Treatment Facility Volume and Survival in Patients with Advanced Prostate Cancer

Abstract

Background:

Despite improvements in the medical management of advanced prostate cancer (aPC), it continues to be a leading cause of cancer death in men. The contemporary management of men with aPC is complex and requires resources more readily available at high volume facilities.

Objective:

To determine the relationship between facility volume and survival in men with aPC.

Design/Setting/Participants:

The National Cancer Database (NCDB) was queried from 2004–2013 for aPC, defined as T4, N+, or M+ disease, identifying 64,815 patients. Six pre-defined patient cohorts were evaluated. Cohort ‘A’ included all patients with aPC. ‘B’ cohorts included only M0 patients. ‘C’ cohorts included only M1 patients. Facilities were divided into quartiles based on median treatment volume (patients/year).

Exposure/Intervention:

Diagnosis and management of aPC at an NCDB-reporting facility.

Outcome & Statistical analysis:

Overall survival (OS) as a function of facility volume. Multivariable Cox Regression models were fitted. Cox regressions using natural cubic splines were used to test for non-linear relationships between volume and OS.

Results & Limitations:

OS improved as facility volume increased (top quartile vs. bottom quartile, HR 0.82, 95% CI 0.77–0.88, p<0.001), and was consistent across patient cohorts. Spline models demonstrate a continuous decrease in hazard of death as volume increases. Limitations include the retrospective analysis and lack of precise treatment information.

Conclusions:

In this retrospective analysis of nearly 65,000 men who presented with aPC, we demonstrate an association between higher facility volume and improvements in OS. This OS advantage persisted with similar magnitudes of effect after narrowing the cohorts by disease and treatment characteristics.

Patient Summary:

In this retrospective review of the NCDB, we analyzed the association between treatment facility volume and survival in men who are diagnosed with advanced prostate cancer. We found that survival improved as volume increased, indicating a possible imbalance of resources and expertise that favors higher volume facilities.

Introduction

Prostate cancer (PC) is the most common malignancy in American men, accounting for 19% of all new cancer cases in 2018.¹ Despite significant improvements in the medical, radiation, and surgical management of PC over several decades, advanced PC continues to be the second leading cause of cancer death in men.¹ Although presentation with de novo advanced/metastatic PC became less frequent after the introduction of PSA screening in the early 1990s, changes in national screening guidelines in 2012 may have increased the incidence of men presenting with advanced disease.²

The optimal management of patients who present with advanced PC is the subject of intensive research, and a growing assortment of treatment options are now available.³ Many of these options require nuanced, multi-disciplinary management and some are uniquely available through ongoing clinical trials. For localized PC and other solid tumors, numerous observational studies have found that facilities treating higher volumes of patients have better short- and long-term treatment outcomes. For prostate cancer, these associations have been demonstrated both in men who receive surgery^4,5 and in men who receive radiation⁶ as their primary treatment. However, it is unclear if this volume-outcome benefit extends to patients with advanced disease. Specifically, it is unknown if survival for men following a diagnosis of advanced PC is related to treatment facility (TF) volume.

We hypothesized that certain tertiary care, research and higher volume facilities that manage a greater number of PC patients might provide specific advantages, such as treatment availability and management expertise, that could affect survival for men with advanced PC. Differences in survival based on TF volume could have implications on the optimal management of these patients. We therefore utilized data from a large national cancer registry to study the relationship between TF volume and survival in patients with advanced PC.

Material and Methods

Data Source

The National Cancer Database (NCDB), a program of the ACS® CoC (Commission on Cancer) and the American Cancer Society, is a national cancer registry and comprehensive clinical surveillance resource for cancer care in the United States. The NCDB compiles data from over 1,500 commission-accredited cancer programs in the United States and Puerto Rico. Though not population-based, the NCDB captures approximately 70% of all newly diagnosed cancer cases in the United States.⁷ Institutional IRB approval was obtained for use of national de-identified registry data.

