A Multimodality Triage Algorithm to Improve Cytoreductive Outcomes in Patients Undergoing Primary Debulking Surgery for Advanced Ovarian Cancer: A Memorial Sloan Kettering Cancer Center Team Ovary Initiative

Abstract

Objective

To describe outcomes using a multimodal algorithm to triage patients with advanced epithelial ovarian cancer (EOC) to primary debulking surgery (PDS) versus neoadjuvant chemotherapy (NACT).

Methods

All patients with EOC treated at our institution from 04/2015–08/2018 were identified. We included patients without contraindication to PDS who underwent prospective calculation of a Resectability (R)-score. A low risk score for suboptimal cytoreduction was defined as ≤6, and a high risk score ≥7. Patients were triaged to laparotomy/PDS, laparoscopic evaluation of resectability (LSC), or NACT depending on R-score.

Results

Among 299 participants, 226 (76%) had a low risk score and 73 (24%) a high risk score. For patients with a low risk score, management included laparotomy/PDS, 181 (80%); LSC, 43 (19%) (with subsequent triage: PDS, 31; NACT, 12); and NACT, 2 (1%). For patients with a high risk score, management included laparotomy/PDS, 9 (12%); LSC, 51 (70%) (with subsequent triage: PDS, 28; NACT, 23); and NACT, 13 (18%). Overall, 83% underwent PDS, with a 75% CGR rate and 94% optimal cytoreduction rate. Use of the algorithm resulted in a 31% LSC rate and a 6% rate of suboptimal PDS.

Conclusions

The multimodal algorithm led to excellent surgical results; 94% of patients achieved an optimal resection, with a very low rate of suboptimal cytoreduction.

Introduction

The mainstay of treatment for advanced epithelial ovarian cancer (EOC) is a combination of debulking surgery and platinum/taxane-based chemotherapy.¹ There is robust evidence supporting the association between minimal volume of residual disease after surgical debulking and improved patient outcomes. The goal of surgery has evolved from achieving an optimal resection (≤1 cm residual disease) to complete gross resection (CGR) of disease.² Randomized trials evaluating neoadjuvant chemotherapy (NACT) versus primary debulking surgery (PDS) have reported CGR rates at PDS of 12–45.5% and optimal debulking rates at PDS of 37–92.8%.^3–9 At high-volume centers, CGR rates for advanced-stage EOC range between 50% and 70%.¹⁰ Despite advances in surgical care, women undergoing debulking surgery remain at risk for suboptimal resection (residual disease >1 cm), with its associated morbidity without proportionate survival benefit. Preoperative assessment of tumor burden with clinical and radiologic data has been the goal of several predictive models^11–13 to minimize the risk of a suboptimal debulking while maximizing tumor resection.

We previously reported on two preoperative PDS predictive models that use both clinical and radiologic factors (Supplemental Table 1). Resectability Score 1 (RS1) model by Suidan et al. in 2014 focused on the risk of suboptimal debulking. The model used three clinical and six radiologic variables to predict the risk of residual disease >1 cm.¹² In the RS1 model, a predictive score ≥7 was associated with a suboptimal debulking rate of greater than 50%. Resectability Score 2 (RS2) model by Suidan et al. in 2017 focused on CGR and was created through a post-hoc analysis of the RS1 cohort using three clinical and nine radiologic variables to predict the likelihood of any gross residual disease.¹³ In the RS2 model, a predictive score ≥6 was indicative of an 87% chance of any gross residual disease at resection. These models demonstrate that increasing preoperative resectability scores are associated with incremental increases in the likelihood of being left with residual disease at the time of resection. In 2019, Kumar and colleagues sought to externally validate both of these models with their institutional ovarian cancer database and found the RS2 model to be independently validated.¹⁴

An alternative approach to radiologic systems for the triage of patients to PDS or NACT are laparoscopic-based scoring systems. These models are designed to predict the likelihood of achieving an optimal resection^11,15–17 and have been associated with a decrease in the number of futile laparotomies in patients undergoing PDS.¹⁷ Fleming et al. described their experience with a quality improvement project implementing a laparoscopic assessment tool published by Fagotti et al.^11,15,16 to triage select women with advanced EOC to PDS or NACT, with the goal of achieving a CGR at the time of PDS or interval debulking surgery (IDS). With the application of diagnostic laparoscopy to potential debulking candidates, the group reported a CGR rate of 88% at the time of PDS, which was used for 23% of newly diagnosed ovarian cancer patients.¹⁸

