External validation of a prognostic nomogram for overall survival in women with uterine leiomyosarcoma

Authors


  • Presented in part as an abstract at the Annual Meeting of the American Society of Clinical Oncology; June 1-5, 2012; Chicago, Illinois.

Abstract

BACKGROUND:

There is no validated system to identify prognostically distinct cohorts of women with uterine leiomyosarcoma (ULMS). By using an independent, pooled, multi-institutional, international patient cohort, the authors validated a recently proposed ULMS nomogram.

METHODS:

The ULMS nomogram incorporated 7 clinical characteristics (age, tumor size, tumor grade, cervical involvement, locoregional metastases, distant metastases, and mitotic index (per 10 high-power fields) to predict overall survival (OS) after primary surgery. Independent cohorts from 2 sarcoma centers were included. Eligible women, at minimum, underwent a hysterectomy for primary, locally advanced, or metastatic ULMS and received part of their care at 1 of the centers between 1994 and 2010.

RESULTS:

In total, 187 women with ULMS were identified who met the above criteria described above (median age, 51 years; median tumor size, 9 cm; median mitotic index, 20 per 10 high-power fields). Tumors generally were high grade (88%), FIGO stage I or II (61%) without cervical involvement (93%) and without locoregional metastases (77%) or distant metastases (83%). The median OS and the 5-year OS rate were 4.5 years (95% confidence interval, 3.2-5.3 years) and 46%, respectively; and 65 women (35%) remained alive at last follow-up. The nomogram concordance index was 0.67(standard error, 0.02), which was as high as the concordance index from the initial cohort used for nomogram development. The concordance between actual OS and nomogram predictions suggests excellent calibration because predictions were within 1% of actual 5-year OS rates for patients with a predicted 5-year OS of less than 0.68.

CONCLUSIONS:

The ULMS nomogram was externally validated using independent cohorts. These findings support the international use of the ULMS nomogram prognostic of OS in ULMS. Cancer 2013. © 2012 American Cancer Society.

INTRODUCTION

Uterine leiomyosarcoma (ULMS) is a rare malignancy with an annual incidence of 0.64 per 100,000 women.1 It accounts for less than 5% of all uterine malignancies and approximately 30% of all uterine sarcomas.2 Although most frequently diagnosed while still confined to the uterus, the clinical course of the disease is difficult to predict, and the overall prognosis is poor. Current staging systems fail to identify which patients are at highest risk for death; thus, it is difficult to select patients for testing potentially beneficial adjuvant strategies.3-6 Neither the International Federation of Gynecology and Obstetrics (FIGO) nor the American Joint Committee on Cancer (AJCC) staging systems for uterine cancer accurately distinguish patients with ULMS into prognostically distinct groups.7, 8 To address this, FIGO recently developed a new classification system specifically for ULMS to include variables such as tumor size, extrauterine spread, and invasion of abdominal tissues.9

Nomograms are prognostic models that provide a score or a probability that a patient will be without disease or will be alive within a specific time frame.10 Nomograms have been used increasingly in the clinical setting to provide patient counseling.11, 12 Because there is still confusion around ULMS staging systems,7, 8 the Memorial Sloan-Kettering Cancer Center (MSKCC) recently developed a novel ULMS-specific nomogram for predicting postresection 5-year overall survival (OS) using 7 clinicopathologic characteristics.13 The utility of the nomogram is that it can more accurately distinguish groups of patients in terms of prognosis compared with either staging system alone. It may be used to provide patients with reasonable expectations. In addition, it may serve as a tool for discussing potential treatments; for identifying patients who have similar risk in order to attain more uniform cohorts within any future study; and, in particular, for identifying patients with high-grade tumors who may benefit from enrollment in clinical trials. The objective of the current study was to validate the MSKCC ULMS-specific nomogram for predicting postresection 5-year OS using an external, independent, international cohort.

MATERIALS AND MATERIALS

The ULMS nomogram incorporates 7 clinical characteristics to predict OS after primary surgery: age at diagnosis, tumor size, tumor grade, involvement of the cervix, locoregional metastases (including direct extrauterine spread or locoregional lymph node), distant metastases (including omentum, peritoneum, bowel serosa, lung, or liver), and mitotic index (mitoses per 10 high-power fields [HPF]). Independent cohorts from 2 international sarcoma centers (Brigham and Women's Hospital/Dana-Farber Cancer Institute [BWH/DFCI], Boston, Mass; and European Institute of Oncology [EIO], Milan, Italy) were included as the “validation cohort.” Eligible women were those who underwent hysterectomy at the institutions between 1994 and 2010. Women with locally advanced or metastatic disease who underwent more extensive surgery were included if the primary tumor (uterus) was also resected, although they did not have to be rendered disease-free. Women who underwent previous resection of the primary tumor or recurrences at other institutions were included if they received follow-up care at 1 of the 2 sarcoma centers. Women who presented with unresectable disease who never underwent surgery and those who had insufficient information on any of the nomogram variables were excluded. The cohort included patients from BWH/DFCI who were reported previously in a study that evaluated the predictive value of the FIGO and AJCC staging systems.7 Institutional review board approval was obtained from each institution for this retrospective data collection.

Histologic tumor grade was determined by microscopic analysis and was based on the assessment of 3 key histologic parameters in the primary uterine tumor: significant cytologic atypia, the mitotic rate, and the presence of coagulative tumor cell necrosis. Cytologic atypia in the form of nuclear enlargement, variation in nuclear shape, and prominent hyperchromasia was considered significant if it could be discernible at low-power magnification (10× objective). Morphologically high-grade tumors had moderate-to-marked cytologic atypia, tumor cell necrosis, and mitotic rates in excess of 10 per 10 HPF. Intermediate grade tumors lacked necrosis but had moderate cytologic atypia and mitotic rates >10 per 10 HPF. Low-grade tumors had no to minimal cytologic atypia but had coagulative tumor cell necrosis and a variable mitotic rate.14

Outcomes

The ULMS nomogram was designed to be prognostic of the 5-year OS rate. OS was calculated from the date of surgery to the date of either last follow-up or death. To compare the predictions of the nomogram with actual outcomes, OS probabilities were estimated using the Kaplan-Meier method and were compared with the nomogram's calculated probabilities.