Study Population

Patients with advanced prostate adenocarcinoma, defined as having T4, N+, or M+ disease, were identified in the NCDB based on ICD-O-3 site codes. This corresponds to Stage IV patients using the AJCC 7^th edition and Stage III & IV patients using the AJCC 8^th edition.⁸ Our study cohort included all patients who were diagnosed with advanced primary prostate adenocarcinoma between 2004 and 2013. Patients were excluded if survival data was unavailable, or if they did not receive any treatment at the reporting facility. To thoroughly examine any association between TF volume and survival, patients were divided into six cohorts defined by varying disease and treatment characteristics (Figure 1). Cohort A included all patients with advanced PC (N = 64,815); cohort B1 was restricted to M0 patients (N = 27,155); cohort B2 was restricted to M0 patients undergoing some active treatment (radiation therapy, definitive surgical therapy, or hormone/endocrine ablation; N = 23,011); cohort C1 included all M1 patients (N = 37,660); cohort C2 was restricted to M1 patients undergoing active treatment (hormone/endocrine; N = 30,643); and cohort C3 was the most restrictive group, including M1 patients who underwent active treatment and who had known metastatic sites (bone/brain/liver/lung, N = 12,452). The purpose of these increasingly restrictive criteria was to confirm that any effects shown in the larger sample were not due to biased patient selection.

Figure 1 –

Study cohort eligibility CONSORT diagram and cohort criteria

Treatment Facility, Patient Volume, and Study Outcome

TFs were divided into quartiles based on average yearly volume, resulting in the following treatment volume ranges: <1.8 patients/year (TF N = 303), 1.8–3.3 patients/year (TF N = 298), 3.4–5.6 patients/year (TF N = 311), and >5.6 patients/year (TF N = 313). Note that these volume numbers exclude any recurrent prostate cancers; only newly diagnosed cancers are available in the NCDB. The primary outcome was overall survival (OS), defined as time from diagnosis until death or loss to follow-up.

Statistical Analysis

First, we compared the distribution of covariates of interest between volume quartiles in the full cohort using frequency tables, Chi-squared tests, and ANOVA tests. Covariates included patient age, race, Hispanic ethnicity, year of diagnosis, insurance type, income, education, location, Charlson-Deyo comorbidity score⁹, facility volume, and clinical characteristics (T/N/M stage). We did not adjust for treatments received, as we considered treatment choice to be on the causal pathway between facility volume and survival outcomes. We then determined which variables were associated with treatment at high-volume (top quartile) facilities simultaneously using multivariable logistic regression with robust standard errors to account for clustering within facility. We also analyzed differences in therapies delivered at hospitals in the highest volume quartile versus hospitals in lower quartiles.

We assessed the relationship between TF volume quartile and OS outcomes using Kaplan-Meier curves and multivariable Cox proportional hazards regressions. To determine if there was a non-linear relationship between volume and survival outcomes, we also fit Cox regressions using natural cubic splines. Degrees of freedom (5) for the spline models were chosen utilizing the full cohort using cross-validation. All regressions adjusted for the aforementioned pre-specified covariates and, once again, adjusted for clustering within facility via robust standard errors. In addition to the main cohorts, we conducted two sensitivity analyses to confirm the associations. First, we created a new cohort based on cohort A where Gleason score was known, and adjusted for Gleason score in the survival model. We also analyzed cohort A adjusting for facility type (academic, comprehensive community, or community); we did not include this in the main model as it is highly collinear with facility treatment volume.

Statistical analysis was done with SAS®, version 9.3 and R version 3.3 with p <0.05 considered statistically significant.

Results

Patient Characteristics

This patient population was predominantly Caucasian (78%), with African American men comprising 18% of the cohort. 73% of men were >60 years old, and only 5% were <50 years old. Many men received care using Medicare (48%) as their primary insurance, while 31% used private insurance or managed care plans. Notably, this cohort appeared to have good access to insurance coverage; only 5% of men used Medicaid, and only 4% were uninsured. Table 1 details patient characteristics pertinent to this study.

Table 1 –

Patient and facility characteristics of the entire cohort (Cohort A, N = 64,815)

