Hospital surgical volume and perioperative mortality of pelvic exenteration for gynecologic malignancies

Abstract

Background and Objectives:

To examine the association between hospital surgical volume and perioperative mortality of pelvic exenteration performed for gynecologic malignancies.

Methods:

A population-based retrospective study utilizing the Nationwide Inpatient Sample was conducted to examine pelvic exenteration for gynecologic malignancies from 2001 to 2011. Annualized hospital surgical volume was defined as the average number of procedures a hospital performed per year in which at least one case was performed, and this was correlated to perioperative mortality.

Results:

A total 1912 exenterations performed at 181 centers were included. Nearly two thirds of exenteration-performing centers had a minimum surgical volume of one case per year (121 centers, 66.9%). Perioperative mortality rate was 1.8%. In multivariable analysis surgical volume remained an independent factor for perioperative mortality (adjusted-odds ratio 0.21; 95% confidence interval, 0.09–0.49; P < .001). Perioperative mortality rates were 3.7% for the centers with minimum surgical volume (1 exenteration a year), 1.4% for the centers performing more than one but two or less exenterations a year, and 0% for the top decile centers (>2 exenterations a year), respectively (P < .001).

Conclusion:

Pelvic exenteration for gynecologic malignancy is a rare surgical procedure with most hospitals performing few cases annually. A higher surgical volume of pelvic exenteration was associated with lower perioperative mortality.

1 ∣. INTRODUCTION

Since the initial report of the surgical volume-outcome relationship was published in 1979, it has been examined for various surgical procedures over the past several decades.^1,2 The fundamental hypothesis of studying the volume-outcome relationship is that a larger hospital surgical volume may be associated with decreased both morbidity and mortality. This may help identify procedures in which regionalizing surgical services may improve outcomes.² Although the volume-outcome relationship can be difficult to assess, volume-outcome relationship have been observed in several highly complex and rare surgical procedures (Table S1).^3-14 The volume-outcome relationship has been demonstrated for cranial surgery, cardiothoracic surgery, transplant surgeries (cardiac, lung, pancreatic, and liver), and pancreatico-duodenectomy.

Pelvic exenteration is a rare and complex radical surgery.^15,16 Pelvic exenteration refers to the en-bloc removal of the visceral pelvic organs that contain tumors, with or without the perineum.^15,16 Pelvic exenteration is typically performed for recurrent cervical, uterine, vaginal, and vulvar cancers located in the central pelvis.¹⁵ Due to the extent of the procedure, pelvic exenteration is a highly morbid surgical procedure with complication rates quoted as high as 51% to 88% and a nearly 40% rate of multiple complications.^17-20 Surgical mortality of this procedure is high and has been reported as 2% to 7%.^20-26

To date, the volume-outcome relationship has not been examined for pelvic exenteration for gynecologic malignancy. This surgery is intended to be curative for women with recurrent gynecologic malignancy. Other nonsurgical options in this setting are generally considered to be only palliative (40%-52% for recurrence-free survival, and 32%-47% for overall survival),^{17,18,21,23,27-29} so identifying any predictors for improved surgical outcome will be useful to maximize the benefit for women who undergo this procedure. Given the rarity of pelvic exenteration, we hypothesized that a larger surgical volume will be associated with better outcomes. The objective of the study was to examine the association between hospital surgical volume and perioperative mortality of pelvic exenteration performed for gynecologic malignancies.

2 ∣. MATERIALS AND METHODS

2.1 ∣. Data source

The Nationwide Inpatient Sample (NIS) is a publicly available, deidentified population-based data platform that is distributed as part of the Healthcare Cost and Utilization Project (HCUP) by the Agency for Healthcare Research and Quality.³⁰ This database includes hospital discharge data for more than 36 million hospitalizations annually, covering more than 90% of the US population when weighted. It provides patient characteristics and resource-use information, such as diagnosis and intervention types, length of stay, and hospital charges, as well as hospital-specific data, including location, bed size, and teaching status. The University of Southern California Institutional Review Board deemed the study exempt due to the use of publicly available deidentified data.

