A Prediction Model and Risk Calculator for Surgical Mortality and Serious Morbidity After Primary Cytoreductive Surgery

Abstract

Objectives:

Primary cytoreductive surgery (PCS) has been the standard initial management for ovarian cancer. In addition to predicting the likelihood of optimal debulking, predicting mortality and serious morbidity after PCS should be considered when deciding between PCS and neoadjuvant chemotherapy (NACT). The goal of this research was to develop a model to predict mortality and serious morbidity after PCS.

Materials and Methods:

This was a retrospective cohort study. Patients who underwent PCS were identified in the National Surgical Quality Improvement Program database (NSQIP) from 2006 to 2012. Outcomes collected included 30-day mortality and serious morbidity (wound dehiscence, abscess, sepsis, septic shock, stroke, coma, myocardial infarction, cardiac arrest, pulmonary embolism, unplanned intubation, and ventilator dependence for more than 48 hours). A backward selection procedure was utilized to identify model parameters. The proposed model (METRICS: Mortality sErious morbidiTy ovaRIan Cancer Surgery) was internally validated.

Results:

Of the cohort (N = 2262), 194 (8.6%) patients with the outcome (30-day mortality and serious morbidity) were identified. Nine preoperative variables were included in the primary model; the model's area under the curve (AUC) was 0.66. To account for advanced-stage disease, a subset of patients with disseminated cancer was identified. The final model included 6 preoperative clinical variables: age, chronic hypertension requiring medication, ascites, white blood-cell count, hematocrit, and serum creatinine. The model AUC was 0.73.

Conclusions:

The risk of mortality and morbidity after PCS should be considered in deciding between PCS and NACT. A model was developed, using 6 preoperative clinical variables, allows risk estimation to help gynecologic oncologists quantify surgical mortality and severe morbidity after PCS. (J GYNECOL SURG 34:1)

Introduction

In 2015, it was estimated that there would be 21,290 new cases of ovarian cancer and an estimated 14,180 people would die of this disease. Approximately two-thirds of patients typically present with widely metastatic advanced-stage disease (stages III and IV).¹ Primary cytoreductive surgery (PCS), followed by adjuvant chemotherapy has been the standard management of ovarian, tubal and primary peritoneal cancer. The primary goal of PCS is optimal cytoreduction.

PCS has been associated with significantly higher surgical complication rates and mortality compared to interval cytoreductive surgery (ICS). Data from historical cohorts have shown that major surgical complication rates range from 11% to 22% and mortality from 0.5% to 1.8%.^2–5 Based on 2 randomized trials (Vergote et al.⁶ and Kehoe et al.⁷) comparing PCS to neoadjuvant chemotherapy (NACT), the rate of major surgical complications after PCS was reported to be high at 20% and 24%, respectively, in the 2 trials. In them, the mortality rates after PCS were 2.7% and 5.6%, respectively, compared to 0.6% and 0.5% in the NACT groups.^6,7 Furthermore, PCS has been associated with delayed receipt of chemotherapy, according to a population-based study conducted by Wright et al.⁸ using Surveillance, Epidemiology and End Results (SEER)–Medicare data from 1991 to 2005. The researchers found that 12% of patients failed to receive any postoperative chemotherapy and 28% of patients received chemotherapy more than 6 weeks after PCS.⁸

Given the importance of avoiding serious surgical morbidity and mortality, and the need for timely administration of adjuvant chemotherapy, preoperative models built to aid in surgical decision-making should account for the risk of postoperative surgical complications and mortality, and not merely optimal cytoreduction when deciding between PCS versus NACT. To date, ovarian cancer-prediction models have focused on predicting optimal cytoreduction without accounting for surgical mortality and serious morbidity. Therefore, the current authors sought to develop a model to predict surgical mortality and serious morbidity after PCS (METRICS: Mortality and sErious morbidiTy after ovaRIan Cancer Surgery).