Descriptive Groups	# of Patients	% of Patients
N	64,815
Patient/Facility characteristics
AGE
<50	3033	5%
51–60	14687	23%
61–70	22109	34%
71+	24986	39%
RACE
Caucasian	50637	78%
African American	11366	18%
Other/Unknown	2812	4%
HISPANIC
No	57151	88%
Yes	3732	6%
Unknown	3932	6%
INSURANCE
Not Insured	2853	4%
Private Insurance / Managed Care	20156	31%
Medicaid	3274	5%
Medicare	31254	48%
Other Government	707	1%
Insurance Status Unknown	6571	10%
MEDIAN HOUSEHOLD INCOME
Less than $38,000	12528	19%
$38,000 - $47,999	14923	23%
$48,000 – $62,999	16746	26%
$63,000 +	19678	30%
FACILITY VOLUME
< 1.8 pts/year	3010	5%
1.8–3.3 pts/year	7521	12%
3.4–5.6 pts/year	13460	21%
>5.6 pts/year	40824	63%
FACILITY TYPE
Community	6534	10%
Comprehensive Community	24541	38%
Academic/Research	27996	43%
Integrated Network	5694	9%
LOCATION
Large metropolitan	32831	51%
Small metropolitan	19404	30%
Suburban	6369	10%
Rural	6211	10%
Disease Characteristics
CHARLSON SCORE
0	52411	81%
1	9260	14%
2	3144	5%
T STAGE
Non-T4	43082	66%
T4	11048	17%
X	10685	16%
N STAGE
0	19985	31%
1	32484	50%
X	12346	19%
BONE METASTASES
No	14199	22%
Yes	14221	22%
Unknown	1582	2%
Treatment Characteristics
SURGERY
No surgery	41776	64%
Total Prostatectomy	13334	21%
Other/Unknown	9,705	15%
RADIATION
None	43931	68%
External Beam	17174	26%
Brachytherapy	373	1%
External Beam + Brachytherapy	273	0%
Unknown/NOS	2923	5%
CHEMOTHERAPY
No	60129	93%
Yes	2,877	4%
Unknown	1809	3%
HORMONE
No	20640	32%
Yes	42807	66%
Unknown	1368	2%
OTHER ENDOCRINE
No	61670	95%
Yes	2278	4%

Treatment and Facility Characteristics

Most (81%) patients were managed at an Academic/Research treatment facility or comprehensive cancer treatment center. Only 10% of patients were managed in community facilities. With regard to the volume of patients treated in each volume quartile, 5% were treated at TFs treating <1.8 patients/year, 12% at 1.8–3.3 patients/year, 21% at 3.4–5.6 patients/year, and the majority of men (63%) at TFs treating >5.6 patients/year.

Several clinical variables were associated with treatment at a facility in the top volume quartile (>5.6 pts/yr). These included age, race, Hispanic ethnicity, education, urban/rural location, Charlson-Deyo score, and stage (Supplementary Table 1).

Differences in treatment delivery were identified when comparing top quartile treatment facilities with all other facilities. Specifically, men with M1 disease who were treated at facilities in the highest quartile were slightly more likely to receive first line hormonal therapy than men treated at lower volume facilities (80.5% vs. 75.5%, respectively; p<0.001). Highest quartile facilities were also less likely to deliver first line hormonal therapy to M0 patients than lower volume facilities (46.8% vs. 55.1%, respectively; p<0.001). Highest quartile facilities were also less likely, in both M1 and M0 patients, to deliver first line radiation therapy. For M0 patients, the difference in radiation use was quite large (21.3% for top quartile vs. 38.1% for bottom 3 quartiles, p<0.001) (Supplementary Table 2).

Survival Outcomes

The median follow-up time for the full cohort was 60.3 months (reverse Kaplan-Meier method). Overall survival improved with each increase in volume quartile (Table 2). The top quartile (>5.6 pts/yr) demonstrated significantly improved survival compared to the bottom quartile (<1.8 pts/yr) [HR 0.82, 95% CI 0.77–0.88, p<0.001]. The improved survival in the top volume quartile remained consistent when analyzed across the six pre-defined cohorts. Table 3 summarizes the survival HRs for each cohort based on volume quartile. Figure 2 displays the unadjusted Kaplan-Meier estimates for Cohort A, divided by volume quartile.