2.2 ∣. Study design and eligibility

Women with gynecological cancer (cervical, uterine, vaginal, and vulvar cancers) who underwent pelvic exenteration from 2001 to 2011 were included in the study. The International Classification of Disease 9th revision (ICD-9) code of 68.8 was used as the surrogate for pelvic exenteration as described previously.^16,20,31 During the study period the ICD-9 for all the covariates in the study remained the same. This study time period was chosen because the NIS program randomly captured approximately 20% of the US hospitals each year and all the consecutive inpatient admissions within the chosen hospitals were recorded during the study period. Cases after 2011 were not utilized as the NIS program changed the data capture mechanism thereafter. Patients who underwent pelvic exenteration for ovarian and fallopian tubal cancers were excluded.²⁰

2.3 ∣. Clinical information

Information abstracted from the database included: patient baseline demographics, disease factors, hospital information, surgical procedure types, and surgical outcomes. Patient demographics included age (<50, 50-69, and ≥70), calendar year of exenteration (2001-2004, 2005-2008, and 2009-2011), race/ethnicity (white, black, Hispanic, and others), medical comorbidities, obesity (yes vs no), median household income (<$39 000, $39 000-$47 999, $48 000-$62 999, and ≥$63 000), and primary expected payer (Medicare, Medicaid, private insurance, and others). Obesity was determined by the ICD-9 coding, defined as body mass index of ≥30 kg/m² per the CDC classification.³²

For medical comorbidities, the Charlson Comorbidity Index was calculated for each patient based on the codes for the specified medical conditions in each category and weighted appropriately to calculate a final score as described previously (Table S2).³³ Disease factors included gynecological cancer type (cervical, uterine, vaginal, and vulvar) and prior radiotherapy (yes vs no). Operative information included performance of lymphadenectomy at exenteration (yes vs no), colostomy (yes vs no), urinary diversion (yes vs no), and vaginal reconstruction (yes vs no). Hospital data included hospital bed size (small, medium, and large), teaching status (rural, urban nonteaching, and urban teaching), and hospital region (Northeast, Midwest, South, and West). Hospital bed size was determined by the HCUP rule per hospital geographic region, urban-rural designation, and teaching status.³⁰

For surgical outcomes, perioperative complications for the index admission were recorded. Perioperative complications included both intraoperative and postoperative complications before hospital discharge. Perioperative complications were defined as the presence of any of the following as described previously: hemorrhage, shock, wound complications, thromboembolism, cerebrovascular disease or stroke, cardiac failure, myocardial infarction, pneumonia, respiratory failure, systemic inflammatory response syndrome (SIRS) or sepsis, ileus or small bowel obstruction, vascular injury, acute kidney injury, pyelonephritis, abscess, fistula, intestinal perforation, position-dependent complications, and death during the index admission (Table S2).^20,34 High-risk complications included shock, respiratory failure, sepsis/SIRS, and thromboembolism based on a prior analysis for mortality risk.²⁰

2.4 ∣. Statistical consideration

The primary objective of analysis was to examine the association of annualized hospital surgical volume for exenteration and perioperative mortality. The secondary objective was to examine perioperative complications, patient characteristics, and surgical procedures per annualized hospital surgical volume for exenteration.

Annualized hospital surgical volume for exenteration was defined as the average number of procedures a hospital performed per year in which at least one case was performed.³⁵ In this study annualized hospital surgical volume was analyzed as a continuous variable. Generalized estimating equations were fitted to examine the association between the annualized hospital surgical volume and patient demographics.

A binary logistic regression model was fitted to identify the independent factors for perioperative death. A threshold of a P < .10 was used for the initial step of covariate selection, and least significant covariate was removed to retain the covariates with a P < .05 in the final stopping model (conditional backward method). Effect size was expressed with odds ratio (OR) and 95% confidence interval (CI). The rationale of this approach was due to the relatively small sample size. The Hosmer-Lemeshow test was used to assess the goodness of fit in the multivariable model, and a P > .05 was interpreted as a good-fit in the model.

A scatter plot diagram was constructed to assess the association of annualized hospital surgical volume and the number of perioperative complications. Curve estimation was tested for various models, and the model which exhibited the largest statistical significance was chosen for interpretation. Cutoffs for annualized hospital surgical volume were automatically determined per the computation. A linear regression model was fitted to assess the association between annualized hospital surgical volume and the number of perioperative complications in the identified segment for the most fitted curve pattern.