In 2014, we created an institutional, multidisciplinary section based within the Gynecology Service at Memorial Sloan Kettering Cancer Center (MSK) dedicated to improving the care of women with EOC; this section was named “Team Ovary.” Within the multidisciplinary framework, and building upon previously reported diagnostic and treatment strategies, we developed and implemented a management algorithm incorporating both radiologic evaluation and selective laparoscopy for patients with newly diagnosed advanced EOC. The objective of the present study is to report the surgical outcomes associated with the implementation of this multimodality algorithm for the management of patients with advanced EOC at our institution. In keeping with our Team Ovary mission for continual improvements in the management of EOC,¹⁹ and following Kumar et al.’s independent validation of our RS2 model, we also performed a post-hoc analysis of this same cohort using the RS2 model.

Methods

Patients

All patients presenting with presumed advanced epithelial ovarian, fallopian tube, or primary peritoneal cancer (all referred to as EOC for this study) evaluated at our institution between April 1, 2015 and August 31, 2018 and considered candidates for PDS were identified. Eligible patients underwent preoperative ovarian cancer radiologic and clinical assessment (ORA) as part of the multimodal triage algorithm. Patients older than 18 years of age were included based on preoperative suspicion for advanced disease and prospective use of the algorithm; final surgical stage was noted and collected but not used as an exclusion criterion. In our determination of PDS vs NACT, since its publication in 2007,²⁰ we have used the criteria described by Aletti et al. to determine patients who may be candidates for PDS. Those who we triage to NACT have all three of the following criteria: 1) age >75 years old, 2) serum albumin <3.5 g/dL, and 3) American Society of Anesthesiologist (ASA) classification score ≥3 or extensive disease. Additionally, patients who have medical comorbidities precluding them from surgery, including venous thromboembolism, and those who have apparent unresectable stage IV disease are not considered candidates for PDS and are excluded from algorithm evaluation. Patients with stage IVA disease, small subcapsular liver lesions, and supradiaphragmatic lymphadenopathy are considered resectable. Unresectable stage IVB disease was defined as follows: multifocal parenchymal liver metastases, pulmonary metastases including diffuse pleural involvement, brain metastases, and bulky thoracic adenopathy (not including cardiophrenic lymphadenopathy). Additional exclusion criteria included the following: younger than 18 years of age at the time of surgery, transfer of care after undergoing initial treatment elsewhere, and non-ovarian histology on final pathology.

Imaging Assessment

All patients underwent pretreatment contrast-enhanced computed tomography (CT) of the abdomen and pelvis; a CT of the chest was recommended as standard management in the alogithm. In addition to a standard-of-care CT imaging report, the scans were prospectively scored by one of five fellowship-trained radiologists with dedicated expertise in ovarian cancer imaging. The ORA included the assessment of the following 11 pre-defined anatomic sites (see Supplemental Table 2 for the specific criteria, including lesion size)^12,13: lesser sac, splenic hilum (splenic vessel entry) or splenic ligaments (gastrocolic, splenocolic, and splenorenal), root of the superior mesenteric artery (SMA), small bowel mesentery, small bowel mesenteric infiltration (defined as small bowel tethering and/or angulation without measurable lesions), retroperitoneal lymph nodes above the renal hilum, supradiaphragmatic lymph nodes, ascites (moderate to large volume), gastrohepatic ligament or porta hepatis lesions, gallbladder fossa or left inter-segmental fissure, and stage IV disease (aside from supradiaphragmatic lymph nodes). Quantitative measurements were performed for all visualized lesions. At each location, the presence of tumor or lymphadenopathy were assessed using a 5-point scale: 1=definitely absent; 2=probably absent; 3=indeterminate; 4=probably present; and 5=definitely present. Only lesions categorized as 4 or 5 were included in the ORA. In addition to CT imaging data, data for three clinical criteria (age, CA-125 levels, and ASA classification score as determined before surgery) were collected prospectively and used to compute a resectability score (RS).