Materials and Methods

Patients were identified who had diagnoses of ovarian, fallopian, and/or primary peritoneal cancer (ICD-9 codes 183.0, 183.2 and 158.0) and who underwent cytoreductive surgery in the National Surgical Quality Improvement Program database (NSQIP) from 2006 to 2012. The NSQIP database is a prospectively collected, nationally validated quality-assessment tool used by the American College of Surgeons. The database is used to collect data on 250 variables, including preoperative demographics, comorbidities, laboratory values, operative details, and 30-day postoperative outcomes and complications. The history and current details of NSQIP, including sampling strategy, data-abstraction procedures, variables collected, outcomes, and structure have been well-described elsewhere.^9,10

Patients were excluded if they had received chemotherapy, or radiotherapy prior to surgery, or were having surgery for recurrent disease. Twenty-three preoperative variables were included in the analysis. Outcome variables collected included 30-day mortality and serious morbidity (wound dehiscence, abscess, sepsis, septic shock, acute renal failure, cerebrovascular accident (CVA), myocardial infarction (MI), cardiac arrest, pulmonary embolism (PE), unplanned intubation, ventilator dependence for more than 48 hours, and coma).

Variables with more than 5% missing data were imputed using multiple imputation via chained equations, using only “complete” variables, or variables with no missing data, to impute the missing data with a total of 10 iterations. The remaining analyses were completed across all 10 imputed datasets. A backward selection procedure was utilized to identify model parameters, with a significance level of >0.20 for removal from the model. t-Test and χ² tests were used to evaluate continuous and categorical variables, respectively. Multivariable logistic regression and tenfold cross validation were performed to validate the model. Model quality was evaluated using c-statistics for discrimination. The area under the curve (AUC) or c-statistic is considered the most relevant measure of model success and refers to the ability of the risk estimate to discriminate cases from noncases. AUC is the probability a randomly selected patient who experienced an event had a higher risk score than a patient who had not experienced the event. The higher the model's AUC is, the higher the probability of predicting the event will be.¹¹ Statistical analysis was performed using both the R version 3.1.1 and Stata version 14.

Results

There were 2262 patients identified who underwent PCS from 2006 to 2012 in the NSQIP database. Of these, 194 (8.6%) experienced the outcomes of interest (30-day mortality and serious morbidity). Among these 194 patients, the mean age was 63.4 (standard deviation [SD]: 11.56) and the mean body mass index was 30 (SD: 8.0). See Table 1. Eighty percent were Caucasians and 9% were African Americans (Table 2).

Table 1.

Clinical Characteristics and Preoperative Laboratory Values of Patients, Continuous Variables

Variables (units)	Mean ± SD without outcome	Mean ± SD with outcome	p-Value
Age (years)	59.86 ± 12.59	63.39 ± 11.56	0.0001
BMI	29.23 ± 7.78	29.96 ± 8.01	0.2101
Sodium (mEq/L)	138.61 ± 2.98	138.28 ± 3.43	0.2153
BUN (mg/dL)	13.56 ± 6.15	14.97 ± 8.53	0.052
Creatinine (mg/dL)	0.79 ± 0.24	0.87 ± 0.43	0.019
Albumin (g/dL)	3.86 ± 0.60	3.65 ± 0.62	<0.001
WBC ( × 10³/mm³)	7.67 ± 2.83	8.46 ± 3.92	0.0063
HCT (%)	37.46 ± 4.90	36.39 ± 4.95	0.0052
Platelets ( × 10³/mm³)	312.46 ± 118.87	329.19 ± 124.00	0.0522

SD, standard deviation; BMI, body mass index; BUN, blood–urea–nitrogen; WBC, white blood-cell count; HCT, hematocrit.

Table 2.