In a sensitivity analysis, we have used clinically relevant cutoffs for annualized hospital surgical volume for exenteration. Centers were grouped as minimum volume centers defined as average one exenteration per year or high-volume centers defined as top decile of annualized hospital surgical volume adopted. Centers that do not meet as minimum-/high-volume criteria were grouped as mid-volume. The concept of top decile was adopted from recent studies demonstrating improved outcome in complex surgeries.^8,36

The surgical volume definition was determined in unweighted model, and the remaining analyses were performed in weighted models. All the statistical analyses were carried out with weighted models. A variance inflation factor was used to assess multicollinearity between the covariates, and a value of ≥2.5 was interpreted as multicollinearity in this study. All hypotheses were two-sided, and a P < .05 was considered statistically significant. Statistical Package for Social Sciences (version 25.0, Armonk, NY) was used for all the analyses. The STROBE guidelines were consulted to outline this observational cohort study.³⁷

3 ∣. RESULTS

A total 1912 exenterations from 181 centers performed from 2001 to 2011 were identified. The median of annualized exenteration volume was average one exenteration per year (Table 1). The majority of study sites had a minimum exenteration volume with an average of one case per year (121 [66.9%] out of 181 centers), and this group performed 683 (35.7%) exenterations during the study period. There were 17 (9.4%) centers in the top decile surgical volume group, and this group performed average more than two exenterations per year accounting for 594 (31.1%) of the exenterations for the study population.

TABLE 1.

Annualized hospital surgical volume for exenteration between 2001 and 2011

Annualized SV	Centers	Proportion (%)
1.0	121	66.9
1.1-2.0	43	23.8
2.1-3.0	12	6.6
>3.0	5	2.7
Total	181	100.0

Note: The median of annualized SV was one exenteration per year that was used to distinguish the low-volume and the mid-volume centers. Annualized SV of more than two exenterations per year represented top decile of annualized SV that was defined as the high-volume centers. Abbreviation: SV, hospital surgical volume for exenteration.

The mean age at surgery was 56.4 (± 13.0) years, and the majority of patients were white (n = 1201, 62.8%), nonobese (n = 1763, 92.2%), privately insured (n = 836, 43.7%), and had at least one comorbidity (Charlson Index ≥ 1; n = 1076, 56.3%; Table 2). The most common disease type was cervical cancer (n = 879, 46.0%) followed by vaginal cancer (n = 549,28.7%). The majority of exenterations were performed at centers with a large bed capacity (n = 1476, 77.2%) and in an urban teaching setting (n = 1630, 85.3%). A recent year of diagnosis, black race, private primary payer, centers with larger bed capacities, urban or teaching centers, and geographic areas with Midwest, South, and West, were the factors associated with higher annualized hospital surgical volume whereas higher comorbidity was associated with a lower surgical volume (all, P < .05; Table 2).

TABLE 2.