Multimodality Algorithm Implementation

Combining the ORA and clinical variables, RS1 was calculated and used by the primary surgeon for clinical decision making. Patients with an RS1 score of 0–6 (deemed “low risk” for suboptimal resection) were recommended to undergo PDS via laparotomy, with an option for diagnostic laparoscopy at the surgeon’s discretion (Figure 1). Patients with an RS1 score of 7 or higher (deemed “high risk” for suboptimal cytoreduction) were recommended to undergo a laparoscopic evaluation for resectability and further triage to PDS or NACT. For patients deemed to have unresectable disease at the time of diagnostic laparoscopy, a second surgeon intraoperative consult to corroborate the clinical plan was recommended.

Figure 1. Preoperative Algorithm.

(EOC, epithelial ovarian cancer; CT, computed tomography; PDS, primary debulking surgery; Dx LSC, diagnostic laparoscopy; NACT, neoadjuvant chemotherapy)

Analysis of Outcomes

Clinical data, including demographics, intra- and perioperative findings, and surgical outcomes, were collected. Residual disease status was categorized as follows: CGR, optimal cytoreduction (CGR and residual disease 0.1–1.0 cm), and suboptimal cytoreduction (residual >1 cm). Clinical outcomes (i.e., residual disease status, rate of diagnostic laparoscopy, rate of suboptimal debulking) and the use of the triage algorithm were reported using descriptive statistics. Suboptimal debulking (futile laparotomy) was further divided into two categories: suboptimal debulking with maximal effort or ”open and close” with no procedures performed.

Post-Hoc Analysis of the RS2 Model

A post-hoc exploratory analysis with the goal to evaluate RS2 in this same cohort was then performed. Using the prospective ORA and clinical data available for all included patients, RS2 was calculated. Clinical outcomes according to RS2 are described.

Results

Algorithm Adherence and Outcomes

During the study period, 406 patients with suspected advanced EOC underwent contrast-enhanced CT including ORA; of these, 299 patients (74%) met all eligibility criteria and were included in this study. The reasons for exclusion were as follows: non-ovarian histology, 43; transfer of care, 28 (NACT started at outside facility prior to consultation at our institution or patient presented for second opinion but received care elsewhere following consultation); unresectable stage IVB disease, 23 (thoracic disease, 18; extra-abdominal adenopathy, 5); medical contraindication to surgery, 12 (medical comorbidities, 8; venous thromboembolus, 4); and patient preference, 1. Demographic data and surgical outcomes are described in Table 1.

Table 1:

Patient Characteristics

Characteristics	N=299
Median Age, years (range)	63 (31–89)
Median BMI, kg/m2 (range)	25.2 (16.4–52)
FIGO stage	n (%)
Stage I	2 (0.7)
Stage II	7 (2)
Stage IIIA/B	14 (5)
Stage IIIC	183 (61)
Stage IV	93 (31)
Histology	n (%)
High-grade serous carcinoma	252 (84)
Low-grade serous carcinoma	8 (3)
Clear cell	9 (3)
Endometroid	8 (3)
Mucinous	20 (7)
Other	2 (0.7)
Median Preoperative Albumin, g/dL (range)	3.9 (2.3–4.9)
Median Preoperative CA-125, U/mL (range)	417 (0–29,680)

BMI, body mass index; FIGO, International Federation of Gynecology and Obstetrics

The median age of our cohort was 63 years (range, 31–89 years). The median body mass index (BMI) was 25.2 kg/m² (range, 16.4–52 kg/m²), and the median preoperative albumin level was 3.9 g/dL (range, 2.3–4.9 g/dL). Ninety-two percent of patients had stage IIIC/IV disease, and 84% had disease of high-grade serous histology (Table 1). Two hundred forty-nine patients (83%) underwent PDS either directly with laparotomy (n=190) or after laparoscopic evaluation of resectability (n=59). In total, 17% of patients (n=50) received NACT—15 received chemotherapy without surgical evaluation and 35 were triaged to NACT based upon laparoscopic findings. Of the 249 patients who underwent PDS, 75.5% (n=188) achieved a CGR, 94% (n= 234) achieved an optimal resection, and 6% (n=15) ended with a suboptimal resection. In patients who underwent PDS, other than hysterectomy, bilateral salpingooophorectomy and omentectomy, the most common procedures performed were diaphagm peritonectomy/stripping (67%) and colon resection (61%) (Table 2).

