Epithelial ovarian cancer (EOC) is the leading cause of death from gynecological cancers in Western countries. Approximately 20% to 30% of patients with early-stage disease and 50% to 75% of those with advanced disease who obtain a complete response after first-line chemotherapy will ultimately develop recurrent disease.1
In this setting, the optimal management for patients at the time of recurrence is still unknown. For primary ovarian cancer, standard therapy is now well established, consisting of complete cytoreductive surgery and chemotherapy.2–4
The role of surgery in recurrence is currently debated, with the following 2 main points of contention: the prognostic value of complete resection at the time of recurrence and the selection criteria for surgery.
Multiple retrospective studies have evaluated the outcomes of women undergoing secondary cytoreduction (SCR), with survivals ranging from 20 to 63 months with optimal surgery2,5–8 and a higher survival than in patients treated with chemotherapy alone.9–11 In 2005, a review of the literature suggested that optimal resection could have the same value as in primary disease.12
In patients for whom complete reduction cannot be obtained, however, extensive surgery with significant morbidity could be harmful, pointing to immediate chemotherapy as a more appropriate modality.
Two main studies have tried to develop a risk model for predicting complete SCR in patients with recurrent EOC (REOC).
In 2006, the AGO DESKTOP OVAR trial identified preoperative predictors of complete surgery,13 with a positive predictive value (PPV) of 79% and a negative predictive value (NPV) of 58%. More recently, Tian et al14 developed a model to predict complete resection in REOC, with an 80.4% sensitivity and a 52.6% specificity.
To date, no study has compared these 2 scores both in prediction capacity and impact on survival.
The aims of this study were to assess the performance of these models in an independent population and to evaluate the impact of complete resection on survival for REOC.
Between January 1, 2000, and December 31, 2010, all patients who were diagnosed with a first recurrence of EOC but without any minimal intervals after the end of the initial treatment in 2 French centers were included (N = 194; Hôpital Tenon, Paris [n = 108], and Institut Claudius Regaud Toulouse [n = 88]). Data were prospectively entered into a database and analyzed retrospectively for this study. Patients with nonepithelial tumor histology or with a second or more subsequent recurrences were excluded. We also excluded patients with missing data on recurrence management. Recurrence was diagnosed at clinical examination, imaging (eg, computed tomography, magnetic resonance imaging, positron emission tomography), and/or CA125 rate. All patients provided their written consent for the data analysis.
The following variables were recorded: patient characteristics (age, body mass index, and medical history), tumor characteristics (International Federation of Gynecology and Obstetrics [FIGO] stage, grade, and histological type), history of initial treatment (including the results of the first surgery), relapse-free interval, CA125 level at recurrence, mode of recurrence diagnosis, imaging at recurrence, treatment of recurrence (including the results of the surgery for recurrence: complete or not), and outcome of the disease.
In the absence of specific recommendation for recurrence, management of recurrence was determined using clinical criteria (especially the general condition and disease-free interval) and imaging (number of sites and localization), adapted to each patient case. The patients eligible for surgery underwent, whenever possible, an initial laparoscopy with macroscopic evaluation of resectability. If complete resection could be assumed, laparotomy and cytoreductive surgery were then performed. The DESKTOP and Tian models were not used at this time to decide surgery.
After surgery, resection was defined as complete if no macroscopically residual disease was present and incomplete otherwise (CC0 according to Sugarbaker15).
The DESKTOP OVAR and Tian model scores were then assessed for our patients.
DESKTOP Trial Score
The AGO DESKTOP OVAR trial13 (2006) retrospectively analyzed 267 patients from 35 centers who underwent surgery for a first recurrence of EOC. Variables associated with complete resection (ie, positive score) were performance status (PS) (Eastern Cooperative Oncology Group [ECOG] 0), history of complete resection (no macroscopic residual tumor) during the first intervention, and absence of ascites of more than 500 mL. Positive score predicts resectability.
Tian et al14 Model
More recently, another model was developed using 1075 patients from 6 centers.14 In addition to the DESKTOP trial model, this model included scores using the following variables: disease-free interval (±16 months), CA125 rate (±105 IU/mL), and initial disease stage (FIGO stage ≥ III). These variables were entered into a risk model and assigned scores ranging from 0 to 11.9. The patients with total scores of 0 to 4.7 were categorized into the low-risk group, which showed a significantly higher proportion of complete cytoreduction than in the high-risk group.
For each model, we compared sensitivity and specificity, PPVs, and NPVs for the threshold reported. Positive predictive value corresponds to the probability of being resectable when the score predicts so, and NPV corresponds to the probability of not being resectable when the score predicts so.
