Objective
Recurrence prediction is a cornerstone of patient management for borderline ovarian tumors. This study aimed to develop a nomogram predicting the recurrence probability in individual patients who had received primary surgical treatment.
Study Design
This retrospective multicenter study included 186 patients with borderline ovarian tumor diagnosed from January 1980 through December 2008. A multivariate logistic regression analysis of selected prognostic features was performed and a nomogram to predict recurrence was constructed. The nomogram was internally validated.
Results
The overall recurrence rate was 34.4% (64/186), with noninvasive and invasive forms in 29% (54/186) and 5.4% (10/186) of cases, respectively. International Federation of Gynecology and Obstetrics stage, age at diagnosis, histologic subtype, completeness of surgery, and type of surgery (radical vs fertility sparing) were associated with an increased risk of recurrence and were included in the nomogram. The predictive model had a concordance index of 0.78 (95% confidence interval, 0.76–0.80) and 0.77 (95% confidence interval, 0.75–0.79) before and after the 200 repetitions of bootstrap sample corrections, respectively, and showed good calibration.
Conclusion
Our results support the use of the present nomogram based on 5 clinical and pathological characteristics to predict recurrence probability with a high concordance, hence to inform patients on surgical management. External validation is required to recommend this nomogram in routine practice.
Borderline ovarian tumors (BOTs) are defined by the presence of cellular proliferation and nuclear atypia without stromal invasion and represent 10-15% of epithelial ovarian tumors. Five- and 10-year survival for stage I, II, and III disease are 99% and 97%, 98% and 90%, and 96% and 88%, respectively. Despite this favorable prognosis, up to 25% of patients experience a recurrence. Several epidemiological and histological factors have been correlated with the risk of recurrence after surgical treatment of BOTs but none of them are sufficiently accurate in routine practice to identify patients at risk of recurrence.
The past decade has been marked by important advances in therapeutic options such as the advent of fertility-sparing surgery. The increased demand for clinicians to help patients make informed decisions regarding the options, benefits, and risks of treatment involves careful insertion of risk factors into the overall treatment recommendations. Predictive models have been developed across most cancer types. One of these predictive models is the nomogram that allows a simple graphical representation of a statistical predictive model that generates a numerical probability of a clinical event. As nomograms can generate individualized predictions, they can be used to stratify patients not only for planning treatment but also for providing better information to the patient about their therapeutic options. This can help clinicians to involve patients in the therapeutic decision-making process and may consequently improve their compliance. Recently, Obermair et al proposed a nomogram to predict the risk of recurrence in patients with BOTs. However, external validation revealed that it had low relevance which was probably linked to differences in the epidemiological, surgical, and histological characteristics of the 2 populations. Therefore, the objective of the current study was to develop a nomogram to predict the probability of recurrence for women with early and advanced stages of BOT.
Materials and Methods
Patients
We conducted a retrospective study using data from patients with BOT who had received primary surgical treatment from January 1980 through December 2008. For each patient, the following parameters were recorded: age, preoperative serum CA 125, type of surgery, completeness of surgery, fertility-sparing surgery, surgical route (laparoscopy or laparotomy), performance of lymphadenectomy, final histological subtype (serous or mucinous), International Federation of Gynecology and Obstetrics (FIGO) staging and occurrence of a recurrence, overall survival, and duration of follow-up.
Fertility-sparing surgery (conservative treatment) was defined as a procedure in which the uterus and at least part of 1 ovary were preserved (ie, unilateral salpingo-oophorectomy [USO] and cystectomy). Surgical treatment was considered nonconservative when bilateral salpingo-oophorectomy (BSO) was performed with or without associated hysterectomy. In line with current French guidelines, initial surgical staging was considered complete once all peritoneal surfaces had been carefully inspected and peritoneal washing, random or oriented multiple biopsies, and infracolonic omentectomy performed. Systematic appendectomy was also a criterion for complete staging of mucinous BOT. Initial surgical staging was considered incomplete in all other cases, independently of the radical or conservative nature. Pelvic and paraaortic lymphadenectomy was not systematically performed. Histological typing was performed according to the World Health Organization’s system. Follow-up included a combination of clinical examination, ultrasonography, and measurement of serum tumor markers. For the first 2 years, follow-up evaluation was performed every 6 months and then patients were evaluated annually.