Table 2:

Summary of Operative Outcomes

Residual Disease following PDS	N=249 (%)
CGR	188 (75.5)
Optimal cytoreduction, ≤1.0 cm	234 (94)
Suboptimal cytoreduction, >1 cm	15 (6)
Procedures performed	N=249 (%)
Diaphragmatic peritonectomy/resection	167 (67)
Large bowel resection	151 (61)
Appendectomy	95 (38)
Splenectomy	74 (30)
Mediastinal lymph node dissection	67 (27)
Porta hepatis lymph node dissection	60 (24)
Partial hepatic resection	47 (19)
Small bowel resection	45 (18)
Cholecystectomy	33 (13)
Partial bladder cystectomy	12 (5)
Partial gastrectomy	9 (4)
Distal pancreatectomy	8 (3)

PDS, primary debulking surgery; CGR, complete gross resection

Based on the prospectively calculated RS1, 226 patients (76%) were designated as low risk for suboptimal cytoreduction and 73 (24%) high risk (Figure 2). In the low-risk group, 181 (80%) underwent PDS via laparotomy, 43 (19%) underwent laparoscopy, and 2 (1%) received NACT. For low-risk patients who underwent laparoscopic evaluation, 72% (n=31/43) went on to undergo laparotomy and PDS and 28% (n=12) were deemed unresectable and triaged to NACT (Table 3). Of the 212 low-risk patients undergoing PDS, cytoreductive outcomes were as follows: CGR, 168 (79%); optimal cytoreduction, 199 (94%); and suboptimal debulking, 13 (6%). Of the 73 patients in the high-risk group, 9 (12%) underwent laparotomy and PDS, 51 (70%) underwent diagnostic laparoscopy (of whom 28 [55%] proceeded to undergo PDS and 23 [45%] went on to receive NACT; Table 3), and 13 (18%) received NACT without any surgical evaluation. Of the 37 high-risk patients who underwent PDS, cytoreductive outcomes were as follows: CGR, 20 (54%); optimal cytoreduction, 35 (95%); and suboptimal debulking, 2 (5%) (Figure 2). Surgical outcomes by RS1 predictive score are shown in Table 3.

Figure 2. Algorithm Adherence and Outcomes Based Upon RS1 Model.

(CT, computed tomography; PDS, primary debulking surgery; LSC, diagnostic laparoscopy; NACT, neoadjuvant chemotherapy; CGR, complete gross resection; O, optimal resection; SO, suboptimal resection)

Table 3:

Outcomes of Primary Debulking Surgery based upon the RS1 Model

Total Predictive Score	Total Patients n (%)	Complete Resection n (%)	Optimal Resection n (%)	Suboptimal Resection n (%)
0	30 (12)	27 (90)	29 (97)	1 (3)
1–2	77 (31)	66 (86)	74 (96)	3 (4)
3–4	61 (24)	43 (70)	56 (92)	5 (8)
5–6	44 (18)	32 (73)	40 (91)	4 (9)
7–8	24 (10)	13 (54)	22 (92)	2 (8)
≥9	13 (5)	7 (54)	13 (100)	-

RS1 model: Resection Score Model 1 (Suidan et al. 2014)

____ denotes the threshold for low and high risk for suboptimal resection: 0–6, low risk and ≥7, high risk

Ninety-four (31%) of 299 patients in the entire cohort underwent diagnostic laparoscopy, of whom 59 (63%) were triaged to PDS. CGR was achieved in 61% (n=36/59) of these patients, and 5 (8%) had a suboptimal resection. Details of all patients who underwent laparoscopic assessment are reported in Supplemental Table 3. Eighteen patients had residual disease of 0.1–1 cm, and of these, 15 had ≤.5 cm residual tumors (13 patients had diffuse carcinomatosis including involvement of the small and large serosa/mesentery and 2 patients had pleural disease identified at the time of diaphragm resection).

In the entire cohort (n=299), 15 patients (5%) underwent a laparotomy with suboptimal debulking. Thirteen (87%) underwent laparotomy with evaluation of the extent of disease and then immediate closure due to intraoperative determination of unresectability. Two patients (13%) underwent maximal cytoreductive effort but were still left with >1 cm residual disease. Reasons for suboptimal debulking were as follows: total colonic involvement (n=8, 53%), extensive left upper quadrant involvement (n=4, 27%), extensive small bowel involvement (n=2, 13%), and inferior vena cava encasement (n=1, 7%).