Clinical Characteristics and Preoperative Laboratory Values of Patients, Categorical Variables

Variables	Levels	Without outcome (%)	With outcome (%)	p-Value
Race	White	1602 (77)	155 (80)	0.3517
	African American	141 (7)	17 (9)
	Other	80(4)	11 (6)
	Unknown	245 (12)	11 (6)
Diabetes		208 (10)	29 (15)	0.0242
Smoking		300 (15)	24 (12)	0.4353
Dyspnea		193 (9)	34 (18)	0.001
Ascites		457 (22)	68 (35)	<0.001
CHTN + M		850 (41)	107 (55)	<0.001
Disseminated cancer		452 (22)	56 (29)	0.0168
Steroid use		62 (3)	10 (5)	0.1033
Weight loss		92 (4)	15 (8)	0.0318
Bleeding disorder		53 (3)	12 (6)	0.0018
SIRS		31 (1)	11 (6)	<0.001
ASA class	I & II	1029 (50)	68 (35)	<0.001
	III	974 (47)	116 (60)
	IV & V	64 (3)	10 (5)
Functional status	Independent	2039 (99)	180 (93)	<0.001
	Dependent	27 (1)	14 (7)

CHTN+M, chronic hypertension requiring medication; SIRS, systemic inflammatory response; ASA, American Society of Anesthesiologists.

Outcomes included 168 patients (7.43%) with serious morbidity and 26 patients (1.15%) who died (Table 3). Infection complications (abscess, wound disruption, and sepsis) were the most-common serious morbidity (63%), followed by PE (15.5%), unplanned intubation (6.5%), ventilator dependence for more than 48 hours (6.5%), CVA (4.7%), MI (2.97%), and cardiac arrest (1.78%). Twenty-three preoperative variables were included in the analysis (Tables 1 and 2). Of these 23 variables, a total of 9 variables were selected for inclusion in the predication model, based on multiple logistic regression and tenfold cross validation. These variables included (age, functional status, chronic hypertension requiring medication, ascites, systemic inflammatory response (SIRS), history of bleeding disorder, white blood-cell count (WBC), serum albumin, and creatinine). The model AUC was 0.66. This model included patients with early and advanced-stage disease.

Table 3.

Mortality and Serious Morbidity Outcomes (8.6%: 194/2262)

Outcome (n = 194)	Count	Percentage (N/2262)
Mortality	26	1.15
Serious morbidity	168	7.43
Abscess	50	2.2
Wound disruption	20	0.89
Sepsis or septic shock	36	1.6
CVA	8	0.35
Coma	0	0
MI	5	0.22
Cardiac arrest	3	0.13
PE	26	1.15
Unplanned intubation	11	0.5
Ventilator dependence	8	0.36
ARF	1	0.04

CVA, cerebrovascular accident; MI, myocardial infarction; PE, pulmonary embolism; ARF, acute renal failure.

Given that surgical complications and mortality were more likely to occur in patients with advanced-stage or widely metastatic disease, a subset of patients who were coded by NSQIP as having “disseminated cancer” were identified. In NSQIP, disseminated cancer was defined as: “(1) Has spread to one site or more sites in addition to the primary site AND (2) In whom the presence of multiple metastases indicates the cancer is widespread, fulminant, or near terminal.” This variable was used as a surrogate marker for having advanced-stage ovarian cancer. Although the definition of the disseminated cancer variable did not specify whether it was a preoperative or an intraoperative observation, the current authors believe that advanced-stage ovarian cancer can be predicted by using preoperative imaging with good certainty. The current authors also believe that this group represented patients in whom disease distribution required complex cytoreductive surgery.

There were 508 patients identified who had disseminated cancer. Of these, 47 patients (9.25%) suffered serious morbidity and 9 patients (1.17%) died within 30 days after PCS, for a total of 56 patients who had severe outcomes (Table 4). Using multiple logistic regression and tenfold crossvalidation, 6 preoperative clinical variables were identified for the final model. The model included: age (odds ratio [OR] = 1.03), chronic hypertension requiring medication (OR = 1.82), ascites (OR = 2.14), WBC (OR = 3.34), hematocrit (OR = 0.11), and serum creatinine (OR = 1.91). See Table 5. The model AUC was 0.73. The final model (METRICS) provided individualized model-based risk estimates of mortality and serious morbidity after PCS calculation. The risks from the model can be estimated by the equation in Box 1.

Table 4.

