	103,983	22%	9,960	26%	20,085	24%	29,622	24%	44,316	19%
Negative	38,596	75%	62,554	76%	94,359	77%	180,805	81%	376,314	78%	28,414	74%	64,402	76%	93,853	76%	189,645	81%
Gleason’s Score on Needle Core Biopsy/TURP
6	8,305	42.8%	14,952	43.5%	20,665	41.7%	37,633	42.3%	81,555	42.4%	5,740	41.9%	15,722	42.8%	21,879	42.6%	38,214	42.2%
7	8,541	44.0%	15,028	43.7%	22,289	45.0%	40,554	45.5%	86,412	44.9%	6,010	43.8%	16,335	44.5%	22,969	44.7%	41,098	45.4%
8	1,615	8.3%	2,855	8.3%	4,306	8.7%	6,965	7.8%	15,741	8.2%	1,237	9.0%	3,048	8.3%	4,327	8.4%	7,129	7.9%
9	887	4.6%	1,466	4.3%	2,123	4.3%	3,722	4.2%	8,198	4.3%	675	4.9%	1,532	4.2%	2,122	4.1%	3,869	4.3%
10	57	0.3%	88	0.3%	127	0.3%	170	0.2%	442	0.2%	49	0.4%	90	0.2%	113	0.2%	190	0.2%
Gleason’s Score on Prostatectomy
6	6,237	30.3%	10,824	30.6%	14,162	27.9%	24,547	26.8%	55,770	28.2%	4,270	30.5%	11,795	30.8%	14,844	28.3%	24,861	26.6%
7	11,902	57.9%	20,453	57.9%	30,961	61.1%	58,056	63.4%	121,372	61.3%	7,874	56.3%	22,226	58.1%	32,014	61.0%	59,258	63.4%
8	1,099	5.3%	1,850	5.2%	2,456	4.8%	4,010	4.4%	9,415	4.8%	824	5.9%	1,928	5.0%	2,580	4.9%	4,083	4.4%
9	1,275	6.2%	2,131	6.0%	3,022	6.0%	4,819	5.3%	11,247	5.7%	981	7.0%	2,230	5.8%	2,946	5.6%	5,090	5.4%
10	48	0.2%	61	0.2%	71	0.1%	101	0.1%	281	0.1%	37	0.3%	64	0.2%	74	0.1%	106	0.1%
NCCN Risk Group Based on Clinical T
Low	5,081	17.2%	9,803	20.4%	14,201	21.0%	28,317	24.1%	57,402	21.8%	3,372	15.2%	10,122	20.0%	15,352	21.8%	28,556	23.8%
Intermediate	17,493	59.1%	27,631	57.4%	39,700	58.6%	67,675	57.5%	152,499	57.9%	13,124	59.2%	29,165	57.5%	40,976	58.1%	69,234	57.8%
High	5,956	20.1%	9,180	19.1%	12,178	18.0%	18,861	16.0%	46,175	17.5%	4,674	21.1%	9,800	19.3%	12,452	17.7%	19,249	16.1%
Very high	1,090	3.7%	1,501	3.1%	1,666	2.5%	2,875	2.4%	7,132	2.7%	999	4.5%	1,606	3.2%	1,692	2.4%	2,835	2.4%
NCCN Risk Group Based on pT
Low	1,013	2.3%	1,869	2.6%	2,592	2.4%	4,504	2.3%	9,978	2.4%	688	2.1%	2,014	2.7%	2,798	2.6%	4,478	2.2%
Intermediate	29,751	68.3%	49,346	68.7%	75,759	71.0%	136,193	69.5%	291,049	69.6%	21,520	66.6%	50,140	68.4%	75,881	70.3%	143,508	70.2%
High	9,269	21.3%	15,122	21.1%	21,096	19.8%	42,791	21.8%	88,278	21.1%	7,129	22.1%	15,566	21.2%	21,813	20.2%	43,770	21.4%
Very high	3,527	8.1%	5,488	7.6%	7,239	6.8%	12,428	6.3%	28,682	6.9%	2,955	9.2%	5,565	7.6%	7,409	6.9%	12,753	6.2%
Regional Lymph Nodes Examined
Median (Interquartile Range)	2 (0,5)		2 (0,4)		2 (0,5)		2 (0,6)		2 (0,5)		2 (0,5)		2 (0,5)		2 (0,5)		2 (0,6)
Range	(0–90)		(0–88)		(0–90)		(0–90)		(0–90)		(0–70)		(0–90)		(0–88)		(0–90)
N+	1,134	2.2%	1,534	1.8%	2,286	1.9%	5,103	2.2%	10,057	2.1%	972	2.5%	1,583	1.8%	2,193	1.8%	5,309	2.2%
Adjuvant Treatment
Adjuvant Treatment Indicated and Given	3,803	7%	5,252	6%	6,080	5%	8,886	4%	24,021	5%	3,938	10%	5,267	6%	6,387	5%	8,429	4%
Adjuvant Treatment Indicated but Not Given	14,461	27.8%	24,024	28.6%	34,747	28.2%	69,420	30.3%	142,652	29%	10,533	26.8%	24,096	28.1%	36,598	29.3%	71,425	29.9%
Adjuvant Treatment Not Indicated but Given	929	2%	1,142	1%	1,152	1%	1,480	1%	4,703	1%	852	2%	1,205	1%	1,241	1%	1,405	1%
Adjuvant Treatment Not Indicated and Not Given	32,896	63%	53,575	64%	81,227	66%	149,315	65%	317,013	65%	23,981	61%	55,156	64%	80,494	65%	157,382	66%