The multimodality algorithm was followed in 92% of cases (n=275/299). Twenty-four patients received care outside the recommendations of the algorithm. Two patients in the low-risk group went on to receive NACT directly, 9 patients in the high-risk group underwent PDS without diagnostic laparoscopy, and 13 patients in the high-risk group received NACT without surgical evaluation. Evaluation of these individual cases showed that patient and provider preferences regarding the specifics of the case guided the treatment decision.

Post-Hoc Analysis of the RS2 Model

A descriptive analysis of the RS2 model was performed, and results are summarized in Table 4. Overall, the RS2 model performed well in that the higher the RS2, the less likely a CGR was achieved. CGR rates by RS2 were as follows: RS2 score 0–2, 92%; RS2 score 3–5, 78%; RS2 score 6–8, 60%; and RS2 score ≥9, 59%.

Table 4:

Outcomes of Primary Debulking Surgery based upon the RS2 Model^*

Total Predictive Score	Total Patients n (%)	Complete Gross Resection n (%)	Gross Residual Disease n (%)
0–2	76 (31)	70 (92)	6 (8)
3–5	82 (33)	64 (78)	18 (22)
6–8	45 (18)	27 (60)	18 (40)
≥9	46 (18)	27 (59)	19 (41)

RS2 model: Resection Score Model 2 (Suidan et al. 2017)

^*The higher the RS2 predictive score, the higher the likelihood of not achieving a complete gross resection

Discussion

In this study, we report on the implementation and cytoreductive outcomes of a multimodal preoperative algorithm to triage patients with advanced EOC to PDS versus NACT. Maximizing the availability of PDS to the most patients and optimizing resource allocation, while minimizing risk and unnecessary/futile intervention, should be the goals of upfront surgical treatment of advanced EOC. The implementation of this multimodality algorithm led to very high rates of complete (76%) and optimal (94%) resection, and modest rates of diagnostic laparoscopy (31%) and suboptimal debulking (6%).

Laparoscopic-based scoring systems have demonstrated very high specificities and positive predictive values (PPVs) for predicting suboptimal resection. The laparoscopy-based scoring system developed and validated by Fagotti and colleagues for predicting a suboptimal resection at the time of debulking surgery showed a predictive index score of 8 or higher has a sensitivity of 30%, specificity of 100%, PPV of 100%, and a negative predictive value (NPV) of 70% for predicting a suboptimal cytoreductive surgery.^11,15,16 A multicenter, randomized controlled trial used diagnostic laparoscopy as the tool to guide choice of primary treatment, showing that laparoscopy reduced the rate of futile laparotomy (10% vs 39%, p<0.001). Of note, the CGR rate achieved in the study was 57% in the PDS group.^{17, 20} Finally, Fleming et al. recently published their single-institution experience with a prospective quality review process for the primary treatment of EOC and broad application of laparoscopy, and reported a CGR rate of 88% in patients who underwent PDS.¹⁸

When evaluating these studies, it is important to note the total population of EOC patients (i.e., the percentage of patients evaluated). In the Fleming et al. paper, 35% of patients who presented to their institution with EOC were candidates for laparoscopic evaluation, resulting in 23% of patients with EOC proceeding to undergo PDS. Most patients were triaged to NACT based on other factors.¹⁸ There are varying definitions of what is considered stage IV unresectable disease. Within our cohort, we identified 406 patients as potential surgical candidates, and after implementation of the multimodality triage algorithm, 61% (249/406) went on to undergo PDS, the majority of whom underwent optimal and complete resections of disease. Having predefined narrow exclusion criteria for patient status and disease location for PDS allows for the maximization of patients undergoing upfront surgery. The ideal rate of futile laparotomy should not be 0%, as this would indicate only patients who have certainty of resectable disease at preoperative assessment undergo PDS. A small margin of error should be allowed.

The aforementioned studies require universal application of diagnostic laparoscopy, and this raises concerns. In addition to increased operative time, resources, and cost, laparoscopy in this context is limited with regard to evaluation of disease burden in the upper abdomen, specifically the lesser sac and porta hepatis.²⁵ In our cohort, 5 patients had a suboptimal debulking after diagnostic laparoscopy.