Patient demographics and association for annualized hospital surgical volume

Characteristic	No	OR (95%CI)	P value
Age,y	56 (IQR, 47-65)
<50	601 (31.5%)	1
50-69	975 (51.0%)	0.78 (0.60-1.01)	.058
≥70	335 (17.5%)	0.82 (0.58-1.17)	.273
Year
2001-2004	518 (27.1%)	1
2005-2008	818 (42.8%)	1.16 (0.87-1.55)	.310
2009-2011	575 (30.1%)	1.33 (1.05-1.69)	.020
Race/ethnicity
White	1201 (62.8%)	1
Black	128 (6.7%)	1.32 (1.04-1.66)	.023
Hispanic	145 (7.6%)	1.12 (0.76-1.64)	.583
Others	92 (4.8%)	1.70 (1.05-2.74)	.030
Missing	345 (18.0%)	1.76 (0.87-3.57)	.119
Obesity
No	1763 (92.2%)	1
Yes	149 (7.8%)	1.04 (0.68-1.59)	.873
Charlson indexa
0	854 (44.7%)	1
1-2	606 (31.7%)	0.88 (0.67-1.14)	.877
3-5	295 (15.4%)	0.95 (0.67-1.33)	.756
≥6	157 (8.2%)	0.65 (0.47-0.91)	.011
Median household income
< $39 000	435 (22.8%)	1
$39 000-$47 999	487 (25.5%)	0.78 (0.36-1.69)	.535
$48 000-$62 999	496 (25.9%)	0.63 (0.30-1.35)	.235
≥ $63 000	419 (21.9%)	0.58 (0.27-1.22)	.151
Missing	75 (3.9%)	0.56 (0.26-1.20)	.136
Primary expected payer
Medicare	516 (27.0%)	1
Medicaid	413 (21.6%)	1.42 (0.97-2.09)	.069
Private including HMO	836 (43.7%)	1.63 (1.13-2.36)	.009
Others/missing	146 (7.6%)	1.56 (1.11-2.21)	.012
Hospital bed size
Small	132 (6.9%)	1
Medium	287 (15.0%)	1.40 (1.19-1.66)	<.001
Large	1476 (77.2%)	1.70 (1.50-1.92)	<.001
Missing	16 (0.8%)	1.35 (0.98-1.87)	.068
Hospital teaching status
Rural	14 (0.8%)	1
Urban nonteaching	251 (13.1%)	1.54 (1.36-1.74)	<.001
Urban Teaching	1630 (85.3%)	2.08 (1.96-2.22)	<.001
Missing	16 (0.8%)	1.69 (1.24-2.29)	.001
Hospital region
Northeast	255 (13.3%)	1
Midwest	410 (21.4%)	1.93 (1.38-2.71)	<.001
South	724 (37.9%)	1.98 (1.43-2.73)	<.001
West	523 (27.3%)	1.44 (1.09-1.90)	.011
Cancer type
Cervical	879 (46.0%)	1
Uterine	275 (14.4%)	0.79 (0.61-1.02)	.073
Vaginal	549 (28.7%)	1.07 (0.81-1.41)	.652
Vulvar	209 (10.9%)	1.10 (0.70-1.73)	.691
History of radiotherapy
No	1753 (91.7%)	1
Yes	159 (8.3%)	0.86 (0.67-1.36)	.802
Lymphadenectomy
No	1023 (53.5%)	1
Yes	889 (46.5%)	1.23 (0.98-1.55)	.076
Colostomy
No	757 (39.6%)	1
Yes	1154 (60.4%)	1.08 (0.86-1.36)	.487
Urinary diversion
No	542 (28.4%)	1
Yes	1369 (71.6%)	1.22 (0.96-1.55)	.102
Vaginal reconstruction
No	1481 (77.5%)	1
Yes	431 (22.5%)	1.44 (1.07-1.95)	.017

Note: Median (interquartile range) or number (percentage per column) is shown. Generalized estimating equation models for P value. Abbreviations: CI, confidence interval; IQR, interquartile range; OR, odds ratio.

^aIndex score for cancer was not included. Total number may not 1912 due to weighted values or due to missing number.

For the entire cohort, nearly half underwent lymphadenectomy at the time of exenteration (n = 889, 46.5%). Rates of accompanying surgeries included: colostomy in about 60% (n = 1154, 60.4%), urinary diversion in approximately 70% (n = 1369, 71.6%), and vaginal reconstruction in roughly 20% (n = 431, 22.5%). Centers performing vaginal reconstruction were more likely to have a higher annualized hospital surgical volume (P = .017). Moreover, albeit statistically non-significant, centers performing lymphadenectomy or urinary diversion at exenteration were more likely to have higher annualized hospital surgical volume (Table 2).

There were 34 (1.8%, 95%CI, 1.2-2.4) deaths during the index admission for pelvic exenteration. In univariable analysis, higher annualized hospital surgical volume was significantly associated with lower perioperative mortality risk (OR, 0.29; 95%CI, 0.15-0.58; P < .001). In multivariable analysis (Table 3), annualized hospital surgical volume remained an independent factor for perioperative mortality, and for every one increase in surgical volume there was an associated decrease, approximately 80%, in perioperative mortality risk (adjusted-OR per surgical volume 0.21, 95%CI, 0.09-0.49; P < .001). More specifically, perioperative mortality rates were 3.7% for the centers with a minimum surgical volume (1 exenteration a year), 1.4% for the centers performing more than 1 but ≤2 exenterations a year, and 0% for the top decile centers for surgical volume (>2 exenterations a year), respectively (P < .001; Figure 1).