Performance regarding the Tian model was quantified with respect to discrimination and calibration. Discrimination (ie, whether the relative ranking of individual predictions was in the correct order) was quantified with the area under the receiver operating characteristic curve (AUC).
In the second step, we analyzed survival regarding management (surgery or chemotherapy), resection status at recurrence (if surgically managed), and the model or score status. We first classified pretherapeutic scores and actual treatment into the following 4 groups: for the DESKTOP scores, (1) pretherapeutic positive score and surgery, (2) pretherapeutic positive score and no surgery, (3) pretherapeutic negative score and surgery, and (4) pretherapeutic negative score and no surgery; for the Tian scores, (1) pretherapeutic low risk and surgery, (2) pretherapeutic low risk and no surgery, (3) pretherapeutic high risk and surgery, and (4) pretherapeutic high risk and no surgery. Subsequently, we analyzed all surgically managed patients in relation to their scores and resection status via the following 4 categories: for the DESKTOP scores, (1) positive score and complete resection, (2) positive score and incomplete resection, (3) negative score and complete resection, and (4) negative score and incomplete resection; for the Tian scores, (1) low risk and complete resection, (2) low risk and incomplete resection, (3) high risk and complete resection, and (4) high risk and incomplete resection.
Univariate survival analyses were performed using Kaplan-Meier curves and log-rank tests. Date of recurrence was considered to be the beginning of follow-up, whereas endpoint corresponded to death. We used a Cox model adjusted for potentially confounding factors.
All analyses were performed using the R software package (http://lib.stat.cmu.edu/R/CRAN/).
Of the 194 patients with REOC in the initial population, 189 patients had no missing data on recurrence management and formed our study population. A total of 162 (85.7%) had all available items for the DESKTOP OVAR trial score, whereas 147 (77.7%) met the criteria for the Tian model. Eighty-one patients were eligible for surgery according to our centers’ criteria, and 78 (40.2%) underwent complete surgical management for their recurrence (eg, debulking surgery). Forty-nine patients (58.8%) had a primary laparotomy, without any exploratory laparoscopy. In the 35 patients with initial laparoscopy for resectability evaluation, 20 (57%) failed because of peritoneal adhesion and had a laparotomy without any exploration. Eleven had an exploratory laparoscopy followed by a laparotomy, and only 3 patients did not have any complementary laparotomy because of an extensive carcinomatosis. Because 7 had no information regarding resection status, our validation population was composed of 74 patients (Appendix). The rate of complete resection was 73% (n = 54). To compare, in primary treatment in our center, 30% of the patients had neoadjuvant chemotherapy and 74.3% of the patients underwent complete resection after initial surgery (70% regardless of initial or secondary surgery).
The clinicopathological and surgical characteristics of the patients are reported in Table 1.
Both surgically and nonsurgically managed populations did not differ in regard to their age, tumor histology, number of sites of recurrence, or presence of ascites at recurrence. The surgically managed patients had a longer median disease-free interval (20 months vs 11.3 months, P < 0.001), were more likely to have undergone complete resection during initial surgery (65.4% vs 50.0%, P = 0.04), and were mostly in good condition (ECOG 0, 85.9% vs 62.6%, P < 0.001). Of note, 29.6% of the patients did not receive chemotherapy at recurrence: most of the cases had single site of recurrence.
Relevance of Scores
In our study, 31.7% of the patients would be eligible for surgery according to the DESKTOP model and 78.9% would be eligible for surgery according to the Tian model.
The surgically managed patients were equally issued from negative or positive DESKTOP scores (52% vs 48%) but were more often from Tian’s low-risk group (71%).
The use of DESKTOP scores on our population showed a similar PPV to that of the original population (80.6% vs 79%), although we had a higher false-negative rate (65.4% vs 42%). Similarly, we had a better PPV with the Tian model than in the original study (74% vs 54% in the original study from Tian et al), although we noted a significantly higher false-negative rate (71.4% vs 20%) (Table 2).
Of the 3 patients deemed unresectable after exploratory laparotomy, 2 were predicted as unresectable by both DESKTOP and Tian scores. One was DESKTOP positive (missing score for Tian score).