Development and predictive accuracy of the model
A nomogram was developed to predict the risk of recurrence at 5 years with or without invasive component. A multivariate analysis was performed using the logistic regression model and including all the factors that were significant at univariate analysis. The complexity of the model was controlled using the Akaike information criteria. P < .05 was considered significant. The final model equation was then organized as a nomogram designed to calculate patient-specific probabilities of recurrence. Values for each of the model covariates were mapped to points on a scale ranging from 0-100, with total points obtained for each model covariate mapped to the probability of recurrence associated with that combination of covariate values. The predictive accuracy of the model was assessed by its discrimination and calibration. The area under the receiver operating characteristic curve (AUC) measures the model’s ability to discriminate between patients with or without recurrence. An AUC of 0.5 indicates that the model provides no predictive discrimination, while a value of 1.0 indicates perfect discrimination between cases with or without recurrence. Measures of predictive accuracy were validated using bootstrap simulation. The model was fit to 200 samples of equivalent size drawn at random with replacement from the original study population. The measures of predictive performance obtained for each statistic in the bootstrap samples were used to estimate the bias in the model statistics attributable to overfitting. Calibration was assessed using plots that overlay model predicted probabilities with the actual probabilities. Results from the bootstrap simulation were also used to calculate the average and the maximal difference error between predicted values.
Additional statistical tests
The categorical variables were analyzed using the χ 2 test. Differences were considered significant at a level of P < .05. All analyses were performed using the R software with the rms, Presence/Absence packages ( http://lib.stat.cmu.edu/R/CRAN ).
Results
Description of the population
In all, 186 patients with primary surgical treatment for BOT were included in the study. The median time of follow-up (from diagnosis to recurrence or last date of follow-up) was 94.9 months (range, 60.00-207.3). Detailed patients’ characteristics at baseline are shown in Table 1 . Among the 186 patients, 37.1%, 3.8%, 6.9%, 12.9%, and 39.2% had FIGO stage Ia, Ib, Ic, II, and III, respectively.
| Parameters | Baseline characteristics | ||
|---|---|---|---|
| Overall cohort, n (%), n = 186 | No recurrence, n (%), n = 122 | Recurrence, n (%), n = 64 | |
| Mean age at diagnosis, mo (median) | 34.42 (31) | 35.9 (33.5) | 31.5 (29) |
| FIGO staging | |||
| Ia | 69 (37.09) | 54 (44.26) | 15 (23.43) |
| Ib | 7 (3.76) | 2 (1.63) | 5 (7.81) |
| Ic | 13 (6.98) | 8 (6.55) | 5 (7.81) |
| II | 24 (12.90) | 14 (11.47) | 10 (15.62) |
| III | 73 (39.24) | 44 (36.06) | 29 (45.31) |
| Histologic subtype | |||
| Serous | 123 (66.12) | 70 (57.37) | 53 (82.81) |
| Mucinous (intestinal + müllerian) | 63 (33.87) | 52 (42.62) | 11 (17.18) |
| Completeness of surgical staging | |||
| Complete | 81 (43.54) | 64 (52.45) | 17 (26.56) |
| Incomplete | 105 (56.45) | 58 (47.54) | 47 (73.43) |
| Surgical route | |||
| Laparoscopy | 41 (22.1) | 28 (22.95) | 13 (20.31) |