Three patients had extensive disease in the left upper quadrant and 2 had extensive bowel disease. Laparoscopic assessment has its own limitations and can fail in the prediction of resectability. Moreover, in addition to underestimating disease in difficult-to-visualize areas, laparoscopy can also overestimate disease burden due to its magnification, lack of adequate tactile feedback, and inability to accurately predict the depth and extent of bowel involvement.

As surgical and therapeutic advances become standard practice, predictive systems for residual disease should be continuously updated. The Fagotti scoring system evaluates and scores seven anatomic locations to determine if the disease is amenable to resection.¹⁵ Considering the surgical advances made in the management of EOC over the last 2 decades, including resection of disease in the upper abdomen,^21–23 some have proposed to increase the laparoscopy score threshold to 10.¹⁸ This is supported by data published by Petrillo et al. who sought to update this scoring system in the setting of upper abdominal surgery, and they found a score of 10 had a PPV of 100% in predicting residual disease. Just as gynecologic oncology surgeons are continually improving intraoperative techniques, predictive models should be modernized with the goal of accurately predicting CGR while simultaneously maximizing (as opposed to depriving) the number of patients who undergo and derive survival benefits from PDS.

Resectability algorithms and scores need to be improved and validated. Following the implementation of the RS1 model in our prospective algorithm, the RS2 model for predicting CGR was independently validated.¹⁴ Our post-hoc analysis of the RS2 model shows promise in that it may be a better predictor of cytoreductive outcome, especially with regard to CGR rates, than the RS1 model. We plan to implement and prospectively analyze the cytoreductive outcomes of the RS2 model within our department. Additionally, innovations in imaging modalitities for ovarian cancer, such as diffusion-weighted magnetic resonance imaging (MRI), are being investigated to incorporate into future multimodal algorithms.

The multimodality algorithm was implemented as a guideline for our department and was adopted service wide. There were specific instances where derivations were made in the algorithm at the discretion of the surgeon. For example, in the high-risk group, 9 patients went straight to PDS and 13 patients went on to NACT. This highlights the role of surgeon expertise and judgment in the management of these patients—a factor very difficult to quantify. During the time period of this study, multidisciplinary collaboration also led to real-time improvements in systems-based practice. For example, after an interdisciplinary discussion of instances of suboptimal debulking due to total colon or stomach involvement, the Radiology team began to comment on the presence or absence of these findings on CT. This ongoing improvement subsequently leads to improvements in patient management, and potential derivations and improvements in the algorithm.

This study was a single-arm quality improvement initiative. The strengths of the study were the large cohort size at a high-volume ovarian cancer center that prioritizes PDS, CGR, and a multispecialty surgical approach to cytoreduction, as needed. We prospectively applied our multimodality algorithm and were able to follow patients throughout their treatment, allowing for an overall description of our patient cohort and their outcomes. Our inclusion criteria were set to capture the cohort that had presumed advanced EOC and describe their outcomes. The success of this initiative is due to multidisciplinary collaboration to improve surgical outcomes. Limitations to the study include the retrospective analysis of the algorithm, which potentially led to selection bias. Additionally, there was no randomization process and it was a single-institution study. Future applications of the algorithm could incorporate a multi-institutional project.

In conclusion, the use of a multimodal algorithm with a dedicated preoperative radiologic and clinical assessment tool and reflex laparoscopic evaluation in this cohort of patients with advanced EOC led to excellent surgical outcomes in most patients, with modest use of diagnostic laparoscopy and a very low rate of futile laparotomy. While universal diagnostic laparoscopy has been suggested as a triage tool, these data suggest that an individualized approach with multidisciplinary input is feasible and allows for more precise resource use, without compromising outcomes. Further investigation into the impact of this tool as an alternative to universal diagnostic laparoscopy on patient morbidity and resource use is warranted.

Highlights.

• A multimodal algorithm combining clinical and radiologic data with diagnostic laparoscopy for advanced ovarian cancer

• A multimodal preoperative algorithm leads to precise resource utilization without compromising outcomes

• Excellent surgical outcomes with a modest utilization of diagnostic laparoscopy and a very low rate of futile laparotomy