Table 2 –

Multivariable survival analysis for entire cohort (Cohort A, N = 64,815)

Variable		H.R.	95% CI			P-value
Facility Volume	1.8–3.3 pts/year	0.97	0.91	-	1.04	0.36
	3.4–5.6 pts/year	0.95	0.89	-	1.01	0.08
	>5.6 pts/year	0.82	0.77	-	0.88	<.0001
	< 1.8 pts/year	ref
Diagnosis Year	Per 5-year increase	0.99	0.98	-	0.99	<.0001
Age	Per 10 year increase	1.38	1.36	-	1.41	<.0001
Race	African American	1.03	0.99	-	1.07	0.10
	Other/Unknown	0.83	0.77	-	0.89	<.0001
	White	ref
Hispanic	Yes	0.82	0.77	-	0.87	<.0001
	Unknown	1.02	0.95	-	1.09	0.6078
	No	ref
Charlson Score	1	1.29	1.25	-	1.34	<.0001
	2	1.82	1.73	-	1.91	<.0001
	0	ref
Location	Large metropolitan	1.00	0.95	-	1.06	0.89
	Small metropolitan	1.03	0.98	-	1.09	0.28
	Suburban	1.00	0.94	-	1.06	0.88
	Rural	ref
Income Quartile	$38,000 - $47,999	0.97	0.93	-	1.01	0.11
	$48,000 – $62,999	0.91	0.87	-	0.96	<.0001
	$63,000 +	0.84	0.79	-	0.89	<.0001
	Less than $38,000	ref
T	T4	1.58	1.51	-	1.65	<.0001
	X	1.39	1.34	-	1.45	<.0001
	Non-T4	ref
N	1	1.10	1.06	-	1.14	<.0001
	X	1.19	1.15	-	1.24	<.0001
	0	ref
Metastasis	1	3.85	3.61	-	4.10	<.0001
	1A	2.84	2.60	-	3.09	<.0001
	1B	3.96	3.73	-	4.21	<.0001
	1C	4.75	4.42	-	5.10	<.0001
	0	ref
Insurance	Status Unknown	0.56	0.29	-	1.08	0.08
	Medicaid	1.38	1.30	-	1.48	<.0001
	Medicare	1.09	1.05	-	1.13	<.0001
	Not Insured	1.34	1.26	-	1.43	<.0001
	Other Government	1.10	0.98	-	1.23	0.09
	Private Insurance / Managed Care	ref
% Without High School Degree	21% or more	1.03	0.97	-	1.09	0.29
	13% to 20.9%	1.03	0.98	-	1.08	0.22
	7% to 12.9%	1.02	0.98	-	1.06	0.26
	Less than 7%	ref

Table 3 –

Overall survival summary by cohort

Cohorts		HR (vs. <1.8 pts/year)	95% CI			P-value
Cohort A All advanced PC patients (T4 or N+ or M+)	1.8–3.3 pts/year	0.970	0.907	-	1.036	0.3643
	3.4–5.6 pts/year	0.947	0.890	-	1.007	0.0839
	>5.6 pts/year	0.821	0.770	-	0.876	<.0001
Cohort B1 M0 patients	1.8–3.3 pts/year	0.849	0.735	-	0.981	0.0265
	3.4–5.6 pts/year	0.827	0.723	-	0.946	0.0055
	>5.6 pts/year	0.698	0.612	-	0.796	<.0001
Cohort B2 M0 patients who underwent active treatment	1.8–3.3 pts/year	0.938	0.799	-	1.101	0.4303
	3.4–5.6 pts/year	0.868	0.746	-	1.009	0.066
	>5.6 pts/year	0.738	0.637	-	0.854	<.0001
Cohort C1 M1 patients	1.8–3.3 pts/year	0.977	0.915	-	1.043	0.491
	3.4–5.6 pts/year	0.964	0.907	-	1.024	0.2323
	>5.6 pts/year	0.839	0.787	-	0.895	<.0001
Cohort C2 M1 patients who underwent active treatment	1.8–3.3 pts/year	1.001	0.930	-	1.077	0.9802
	3.4–5.6 pts/year	0.998	0.933	-	1.068	0.9638
	>5.6 pts/year	0.870	0.813	-	0.931	<.0001
Cohort C3 M1 patients who underwent active treatment, and with known metastatic sites	1.8–3.3 pts/year	0.987	0.872	-	1.117	0.8346
	3.4–5.6 pts/year	1.026	0.916	-	1.149	0.6541
	>5.6 pts/year	0.864	0.773	-	0.966	0.0099
Cohort A Sensitivity analysis adjusted for Gleason score	1.8–3.3 pts/year	0.978	0.906	-	1.056	0.5747
	3.4–5.6 pts/year	0.940	0.874	-	1.012	0.0993
	>5.6 pts/year	0.812	0.754	-	0.875	<.0001
Cohort A Sensitivity analysis adjusted for facility type	1.8–3.3 pts/year	0.966	0.901	-	1.035	0.3245
	3.4–5.6 pts/year	0.951	0.887	-	1.019	0.1512
	>5.6 pts/year	0.855	0.791	-	0.925	<.0001

Figure 2 –

Unadjusted Kaplan-Meier survival estimates based on volume quartile of treatment facility.