Mortality and Serious Morbidity Outcomes in Patients with Disseminated Cancer (Advanced Stage Disease; 9.25%: 56/508)

Outcome (n = 56)	Count	Percentage (n/508)
Mortality	9	1.17
Serious morbidity	47	9.25
Abscess	20	3.39
Wound disruption	1	0.19
Sepsis or septic shock	12	2.36
CVA	2	0.39
Coma	0	0
MI	1	0.19
Cardiac arrest	0	0
PE	6	1.18
Unplanned intubation	1	0.19
Ventilator dependence	4	0.78
ARF	0	0

CVA, cerebrovascular accident; MI, myocardial infarction; PE, pulmonary embolism; ARF, acute renal failure.

Table 5.

Predictors of Mortality and Serious Morbidity After PCS in Patients With Disseminated Cancer (Advanced Stage Disease), Multivariable Logistic Regression

Variables (units)	OR	95% CI	p-Value
Age (year)	1.03	(1.00–1.06)	0.041
Ascites	2.14	(1.17–3.91)	0.014
CHTN + M	1.82	(0.95–3.48)	0.071
WBC ( × 10³/mm³)	3.34	(1.54, 7.14)	0.002
HCT (%)	0.11	(0.01–0.95)	0.044
Creatinine (mg/dL)	1.91	(0.73–4.98)	0.184

OR, odds ratio; CI, confidence interval; CHTN+M, chronic hypertension requiring medication; WBC, white blood-cell count; HCT, hematocrit.

Discussion

The estimation of surgical risk is essential for decision-making by both gynecologic oncologists and patients. Clinicians often base their risk estimation on clinical judgment, surgical literature, and personal experience. This method has variable accuracy and is subject to bias.^12–15 More accurate, individual patient data-driven estimates of surgical risk of PCS could inform patients and gynecologic oncologists better and will improve outcomes.

METRICS predicts 30-day mortality and serious morbidity after PCS using preoperative patient-specific data. The proposed predictive model for morbidity and mortality was statistically validated and demonstrated good accuracy and discrimination. As such, the model could be a useful tool for patient counseling and informed decision-making during preoperative evaluation of patients with ovarian cancer. Furthermore, this model was developed using NSQIP, a high-quality, prospectively collected surgical database. By developing a predictive model for morbidity and mortality, we can determine optimal treatment pathways better for patients, minimize severe surgical complications, and, ultimately, improve the quality of care for patients who have ovarian cancer.^16,17

Barber et al. recently published a risk assessment for only elderly patients with ovarian cancer (more than 65-years-old) using the NSQIP database.¹⁸ The METRICS model will calculate risk of mortality and serious morbidity for all patients who have ovarian cancer regardless of their ages. The final model is more specific to patients with advanced-stage disease, compared to the Barber et al. model,¹⁸ because the current authors used disseminated cancer as a surrogate for advanced-stage disease. This was demonstrated in the better AUC, using a simpler model that included only 6 variables (compared to 8 categorical variables in Barber et al.¹⁸). Finally, contrary to the abovementioned model, the METRICS model is internally crossvalidated. Patankar et al. published risk stratification for patients undergoing surgery for ovarian cancer.¹⁹ The researchers used intraoperative variables in addition to baseline factors, precluding this risk-stratification's use as a preoperative decision tool.

Stashwick et al. published a model predicting “successful surgery” (defined as optimal residual disease and no major perioperative complication) in patients who underwent PCS. These researchers identified low albumin and splenic disease as the only 2 parameters associated with a higher risk of major perioperative complications.²⁰ This model was based on 106 patients from a single institution, and there were 24 major complications. The small number of patients and major complications limit this model's generalizability. Another model was published by Gerestein et al.³ of a prediction normogram for 30-day morbidity based on a cohort of 293 patients. The researchers' final model included intraoperative variables, such as operative time and extent of surgery, in addition to baseline factors.³ For this reason, this model would also not be applicable as a preoperative decision tool. In contrast, the METRICS model is based solely on preoperative variables and includes the largest number of patients, compared to any other published models.