The median follow-up duration was 27.9 months. The median overall survival in the nonsurgically managed group was 20 months (95% confidence interval [CI], 16.7–24.5) versus 41.5 months (95% CI, 36.9-infinite) in the surgically managed group. In the univariate analysis, the difference was statistically significant (log-rank P < 0.001, HR, 0.43; 95% CI, 0.28–0.67) (Fig. 1A). In the multivariate analysis, progression-free interval of 16 months or less (HR, 2.14; 95% CI, 1.27–3.60; P = 0.004), initial resection status (HR, 1.70; 95% CI, 1.07–2.70; P = 0.02), ascites (HR, 1.94; 95% CI, 1.12–3.36; P = 0.01), and CA125 rate at recurrence (HR, 2.40; 95% CI, 1.26–4.58; P = 0.007) were significant; only surgery failed to show statistical significance (HR, 0.92; 95% CI, 0.53–1.62; P = 0.7).
However, we observed significantly higher overall survival in the group with complete resection (59.4 vs 17.9 months, Fig. 1B, log-rank P < 0.01). This difference persisted after adjustment for confounding variables (HR, 2.53; 95% CI, 1.01–6.3; P = 0.04).
The DESKTOP score failed to predict appropriate survival; the median survival after complete surgery in the resectable and nonresectable DESKTOP groups was 39.5 months and 59.4 months, respectively (P = 0.08) (Fig. 2A). On the contrary, the high-risk groups according to the Tian model had a median survival of 41.5 months, whereas it was not reached in the low-risk group (P = 0.02) (Fig. 2B).
Differences in survival were analyzed according to both scores and initial management of recurrence (surgery or no surgery) (log-rank P < 0.001 for both scores, Fig. 3). As expected, a significantly lower survival was observed in the patients with negative DESKTOP scores or Tian’s high-risk scores as well as in the patients with an initial decision of no surgical management (HR, 2.9; 95% CI, 1.6–5.4, and HR, 4.2; 95% CI, 2.2–8.0, P < 0.001, respectively). An increased survival was observed in the patients with predicted complete resection with the DESKTOP score but without surgical management. However, the hazard ratios were not statistically significant (HR, 0.4; 95% CI, 0.15–1.3; P = 0.14, after adjustment for confounding factors).
This study aimed to validate the relevance of various clinical criteria regarding patient selection for second surgical resection associated with EOC. In our French sample, currently available models (DESKTOP OVAR and Tian) showed good PPVs (80.6% and 74%, respectively) but unacceptable false-negatives rates (65.4% and 71.4%, respectively) because it missed potentially meaningful debulking.
These results confirmed that surgery should be performed in patients with scores predicting resectability. However, in patients with scores predicting nonresectability, other criteria are needed to decide eligibility for surgery.
As for the initial management of EOC, recurrences are effectively managed by complete surgical resection.12,13,16 Our data are also consistent with this assumption. In practice, however, patient selection for surgery is difficult given the lack of evidence for extensive surgery and significant morbidity associated with the procedure. As such, most recurrences are currently treated with chemotherapy alone.17
Using simple criteria, the DESKTOP score sought to identify 75% of the patients for whom complete resection could be achieved. This score has been the subject of prospective external validation, with good results (DESKTOP 2).18 In our population, PPV was 80%, corresponding to the objectives set by the DESKTOP authors. However, this score also neglected nearly two thirds of the false negatives who could favorably undergo surgical resection.
Similarly, the model developed by Tian et al and concurrently validated in a separate population predicted complete resection in 75% of the low-risk patients in our population. However, the Tian scores neglected 71% of false negatives, compared with only 20% in the original publication. Moreover, discrimination was ineffective in our study population (AUC, 0.57), and the resection rates were nearly the same regardless of the result of the score (74.0 vs 71.4%), limiting widely its applicability.
With both scores, patient management based on their initial scores (ie, a score predicting the failure of resection was associated with no surgical management) was associated with a significantly worse survival. In contrast, patient survival in those with favorable scores who underwent surgery was no different than in the patients without surgical management.
In both models, complete resection was the better predictor of survival. The patients for whom complete resection could be obtained, despite model predictions, displayed better survival than did the patients for whom resection had failed, contrarily to the optimistic predictions of the scores. There is therefore a real need to precisely differentiate these patients to offer surgery only to those patients for whom the procedure would provide measurable benefit.