Multivariable Survival Analysis

The overall survival improved by higher annual caseload in the multivariable Cox proportional hazards ratios model. For FAC, there was a 30% increase in all-cause mortality risk associated with performing radical prostatectomy at a below median Volume Group (VG1) compared to a top 10 percentile Volume Group (VG4) facility- HR 1.30 95% CI: 1.23–1.36, p<0.0001. For FASC the increase in all-cause mortality risk was 25%- HR 1.25 95% CI: 1.19–1.32, p<0.0001 when VG1 and VG4 were compared.

There was no statistically significant difference in outcomes for low vs intermediate risk group patients who underwent surgery. However, high risk, very high risk and node positive disease predicted for increasingly worse outcomes with 16, 36 and 111 percent and 16, 36 and 112% increased risk of all-cause mortality based on FAC and FASC, respectively- Table 4.

Table 4.

Univariable and multivariable analysis results.

	Volume Group based on Facility Annual Caseload (FAC)			Volume Group based on Facility Annual Surgical Caseload (FASC)
Variable	Hazard Ratio	95% CI	p	Hazard Ratio	95% CI	p
Univariable Results
Caseload
Volume Group 1, <50th percentile	1.50	(1.448, 1.56)	<.0001	1.52	(1.457, 1.579)	<.0001
Volume Group 2, 50–74th percentile	1.35	(1.302, 1.392)	<.0001	1.40	(1.354, 1.446)	<.0001
Volume Group 3, 75–89th percentile	1.23	(1.194, 1.269)	<.0001	1.25	(1.217, 1.292)	<.0001
Volume Group 4, ≥90th percentile	Ref			Ref
Multivariable Results
Caseload
Volume Group 1, <50th percentile	1.30	(1.232, 1.362)	<.0001	1.25	(1.191, 1.32)	<.0001
Volume Group 2, 50–74th percentile	1.20	(1.15, 1.244)	<.0001	1.22	(1.176, 1.267)	<.0001
Volume Group 3, 75–89th percentile	1.16	(1.121, 1.199)	<.0001	1.13	(1.097, 1.171)	<.0001
Volume Group 4, ≥90th percentile	Ref			Ref
Definitive Surgical Procedure, Days from Diagnosis
2–90 days	Ref			Ref
91–365 days	1.10	(1.066, 1.129)	<.0001	1.09	(1.063, 1.125)	<.0001
>365 days	1.42	(1.188, 1.705)	0.0001	1.42	(1.184, 1.7)	0.0001
Surgical Margin Status
Negative	Ref			Ref
Positive	1.16	(1.113, 1.207)	<.0001	1.16	(1.111, 1.205)	<.0001
Unknown	1.01	(0.915, 1.12)	0.8182	1.01	(0.912, 1.116)	0.8634
NCCN Risk Group Based on pT
Low	Ref			Ref
Intermediate	1.16	(0.985, 1.361)	0.0747	1.16	(0.989, 1.366)	0.0679
High	1.41	(1.198, 1.665)	<.0001	1.42	(1.201, 1.669)	<.0001
Very High	2.22	(1.878, 2.625)	<.0001	2.23	(1.885, 2.635)	<.0001
N+	3.31	(2.786, 3.937)	<.0001	3.33	(2.805, 3.962)	<.0001
Charlson-Deyo Score
0	Ref			Ref
1	1.57	(1.52, 1.615)	<.0001	1.57	(1.522, 1.617)	<.0001
≥2	2.59	(2.437, 2.747)	<.0001	2.59	(2.437, 2.748)	<.0001
Primary Payer
Not insured	Ref			Ref
Private insurance	0.70	(0.632, 0.773)	<.0001	0.70	(0.634, 0.775)	<.0001
Medicaid	1.27	(1.116, 1.442)	0.0003	1.27	(1.113, 1.438)	0.0003
Medicare	0.87	(0.781, 0.959)	0.0057	0.87	(0.783, 0.962)	0.007
Other government	0.88	(0.759, 1.024)	0.0992	0.88	(0.76, 1.026)	0.1035
Unknown	0.75	(0.657, 0.858)	<.0001	0.75	(0.654, 0.853)	<.0001
Facility Type
Community Cancer Program	Ref			Ref
Comprehensive Community Cancer Program	1.01	(0.952, 1.065)	0.8042	0.99	(0.94, 1.048)	0.778
Academic/Research Program	0.98	(0.92, 1.042)	0.5143	0.95	(0.896, 1.01)	0.1047
Integrated Network Cancer Program	1.07	(1, 1.151)	0.0504	1.03	(0.962, 1.102)	0.3942
Other or unknown	2.54	(1.563, 4.122)	0.0002	2.47	(1.518, 4.002)	0.0003