TABLE 3.

Multivariable analysis for perioperative mortality

Characteristic	Adjusted-OR (95%CI)	P value
Age	1.12 (1.06-1.18)	<.001
Race/ethnicity		.002 *
White	1
Black	16.6 (3.95-69.3)	<.001
Others	1.24 (0.22-6.96)	.810
Missing	0.55 (0.14-2.10)	.381
Charlson Index	1.87 (1.55-2.27)	<.001
Hospital bed size		<.001 *
Small	1
Medium	0.36 (0.09-1.38)	.136
Large	0.05 (0.02-0.18)	<.001
Missing	na	.998
Hospital region
Northeast	1.59 (0.44-5.81)	.480
Midwest	0.16 (0.04-0.63)	.009
South	0.17 (0.04-0.63)	.008
West	1
Annualized surgical volume	0.21 (0.09-0.49)	<.001

Note: A binary logistic regression model for multivariable analysis (conditional backward method). All the covariates with P < .10 in univariable analysis were entered in the initial model except for hospital teaching status that showed a multicollinearity for hospital bed size. Conditional backward method was used to retain only the covariates with P < .05 in the final model. All the listed covariates were entered in the final model. Hosmer-Lemeshow test, P = .869, indicating a good-model. Abbreviations: CI, confidence interval; OR, odds ratio.

^*P value for interaction. Significant P values are emboldened.

FIGURE 1.

Exenteration mortality rate per annualized hospital surgical volume. the Chi-square test for P value. Perioperative mortality rate is shown per annualized hospital surgical volume for exenteration. Observed values with 95% confidence intervals are displayed

More than two thirds of women who had exenteration had at least one perioperative complication (n = 1334, 69.8%), and nearly 40% had multiple complications (n = 763, 39.9%). The median number of perioperative complication per exenteration was two (interquartile range, 1-3). Among the tested curve estimations, the cubic curve pattern most fits to the study cohort (Figure 2). The automated computation identified the two reflection points for annualized hospital surgical volume as 1.8 and 4.0. Between the annualized hospital surgical volume of 1.8 and 4.0, the number of perioperative complications significantly decreased by 24% per every increase in surgical volume of one (y = −0.24x + 2.24; P = .002).

FIGURE 2.

Association between annualized hospital surgical volume and extents of perioperative complications. The most fitted curve between the extent of complications and annualized hospital surgical volume was cubic pattern. There was a significant decreasing association for the number of perioperative complication between 1.8 and 4.0 for annualized hospital surgical volume (y = −0.24x + 2.24, P = .002). ASV, annualized hospital surgical volume

There were 462 (24.2%) exenterations that had high-risk perioperative complication. Women who developed any of the high-risk perioperative complications were significantly more likely to die during the index admission compared to those who did not (5.8% vs 0.5%; OR, 12.8; 95%CI, 5.53-29.6; P <.001). Between the annualized hospital surgical volume of 1.8 and 4.0, the number of high-risk perioperative complications significantly decreased per each surgical volume increase of one (y = −0.19x + 0.82; P < .001).

4 ∣. DISCUSSOIN

The two main results of this study are that (a) pelvic exenteration for gynecologic malignancies is a rare surgical procedure with nearly two thirds of exenteration-performing centers having a surgical volume of only one case per year and (b) there was volume-outcome relationship between hospital surgical volume for pelvic exenteration and perioperative mortality risk with higher surgical volume being associated with lower surgery-related death.

Lower rates of perioperative death after pelvic exenteration in the higher surgical volume centers are encouraging findings in this study. Specifically, every increase in hospital surgical volume of one was associated with an approximately 80% decrease in perioperative mortality risk. The top decile of exenteration-performing centers performed nearly 30% of the surgeries during the study period and notably, performed more associated invasive surgical procedures including lymphadenectomy, urinary diversion, and vaginal reconstruction. Nevertheless, the higher volume centers had lower perioperative morbidity and mortality.