Sensitivity analyses were conducted on Cohort A, adjusting for Gleason score and facility type, which are variables likely to directly influence survival. OS remained significantly better in the top volume quartile after inclusion of Gleason score (HR 0.81, 95% CI 0.75–0.88, p<0.001) and facility type (HR 0.86, 95% CI 0.79–0.93, p<0.001) in these sensitivity analyses.

Other significant variables of note in the multivariable model included age (HR 1.38, 95% CI 1.36–1.41 per 10-year increase, p<0.001), Hispanic ethnicity (HR 0.82, 95% CI 0.77–0.87, p<0.001), and comorbidity score (CCI 1: HR 1.29, 95% CI 1.25–1.34; CCI 2: HR 1.82, 95% CI 1.73–1.91, both p<0.001). Median household income also appeared to play a role in OS, with each increase in income quartile associated with improvement in survival, despite the high insured rate in this population.

Results of the spline models show that in the combined stage IV cohort, any increase in volume is associated with reduced hazard of death (Figure 3a). This was also true within the M0 cohorts (Figure 3b). However, in the M1 cohorts the largest benefits occurred in facilities seeing >10 newly-diagnosed advanced PC cases/year (Figure 3c).

Figure 3 –

Adjusted spline models demonstrating changes in the hazard ratio for survival as a function of TF volume.

Figure 3a – Effect of volume on HR for death for all Stage IV (AJCC 7^th ed) patients

Figure 3b – Effect of volume on HR for death for all M0 patients

Figure 3c – Effect of volume on HR for death for all M1 patients

Discussion

In this retrospective analysis of nearly 65,000 men who presented with advanced prostate cancer, we demonstrate that management at a high-volume facility (top quartile, >5.6 pts/yr) confers a significant survival advantage when compared to management at a facility in the lowest quartile (<1.8 pts/yr). This survival advantage persisted with similar magnitudes of effect after narrowing the cohorts by disease and treatment characteristics (Cohorts B and C). Results of the spline models show that higher TF volume results in a lower hazard of death for advanced PC, with a benefit that increases continuously with increasing volume. In total, these findings imply that there may be differences in the management of advanced PC between lower- and higher-volume TFs that affect survival. These findings are consistent with data demonstrating survival advantages following management at high-volume TFs in other systemic/advanced disease states, such as multiple myeloma¹⁰ and metastatic renal cell carcinoma.¹¹

In this cohort, we were surprised to note that African American men and other racial minority men were more likely to be treated at top-quartile facilities (HR 1.28, and 1.38, respectively, see Supplementary Table 1). We presume that this finding may be artifactual given the low numbers of minority men in the overall cohort (combined ~20%). However, if minority men do get treated more frequently at higher volume facilities, then explaining outcome disparities for these men will require further research. Interestingly, insurance status did not appear to affect the likelihood of treatment at a top-quartile facility.

In localized PC, evidence has consistently demonstrated a volume-outcome relationship that benefits patients treated at higher volume facilities or by higher volume surgeons.^5,12 Some examples of these benefits include decreased postoperative complications, length of stay, readmission risk, and lower hospital charges. Large observational series using national databases, both in the United States and Europe, have also shown an association between TF volume and survival for men treated with localized PC, though these mortality statistics usually refer to in-hospital or short-term mortality related to treatment.^5,13–15 The volume-outcome relationship for oncologic procedures, and for radical prostatectomy in particular, has led to initiatives that advocate for volume-based referral to improve quality of care.^16–18

Although the treatment landscape for advanced PC is no longer limited to hormone ablation and chemotherapy, androgen deprivation therapy (ADT) followed by docetaxel for castration-resistant disease were the standard-of-care for systemic treatment within the majority of the time period captured by this NCDB analysis. Provider-specific variations in the timing of ADT and chemotherapy may be one reason for differences in survival, but the mechanisms driving the survival disparity found in this study are likely multifactorial and more complex. Notably, however, we did note differences first-line treatments between top quartile facilities and all other facilities, specifically with regard to the use of hormonal therapy and radiation therapy. We cannot conclude, however, that these differences played a part in the outcome disparities noted in this study.