Private insurance predicted for best outcome while Medicaid predicted for worse outcome compared to the uninsured group with 30% all-cause mortality risk reduction for private insurance based on both FAC and FASC and 26% and 27% all-cause mortality risk increase for FAC and FASC.

Overall Survival-Kaplan-Meier Curves

At a median follow up period of 60.75 months (range: 0.03–143.14) median overall survival (OS) was not reached. At 90^th percentile survival, the adjusted overall survival curves revealed a survival advantage as volume increased. When the top 10 percentile group (VG4) was compared to below median volume group (VG1) the OS advantage reached 13.2 months for FAC (OS of 93 months for VG1 and 106.2 months for VG4) and 11.3 months for FASC (OS of 94.4 months for VG1 and 105.7 months for VG4)- Figure 2. These values for unadjusted 90^th percentile OS comparing VG4 with VG1 were 21.2 and 22.2 months, respectively- Figure 3.

Figure 2.

Adjusted survival curves. Top volume groups by FAC. Bottom volume groups of FASC.

Figure 3.

Unadjusted Kaplan-Meier curves. Top volume groups by FAC. Bottom volume groups by FASC.

Sensitivity Analyses

When patients who died within 90 days after surgery were removed the analyses, the HRs remained similar across volume groups- Supplementary Table 2. The differences in OS by FAC volume groups were smaller in patients diagnosed in 2010 or earlier than those diagnosed in 2011 or later. But the HRs were similar by FASC volume groups across year of diagnosis – Supplementary Table 3. There were no significant interactions between FAC or FASC volume groups and Gleason’s scores on OS. However, the differences in OS by FAC or FASC volume groups were stronger among patients with low or intermediate NCCN risks than these with high NCCN risks P for interaction <0.001, – Supplementary Table 4.

Patients in this analysis may have received treatment at more than one facility. When the variable indicating the case was reported by more than one facility was included in the analysis, the HRs for mortality were higher for lower volume groups in both FAC and FASC volume groups, showing significant interaction with treatment at more than one facility predicting for significantly worse outcome p < 0.0001. Once again FAC was a stronger predictor of outcome. - Supplementary Table 5.

Discussion

The results of this analysis reveal an overall survival benefit associated with annual caseload of the facility where radical prostatectomy is performed. This absolute all-cause mortality risk difference was up to 30% when the two ends of the caseload spectrum were compared. Even though the median survival was not reached due to the curative nature of radical prostatectomy, there was a stark difference in outcomes between the volume groups reaching 13 months when the top 10 percentile volume centers were compared to below median volume facilities. Interestingly, when Volume Groups were defined based on facility prostate cancer caseload, which included all prostate cancer cases metastatic or non-metastatic, the survival benefit conferred by the high-volume facilities were larger than when Volume Groups were constructed by facility surgical (radical prostatectomy) caseload. This may indicate that the benefit is not entirely due to better surgery performed by individual surgeons, but also due to comprehensive case management capability.