A possible explanation for the aforementioned discrepancies is, in addition to the lack of power in our data, the preselection of patients for surgery. In both models, as in our study, the data were retrospective and the models were developed for patients already selected for surgery, according to the decision criteria specific to each location, although this was not specified in the methodology. The rate of complete resection ranged from 40.4% to 49.8% in the Tian and DESKTOP studies, respectively. The rate of complete resection in our series was 73%. Maybe our selection criteria were more stringent, explaining this better resection rate, and could limit part of score applicability in our population. In both models, intraoperative findings largely improved the model performance (AUC improved from 0.72 to 0.8 for Tian, after inclusion of tumor characteristics during surgery; false-negative rates improved from 42% to 23% for DESKTOP, if surgery was abandoned because of laparoscopic visualization of peritoneal carcinomatosis). However, laparoscopy may be impractical in the context of recurrence, mainly because of postoperative adhesions related to the first intervention, leading to random evaluations of the tumor mass. In our study, we renounced laparoscopy for an immediate laparotomy because of adhesion in more than half of the patients (57%).
Furthermore, the goal of these models was to identify patients before any surgery.
Some factors predicting complete resection are now well known and used in practice, such as ascites19–21; residual disease at initial surgery5,6,20–22; ECOG PS5,13,14; number of sites of recurrence2,6,23,24; size of tumor recurrence5; and recurrence-free interval with a threshold at 12 months,2,22,25 16 months,14 or 18 months.6 Other parameters are still debated, such as the CA125 rate.24
However, other factors may also influence the outcome of surgery, such as intraoperative complications (surgical or anesthetic) or surgeon expertise.26 The latter, which may be a decisive variable, is difficult to detect using mathematical models.
Our study confirms that complete resection in the secondary surgery should be the element to privileging, and the Tian model, the AGO DESKTOP OVAR study, and the present study illustrate how survival is improved in patients optimally resected, including among those initially at high risk for nonresection. It will be essential for these scores to minimize as much as possible false negatives. Indeed, the AGO DESKTOP OVAR score was not developed to predict surgical outcome in patients with negative score, and our findings confirm these results.
In conclusion, the management of REOC remains unclear. If complete resection is the strongest survival determinant in patients with REOC undergoing SCR surgery, patient selection for such management becomes important. Mathematical models are one way to select these patients, although current models developed using retrospective data did not take into account the preselection decision, which was made upstream and was likely different from one center to another. Regarding the adoption of SCR as the standard of care for patients with REOC, an ongoing randomized trial is comparing randomizing chemotherapy alone versus chemotherapy and surgery to try to answer these questions (DESKTOP III, NCT01166737). Meanwhile, factors such as the general condition of the patient, disease-free interval, result of the initial surgery, and presence of ascites remain relevant indicators to help in decision making regarding surgery.
1. Gadducci A, Cosio S, Zola P, et al. Surveillance procedures for patients treated for epithelial ovarian cancer
: a review of the literature. Int J Gynecol Cancer
. 2007; 17: 21–31.
2. Chi DS, McCaughty K, Diaz JP, et al. Guidelines and selection criteria for secondary cytoreductive surgery
in patients with recurrent, platinum-sensitive epithelial ovarian carcinoma. Cancer
. 2006; 106: 1933–1939.
3. Hoskins WJ, McGuire WP, Brady MF, et al. The effect of diameter of largest residual disease on survival after primary cytoreductive surgery
in patients with suboptimal residual epithelial ovarian carcinoma. Am J Obstet Gynecol
. 1994; 170: 974–979; discussion 979–980.
4. Hunter RW, Alexander ND, Soutter WP. Meta-analysis of surgery
in advanced ovarian carcinoma: is maximum cytoreductive surgery
an independent determinant of prognosis? Am J Obstet Gynecol
. 1992; 166: 504–511.
5. Eisenkop SM, Friedman RL, Spirtos NM. The role of secondary cytoreductive surgery
in the treatment of patients with recurrent epithelial ovarian carcinoma. Cancer
. 2000; 88: 144–153.
6. Salani R, Santillan A, Zahurak ML, et al. Secondary cytoreductive surgery
for localized, recurrent epithelial ovarian cancer
: analysis of prognostic factors and survival outcome. Cancer
. 2007; 109: 685–691.
7. Oksefjell H, Sandstad B, Tropé C. The role of secondary cytoreduction in the management of the first relapse in epithelial ovarian cancer
. Ann Oncol
. 2009; 20: 286–293.
8. Tian W-J, Jiang R, Cheng X, et al. Surgery
in recurrent epithelial ovarian cancer
: benefits on survival for patients with residual disease of 0.1–1 cm after secondary cytoreduction. J Surg Oncol
. 2010; 101: 244–250.
9. Alberts DS, Liu PY, Wilczynski SP, et al. Randomized trial of pegylated liposomal doxorubicin (PLD) plus carboplatin versus carboplatin in platinum-sensitive (PS) patients with recurrent epithelial ovarian or peritoneal carcinoma after failure of initial platinum-based chemotherapy (Southwest Oncology Group Protocol S0200). Gynecol Oncol
. 2008; 108: 90–94.