The surgical mortality rate at the minimum volume centers was considerably increased, and one in 27 women who underwent this procedure died after pelvic exenteration in this group. This is a clear difference, as none of the women died after pelvic exenteration in the high-volume group. Historically, the surgical mortality of pelvic exenteration was 6% to 7% in the late-1980s, and has decreased to approximately 2% to 4% in more recent years.^20-26 While the overall surgical mortality of pelvic exenteration has improved over the past several decades, the actual surgical mortality rate significantly varies across centers related to surgical volume and even recently, has ranged from 0% to nearly 4%.

The reason for improved surgical mortality in the higher surgical volume centers is unknown but it is likely that the lower incidence of critical perioperative complications such as shock, sepsis/SIRS, and respiratory failure in the higher surgical volume centers may affect this association. These complications possess significantly higher mortality risk (5.8%), and the reduction in these complications in the higher surgical volume centers would be the main factor for minimizing the mortality risk.

Another causality may be the difference in failure to rescue. That is, higher surgical volume centers have a higher capability to salvage the patients who developed perioperative complications. Indeed, our results showed that the association between the exenteration volume and the number of perioperative complication was not a linear inverse pattern throughout. As mentioned earlier, these high surgical volume centers do more associated invasive surgical procedures. Thus, it is likely that high surgical volume centers have a large bed capacity and are teaching hospitals where the infrastructural support is in place to appropriately treat these complications, resulting in minimum mortality risk.

The majority of centers in this study performed few pelvic exenterations annually. Even in high surgical volume centers, with case volumes representing the top 10th percentile among the exenteration-performing centers, the median surgical volume was three cases per year. A recent study from a center with an average of eight exenterations performed per year, reported a relatively low number of exenterations per individual surgeon (median, 6 per 21 years).³⁸ In comparison to the definition of high surgical volume centers for other surgical procedures (Table S1), these statistics for pelvic exenteration are low.^3-14 Collectively, these data point out that pelvic exenteration for gynecologic malignancies is a rare surgical procedure.

Based on our observation, regionalizing surgical service for pelvic exenteration to large bed capacity urban teaching centers may be beneficial for the reduction of surgical morbidity and mortality. Our study suggests an incremental improvement in perioperative outcomes for every additional exenteration performed per year.² Whether or not a minimum volume standard should be applied to pelvic exenteration in gynecologic malignancies merits investigation. Our analysis showed that the number of perioperative complications started decreasing at an annualized hospital surgical volume of 1.8 cases. Thus, this number might be a possible target goal for improvement of perioperative outcomes.

We recognize a number of important limitations. First, this is a retrospective study and there may be several unmeasured factors which confound the analysis. For instance, as the NIS database is designed per the codes rather than for specific research purposes, the exact indication for exenteration was not available and it is unknown if the exenteration was performed with a therapeutic vs a palliative intent. Other salient information missing in the database that also likely affects the outcomes includes patient performance status, frailty, nutritional status, operative details (surgical duration and estimated blood loss), type of surgeon, particularly for reconstruction (gynecologic oncologist vs other subspecialists), and hospital's quality of perioperative care.

Second, due to the lack of specific ICD-9 codes for exenteration, distinguishing the exact type of exenteration (total, anterior, and posterior) was not feasible for this study. This information would significantly impact perioperative outcomes. Third, this study only examined hospital surgical volume, and it is unknown if surgeon skills and experiences impacted outcomes. One prior study found no association between surgeon’s surgical volume for pelvic exenteration and oncologic outcome, but, their sample size was relatively small (n = 167).³⁸ Fourth, the NIS database captures perioperative complications only for the inpatient admission, and complications after discharge are not able to be assessed. Similarly, long-term oncologic outcomes are not available in this database but are alternative key endpoints of these surgical procedures. Finally, this study examined the US population only, and the generalizability in different patient populations is unknown.

In summary, pelvic exenteration for gynecologic malignancies is a rare and complex radical surgery with high morbidity and mortality. Lower perioperative mortality in higher surgical volume centers is a promising finding for women with gynecologic malignancies who undergo pelvic exenteration. Further study is warranted to identify whether a high surgical volume for pelvic exenteration is associated with improved oncologic outcomes.

DATA AVAILABILITY STATEMENT

The Nationwide Inpatient Sample is a publicly available, deidentified population-based data platform that is distributed as part of the Healthcare Cost and Utilization Project by the Agency for Healthcare Research and Quality. Information is available at https://www.hcup-us.ahrq.gov/.