Only the top 10 percentile facilities- designated as such by either method- had a lower rate for positive surgical margins than the average rate. As expected, facilities within the top 10 percentile of surgical caseload had a lower rate compared to facilities within the top 10 percentile of all prostate cancer caseload. The average number of nodes dissected and the node positive rates were similar across different Volume Groups. Clearly, the surgical outcomes improved by surgical experience, yet, the overall survival benefit showed a stronger correlation with the all-prostate-cancer caseload than the radical-prostatectomy caseload.

These results do not support a case-mix difference across volume groups as the explanation for the outcome difference. The mix of Gleason’s score and NCCN risk group as well as Charlson-Deyo comorbidities index were only modestly different among volume groups. Gleason’s score and NCCN risk groups significantly changed after radical prostatectomy with a net decrease in low risk patients, but the distribution of the Gleason’s scores and NCCN risk groups remained similar across volume groups.

In the multivariable analysis despite inclusion of commonly used prognostic factors, the role of annual caseload remained highly significant.

The exact factors responsible for this phenomenon remained elusive. What this study was able to do, however, was to demonstrate that the role of experience and annual caseload is much larger than previously thought and is not limited to rates of PSA recurrence or continence or transfusion. The results demonstrated- to our knowledge for the first time- that there was a survival advantage. While the decade long span of the data appears long, the treatment studied is associated with a long survival period and therefore median overall survival for patients was not reached. However, estimates of the 90^th percentile survival benefit was associated with a very large advantage of more than 1 year in favor of high-volume facilities.

As the exact cause of this difference in overall survival remains elusive, regionalization of radical prostatectomy appears to be the most viable approach. Previous studies have explored the feasibility of referrals for the Medicare population within driving distance with encouraging results⁸ and also demonstrated savings that could be used towards referral costs⁷.

Future studies should consider exploring the impact of central or regional pathology review on these outcomes as the pathology report sets the treatment plan in post-operative setting and could potentially result in undertreatment and poor outcomes.

This is study is limited by the fact that it is a retrospective analysis of registry a dataset. The variables are not comprehensive, may have missing values and do not offer a complete picture of the care provided. The role of referral and facility selection biases on the part of patients, or providers could not be examined by this dataset. The accuracy of the pathology reporting, which guides the post-operative treatment, could not be independently examined and adjusted using this dataset.

Conclusion

This study of a large population of prostate cancer patients reveals a significant and clinically important survival advantage for performing radical prostatectomies at highly experienced centers. These data support the regionalization of radical prostatectomy in an effort to improve oncologic outcomes for this patient population.

Key Points.

Question:

What is the impact of facility caseload on oncological outcome of radical prostatectomy, namely overall survival?

Findings:

In this review of 488,389 cases of radical prostatectomy, as the annual caseload of prostate cancer care at the facility decreased, the all-cause mortality for the patients increased significantly. At 90^th percentile survivorship, the adjusted overall survival difference between the top 10^th percentile and the bottom 50^th percentile of annual caseload reached 11.3 to 13.2 months (p<0.0001).

Meaning:

Caseload of the facility where radical prostatectomy was performed had a direct and significant impact on the overall survival for the patients.

Narrative:

For years there have been discussions on the role of experience and outcomes for prostate cancer. Vickers groups from MSKCC published showing the difference, mostly using biochemical failure as an endpoint. In the absence of solid survival data, the role of experience on outcomes did not impact any policy or practice.

This is an in-depth analysis of carefully modeled history of almost half a million cases of radical prostatectomies over a 10 year period and more than 1000 facilities. It does show a survival difference based on the caseload and interestingly enough experience with all stages of prostate cancer was a stronger predictor of outcomes compared to just the number of surgeries.

The point here is not to make surgeons look bad.

Far from it. In the words of a wise surgeon “How many ways can one mess up a radical prostatectomy?”

The technical part comes down mostly to nodal dissection and positive margins. Controlling for all factors captured in NCDB that could impact outcome, the effect persists and experience with all stages of prostate cancer remains a stronger predictor than just surgical volume.

This leads our team to believe the effect is interdisciplinary and the experience of the pathologist, medical oncologist and radiation oncologist also impact the outcome.

We believe this could start an import dialog. We would love to hear feedback from peer reviewers and have this published at Cancer with editorial commentary to increase awareness. We would be happy to share all aspects of analysis with peer reviewers and editors.

While we are acutely aware of all of the limitations of a retrospective NCDB analysis, we believe we should not overlook the findings here as this analysis is powered by 488,000 patients and demonstrates a 13 months OS benefit just based on where the surgery was done. Despite all of its limitations, it is something to be discussed at high levels of our community to increase awareness and to generate new ideas and research.