The study period did overlap with the development of newer agents such as abiraterone acetate, enzalutamide, sipuleucel-T, cabazitaxel, and radium-223. All of these agents have demonstrated survival benefits for patients with advanced PC^19–23, and some of the patients in our study cohort were likely participants in these pivotal clinical trials over the last decade. Indeed, recruitment to clinical trials, though present at varying degrees in all TFs, is likely more robust at higher volume centers. This, in part, may contribute to the improved survival seen at high-volume TFs in our cohort.

Additionally, thoughtful multi-disciplinary care of men with advanced PC may be better coordinated in large volume centers, enabling individualized treatment decisions based on multiple clinical factors. Techniques such as local treatment of oligo-metastatic disease^24,25, utilization of novel imaging modalities, and salvage surgical strategies are rapidly becoming more common. When and how to utilize the above is often a matter of judgment that only improves with practice.

The management of patients undergoing systemic treatment beyond ADT requires infrastructure that supports frequent treatment, available nursing services, financial assistance, and prompt treatment- and disease-related medical management. This infrastructure can be difficult to create and maintain at lower-volume TFs, and it has been theorized that facilities that meet volume thresholds (such as Leapfrog thresholds) may have inherent structural and procedural elements that promote improved outcomes.²⁶ While there have been calls for regionalization of care in localized PC, it is uncertain if such regionalization would be reasonable for the management of advanced PC. Concentrating care to fewer facilities could raise unique difficulties and would demand the reallocation of funds and resources to those facilities, causing strain on entire care networks. Treatment regionalization can also dissuade some patients from traveling long distances to obtain care, can lead to treatment delay, and can potentially exacerbate existing disparities in outcomes by socioeconomic status or race.^27,28

Improving outcomes at low volume TFs is difficult and requires a multidisciplinary approach to managing patients with advanced PC. Although high volume TFs enjoy certain benefits, these advantages can be shared with centers that have less robust clinical expertise. This includes facilitating consultations to obtain expert advice, creating pathways for inter-hospital patient co-management, and sharing access to resources available at centers of excellence. Experienced centers can also help navigate patients, through these conduits, to clinical trials available for patients with advanced PC.

There are several important limitations to this study. First, there are inherent limitations to using the NCDB as a tool to study patient outcomes, a topic which has previously been described in depth.²⁹ This study is retrospective in nature and is limited to specific patient, treatment, and facility characteristics collected by the NCDB. The NCDB does not provide cancer-specific mortality data, which limits interpretation. However, we feel it is ultimately OS that is most relevant in men with advanced prostate cancer. The database also does not include specific information on which systemic or hormonal therapy regimen was used for each patient, which limits our ability to determine if specific treatments or treatment sequences are responsible for the observed volume-outcome associations. Provider-level treatment volume is known to affect treatment outcomes in localized PC^5,30, though this association is not well-established for systemic treatment for advanced PC. The NCDB does not include provider-specific data that could be used to test this association. Additionally, our definition of volume was limited to the number of newly diagnosed advanced PC cases reported by an institution. We assume that this is reflective of the overall volume of advanced PC cases (including recurrent cases), but we are unable to test this assumption. As in any observational study, unmeasured confounders could have biased the association between TF volume and survival. To address confounding as much as possible, we adjusted for measured covariates and used various cohort definitions to select for patients whose disease and treatment characteristics were more homogeneous. Using this approach, we saw that the volume-outcome relationship established for the entire cohort remained consistent across all cohorts and models.

Conclusion

In this NCDB review of nearly 65,000 men who presented with advanced PC, management at a high-volume TF was significantly associated with better OS, with the effect on OS increasing continuously as a function of TF volume. This association remained true even after narrowing the cohort to select for more homogeneous groups with regard to disease and treatment characteristics. Although the precise underlying mechanism for this association is not known, these findings may be used by both clinicians and researchers to optimize treatment strategies for men with advanced PC.

Supplementary Material

1

Supplementary Table 1 - Patient and facility characteristics associated with treatment at a high-volume (top quartile) facility (N = 64,815)

Supplementary Table 2 – Differences in delivery of first-line treatment between top volume quartile treatment facilities and bottom 3 quartile facilities.