Overall, the METRICS model performed well with an AUC of 0.73. While an AUC of 0.99 is theoretically achievable for an outcome with narrowly distributed risk factors, for a composite outcome with a large variance, the maximum possible value for the c-statistic is in the range of 0.62–0.78.¹¹ As such, the METRICS model with an AUC of 0.73 is a very good discrimination tool. The current authors recognize that between the 2 reported models (all ovarian cancer patients who underwent cytoreductive surgery versus the subset with disseminated disease); there was an absolute difference in c-statistic of 0.08. While this difference between the 2 models might not seem great, this represents an improvement of at least 10% and is statistically and potentially clinically significant when working with an AUC of 0.6–0.8.

To illustrate the application of the current author's model, 2 patient examples are provided below, using METRICS as a risk-predictor model:

(1) A 60-year-old patient with no ascites, no chronic hypertension, and the following laboratory values—hematocrit (HCT), 38%; WBC, 7000; and creatinine, 0.7 mg/dL—would have a 30-day mortality and serious morbidity of 4% following PCS.

(2) A 60-year-old patient with ascites, chronic hypertension (requiring medications), and the following laboratory values—HCT, 38%; WBC, 7000 and creatinine, 1.0 mg/dL—would have a risk of 17% of serious surgical morbidity or mortality.

While the model provides a risk estimate, each surgeon would, of course, have to determine his or her threshold of what would be an acceptable cutoff point (% risk) to proceed with PCS versus NACT. An electronic version of the equation is available for use (Supplementary Data; Supplementary Data are available online at www.liebertpub.com/gyn).

The strengths of the current study are many. First, the METRICS model was developed using a large national patient sample. Second, the data represent current practice. Third, the model includes only preoperative variables and baseline information. Fourth, the model was internally validated. Fifth, it provides individualized risk estimation.

Box 1.

Final METRICS Model Calculator Equation

\documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\usepackage{upgreek}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} $$\ Prob ( outcome ) = \frac { 1 } { { 1 + \exp ( - 0.901 - 0.03*A - 0.76*B - 0.60*C - 1.20* ln ( D ) - 0.65* ln ( E ) + 2.19* ln ( F ) ) } } $$ \end{document}

A: Age (year) (continuous)

B: Ascites (categorical)

C: Chronic Hypertension requiring medication (categorical)

D: White blood cells count (x10³/mm³) (continuous)

E: Serum creatinine (mg/dL) (continuous)

F: Hematocrit (%) (continuous)

METRICS: Mortality sErious morbidiTy ovaRIan Cancer Surgery

The present study also had several limitations. One of the limitations was the retrospective design of the study. However, NSQIP database is collected prospectively and the current authors are interested in predictive power rather than in causal effects. Another potential bias is the selection bias that could arise from NSQIP naturally gathering data from member institutions and hospitals that are motivated to improve the quality of surgical care. However, the database is still diverse with respect to size, region, and academic versus community hospital status, and, therefore, supports the generalizability of the current study's findings. Other limitations include the absence of staging, grade, or histology information in the NSQIP database. However, it was possible to use disseminated disease as a surrogate for patients with advanced-stage ovarian cancer and the current authors know that the majority of epithelial ovarian cancers are of higher grade and have aggressive histology that would not be likely to change the risk of surgical complications. When the subset analysis of patients with disseminated cancer was performed, there was better discrimination using the METRICS model.

Conclusions

For the future, the current authors aim to validate the model externally using multi-institutional–level data. A future model that combines predicting optimal cytoreduction and the risk of mortality and serious morbidity after PCS using a large data source would be the model of choice to aid in decision-making between PCS and NACT. The cost of serious morbidity after PCS is another area that needs further research and evaluation given that the cost of healthcare is increasing, and we must consider means to improve surgical outcomes of patients undergoing ovarian cancer.

Footnotes

Author Disclosure Statement

All of the authors report no conflicts of interests.

References

Cancer Statistics. Cancer Stat Facts: Ovarian Cancer. Online document at: http://seer.cancer.gov/statfacts/html/ovary.html Accessed March 2, 2017.

Eisenkop

, Friedman

, Wang

. Complete cytoreductive surgery is feasible and maximizes survival in patients with advanced epithelial ovarian cancer: A prospective study. Gynecol Oncol, 1998; 69:103.