10. Pfisterer J, Plante M, Vergote I, et al. Gemcitabine plus carboplatin compared with carboplatin in patients with platinum-sensitive recurrent ovarian cancer
: an intergroup trial of the AGO-OVAR, the NCIC CTG, and the EORTC GCG. J Clin Oncol
. 2006; 24: 4699–4707.
11. Power P, Stuart G, Oza A, et al. Efficacy of pegylated liposomal doxorubicin (PLD) plus carboplatin in ovarian cancer
patients who recur within six to twelve months: a phase II study. Gynecol Oncol
. 2009; 114: 410–414.
12. Harter P, Du Bois A. The role of surgery
in ovarian cancer
with special emphasis on cytoreductive surgery
. Curr Opin Oncol
. 2005; 17: 505–514.
13. Harter P, Du Bois A, Hahmann M, et al. Surgery
in recurrent ovarian cancer
: the Arbeitsgemeinschaft Gynaekologische Onkologie (AGO) DESKTOP OVAR trial. Ann Surg Oncol
. 2006; 13: 1702–1710.
14. Tian W-J, Chi DS, Sehouli J, et al. A risk model for secondary cytoreductive surgery
in recurrent ovarian cancer
: an evidence-based proposal for patient selection. Prognostic features of 51 colorectal and 130 appendiceal cancer patients with peritoneal carcinomatosis treated by cytoreductive surgery
and intraperitoneal chemotherapy. Ann Surg
. 1995; 221: 124–132.
15. 15. Sugarbaker PH, Jablonski KA. Prognostic features of 51 colorectal and 130 appendiceal cancer patients with peritoneal carcinomatosis treated by cytoreductive surgery
and intraperitoneal chemotherapy. Ann Surg
. 1995; 221: 124–132.
16. Zang RY, Harter P, Chi DS, et al. Predictors of survival in patients with recurrent ovarian cancer
undergoing secondary cytoreductive surgery
based on the pooled analysis of an international collaborative cohort. Br J Cancer
. 2011; 105: 890–896.
17. Loehr A, Harter P, Traut A, et al. Cytoreductive surgery
in recurrent ovarian cancer
. J Cancer Res Clin Oncol
. 2004; 130(suppl 1) (suppl 1). S122 (abstract).
18. Harter P, Sehouli J, Reuss A, et al. Prospective validation study of a predictive score for operability of recurrent ovarian cancer
: the Multicenter Intergroup Study DESKTOP II. A project of the AGO Kommission OVAR, AGO Study Group, NOGGO, AGO-Austria, and MITO. Int J Gynecol Cancer
. 2011; 21: 289–295.
19. Eltabbakh GH, Mount SL, Beatty B, et al. Factors associated with cytoreducibility among women with ovarian carcinoma. Gynecol Oncol
. 2004; 95: 377–383.
20. Zang R-Y, Li Z-T, Tang J, et al. Secondary cytoreductive surgery
for patients with relapsed epithelial ovarian carcinoma: who benefits? Cancer
. 2004; 100: 1152–1161.
21. Li Y-F, Li M-D, Liu F-Y, et al. How to increase the optimal rate of secondary cytoreductive surgery
in recurrent epithelial ovarian cancer
[in Chinese]. Ai Zheng
. 2003; 22: 1193–1196.
22. Tebes SJ, Sayer RA, Palmer JM, et al. Cytoreductive surgery
for patients with recurrent epithelial ovarian carcinoma. Gynecol Oncol
. 2007; 106: 482–487.
23. Gronlund B, Lundvall L, Christensen IJ, et al. Surgical cytoreduction in recurrent ovarian carcinoma in patients with complete response to paclitaxel-platinum. Eur J Surg Oncol
. 2005; 31: 67–73.
24. Boran N, Hizli D, Yilmaz S, et al. Secondary cytoreductive surgery
outcomes of selected patients with paclitaxel/platinum sensitive recurrent epithelial ovarian cancer
. J Surg Oncol
. 2012; 106: 369–375.
25. Tay EH, Grant PT, Gebski V, et al. Secondary cytoreductive surgery
for recurrent epithelial ovarian cancer
. Obstet Gynecol
. 2002; 99: 1008–1013.
26. Bristow RE, Tomacruz RS, Armstrong DK, et al. Survival effect of maximal cytoreductive surgery
for advanced ovarian carcinoma during the platinum era: a meta-analysis. J Clin Oncol
. 2002; 20: 1248–1259.