| Laparotomy | 42 (22.5) | 27 (22.13) | 15 (23.43) |
| NA | 103 (55.4) | 67 (54.91) | 36 (56.25) |
| Surgery procedure | |||
| BSO ± TH | 75 (40.32) | 59 (48.36) | 16 (25.00) |
| USO | 43 (23.11) | 39 (31.96) | 29 (45.31) |
| Cystectomy | 68 (36.55) | 24 (19.67) | 19 (29.68) |
| Lymphadenectomy | |||
| No | 141 (75.8) | 97 (79.50) | 44 (68.75) |
| Pelvic ± paraaortic | 33 (17.8) | 21 (17.21) | 12 (18.75) |
| NA | 12 (6.4) | 4 (3.27) | 8 (12.50) |
| CA 125 >35 IU/mL | |||
| No | 9 (4.83) | 4 (3.27) | 5 (7.81) |
| Yes | 177 (95.16) | 118 (96.72) | 59 (92.18) |
The histologic types were serous and mucinous in 66.1% (123/186) and 33.9% (63/186), respectively ( Table 1 ). In all, 105 patients (56.5%) underwent incomplete surgical staging. Lymphadenectomy was performed in 33 of the 186 patients (17.8%), all of which included both pelvic and paraaortic areas. In all, 111 patients (63.4%) underwent fertility-sparing surgery ( Table 1 ).
Recurrence rate and survival
The overall recurrence rate was 34.4% (64/186), with noninvasive and invasive forms in 29% (54/186) and 5.4% (10/186) of cases, respectively. Recurrences were diagnosed in 25% (16/64) of patients of the BSO group, 29.7% (19/64) of patients of the USO group, and 45.3% (29/64) of patients of the cystectomy group. Cystectomy was associated with a higher risk of recurrence compared to USO or BSO but without significant difference between these 2 groups. In these subgroups the median time to first recurrence was 54.3 months (range, 7.0–132.8), 49.8 months (range, 13.1–170.4), and 35.1 months (range, 4.2–145.0), respectively. The median follow-up for each surgical group was 78.2 months (range, 60.0–207.3), 98.6 months (range, 61.1–189.1), and 78.6 months (range, 60.0–138.6), respectively. No difference in median follow-up was noted according to the groups. For the whole population, overall survival at 5 years was 100% and 94.7% (95% confidence interval [CI], 88.9–100) at 10 years.
Recurrence-predicting model and nomogram validation
FIGO stage, age at diagnosis, histological subtype, completeness of surgery, and type of surgery were included in the logistic regression model ( Table 2 and Figure 1 ).
| Variable | Multivariate analysis | |
|---|---|---|
| OR (95% CI) | P value | |
| Age | 0.99 (0.95–1.02) | .365 |
| FIGO staging | ||
| Ia | Reference | |
| Ib | 0.34 (0.08–1.41) | |
| Ic | 1.17 (0.24–5.64) | |
| II | 2.12 (0.34–13.19) | |
| III | 2.59 (0.24–27.82) | .073 |
| Histologic subtype | ||
| Serous vs mucinous | 2.56 (0.77–8.52) | .126 |
| Surgical procedure | ||
| BSO ± TH | Reference | |
| USO | 9.43 (3.1–28.61) | |
| Cystectomy | 11.35 (4.01–32.08) | < .001 |
| Completeness of surgical staging | ||
| Incomplete vs complete | 2.08 (0.81–5.36) | .128 |
The predictive model had an AUC of 0.78 (95% CI, 0.76–0.80) and 0.77 (95% CI, 0.75–0.79) before and after the 200 repetitions of bootstrap sample corrections, respectively. The predicted probability obtained from the bootstrap correction and the actual probabilities of recurrence are shown in the calibration plot ( Figure 2 ). No significant difference was observed between the predicted percentages and the observed probabilities of recurrence ( P = 1.00), meaning that the nomogram was well calibrated. The average difference and the maximal difference in predicted and calibrated probabilities of recurrence were 0.02% and 7.9-15% respectively.