Gerestein

, Damhuis

, Burger

, Kooi

. Postoperative mortality after primary cytoreductive surgery for advanced stage epithelial ovarian cancer: A systematic review. Gynecol Oncol, 2009; 114:523.

Chi

, Zivanovic

, Levinson

, et al. The incidence of major complications after the performance of extensive upper abdominal surgical procedures during primary cytoreduction of advanced ovarian, tubal, and peritoneal carcinomas. Gynecol Oncol, 2010; 119:38.

Rafii

, Stoeckle

, Jean-Laurent

, et al. Multi-center evaluation of post-operative morbidity and mortality after optimal cytoreductive surgery for advanced ovarian cancer. PLoS One, 2012; 7:e39415.

Vergote

, Tropé

, Amant

, et al.; European Organization for Research and Treatment of Cancer—Gynaecological Cancer Group; NCIC Clinical Trials Group. Neoadjuvant chemotherapy or primary surgery in stage IIIC or IV ovarian cancer. N Engl J Med, 2010; 363:943.

Kehoe

, Hook

, Nankivell

, et al. Primary chemotherapy versus primary surgery for newly diagnosed advanced ovarian cancer (CHORUS): An open-label, randomised, controlled, non-inferiority trial. Lancet, 2015; 386:249.

Wright

, Herzog

, Neugut

, Burke

, Lu

, Lewin

, Hershman

. Effect of radical cytoreductive surgery on omission and delay of chemotherapy for advanced-stage ovarian cancer. Obstet Gynecol, 2012; 120:871.

Cohen

, Bilimoria

, Ko

, Hall

. Development of an American College of Surgeons National Surgery Quality Improvement Program: Morbidity and mortality risk calculator for colorectal surgery. J Am Coll Surg, 2009; 208:1009.

10.

Bilimoria

, Liu

, Paruch

, Zhou

, Kmiecik

, Ko

, Cohen

. Development and evaluation of the universal ACS NSQIP surgical risk calculator: A decision aid and informed consent tool for patients and surgeons. J Am Coll Surg, 2013; 217:833.

11.

Cook

NR.

Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation, 2007; 115:928.

12.

Ferguson

, Stromberg

, Celauro

. Estimating lung resection risk: A pilot study of trainee and practicing surgeons. Ann Thorac Surg, 2010; 89:1037;discussion:1042.

13.

Karliczek

, Harlaar

, Zeebregts

, Wiggers

, Baas

, van Dam

. Surgeons lack predictive accuracy for anastomotic leakage in gastrointestinal surgery. Int J Colorectal Dis, 2009; 24:569.

14.

Markus

, Martell

, Leister

, Horstmann

, Brinker

, Becker

. Predicting postoperative morbidity by clinical assessment. Br J Surg, 2005; 92:101.

15.

Volk

, Roney

, Merion

. Systematic bias in surgeons' predictions of the donor-specific risk of liver transplant graft failure. Liver Transpl, 2013; 19:987.

16.

Fisher

, McClellan

, Safran

. Building the path to accountable care. N Engl J Med, 2011; 365:2445.

17.

Bekelis

, Desai

, Bakhoum

, Missios

. A predictive model of complications after spine surgery: The National Surgical Quality Improvement Program (NSQIP) 2005–2010. Spine J, 2014:14:1247.

18.

Barber

, Rutstein

, Miller

, Gehrig

. A preoperative personalized risk assessment calculator for elderly ovarian cancer patients undergoing primary cytoreductive surgery. Gynecol Oncol, 2015; 139:401;corrigendum: Gynecol Oncol 2016:142:380.

19.

Patankar

, Burke

, Hou

, et al. Risk stratification and outcomes of women undergoing surgery for ovarian cancer. Gynecol Oncol, 2015; 138:62.

20.

Stashwick

, Post

, Arruda

, Spillman

, Behbakht

, Davidson

, Kelly

. Surgical risk score predicts suboptimal debulking or a major perioperative complication in patients with advanced epithelial ovarian, fallopian tube, or primary peritoneal cancer. Int J Gynecol Cancer, 2011; 21:1422.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.01 MB