Statistical Considerations

The performance of the nomogram in the validation cohort was assessed through estimates of discrimination and calibration. The nomogram's predictive accuracy and its ability to separate patients with different outcomes were measured using the concordance probability estimate (CPE).15 CPE quantifies the level of concordance between predicted probabilities and the actual chance of having the event of interest. The CPE denotes the probability that, of 2 randomly selected patients, the patient who survives longer will have a higher survival probability obtained from the nomogram than the patient with shorter survival. Calibration assesses how far predictions are from actual outcomes. The calibration of the nomogram was assessed using a calibration plot by plotting the nomogram's predicted 5-year survival probability against the patient-observed or actual probability as calculated by the Kaplan-Meier method for 4 subcohorts. Each subcohort was obtained by ranking patients based on their predicted probabilities and dividing the entire validation cohort into quartiles. A perfectly accurate nomogram would result in a plot in which the observed and predicted probabilities for given groups fall along the 45-degree line. The distance between the pairs and the line of unity (along the 45-degree line) is a measure of how far the nomogram's predictions are from actual outcomes. The 7 nomogram factors described in the original publication proposing the nomogram were collected and grouped in the same way. Patients' baseline characteristics were compared using the Wilcoxon test for continuous variables and the chi-square test for categorical variables. All tests were 2-sided. Statistical analyses were performed using the SAS software package (SAS Institute, Inc., Cary, NC)and SPlus software (S-Plus version 2000 Professional; Microsoft Corporation, Redmond, Wash) with the Design and Hmisc libraries16 and library PHCPE in R version 2.12.2 (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Patients

One-hundred eighty-seven eligible patients were included in this study in the validation cohort (129 women from BWH/DFCI and 58 women from EIO). Characteristics of patients in the original cohort from MSKCC who were used to construct the nomogram (the “model-development cohort”) and in the validation cohort are provided in Table 1. The median age for the validation cohort was 51 years, median tumor size was 9 cm, and median mitotic index was 20 per 10 HPF. Tumors generally were morphologically high grade (88%) FIGO stage I or II (61%) tumors without cervical involvement (93%) and without locoregional metastasis (77%) or distant metastases (83%) at initial presentation. The validation and model-development cohorts were very similar in terms of patient and pathologic characteristics. The mitotic index had a wider range but a smaller median in the model-development cohort compared with the validation cohort. The model-development and validation cohorts were similar in terms of disease stage and the number of patients with metastatic disease. However, a 2-way comparison of patient characteristics revealed significant differences between patients from EIO versus MSKCC and from EIO versus BWH/DFCI (Table 2). Patients from EIO were younger (median age, 45 years), had smaller tumors (median size, 7 cm), and had lower mitotic indices (median, 11.5 per 10 HPF). Fewer patients from EIO had high-grade tumors (76% vs 91%) or locoregional metastasis (14% vs 28%) compared with patients from the MSKCC or BWH/DFCI subsets.

Table 1. Characteristics of Patients in the Model-Development and Validation Cohorts
 No. of Patients (%)
CharacteristicModel-Development Cohort, N = 185Validation Cohort, N = 187
  1. Abbreviations: FIGO, International Federation of Gynecology and Obstetrics; HPF, high-power fields.

Age: Median [range], y51 [23-81]51 [26-81]
Tumor size: Median [range], cm9 [0.9-28]9 [2-30]
Tumor grade
 High169 (91)165 (88)
 Low16 (9)22 (12)
Cervical involvement
 Yes21 (11)14 (7)
 No164 (89)173 (93)
Locoregional metastasis
 Yes52 (28)43 (23)
 No133 (72)144 (77)
Distant metastasis
 Yes34 (18)32 (17)
 No151 (82)155 (83)
FIGO stage
 I107 (58)105 (57)
 II14 (8)7 (4)
 III31 (17)30 (16)
 IV33 (18)41 (22)
Mitotic index: Median [range], mitoses per 10 HPF15 [1-200]20 [1-97]
Table 2. Two-Way Comparison of Patient Characteristics Within the 3 Different Hospitals/Cancer Centers
 P
CovariateModel-Development Cohort (MSKCC) vs Validation Cohort (BWH/DFCI, EIO)BWH/DFCI vs MSKCCEIO vs MSKCCEIO vs BWH/DFCI
  1. Abbreviations: BWH/DFCI, Brigham and Women's Hospital/Dana-Farber Cancer Institute; EIO, European Institute of Oncology; FIGO, International Federation of Gynecology and Obstetrics; MSKCC, Memorial Sloan-Kettering Cancer Center; NS: not significant.

AgeNSNS.003.0001
SizeNSNS.03.0031
GradeNSNS.002.0004
Cervical involvementNSNS.07NS
Locoregional metastasisNSNS.027.045
Distant metastasisNSNSNSNS
Mitotic index.0012< .0001NS< .0001
FIGO stageNSNSNS.01

Nomogram Concordance

For the validation cohort, the median survival was 4.5 years (95% confidence interval [CI], 3.2-5.3 years), and the 5-year OS rate was 46% (Fig. 1). Sixty-five women (35%) were alive at last follow-up. The median follow-up was 6.6 years (range, 0.25-19 years). Figure 2 displays the published ULMS nomogram. The nomogram includes age at diagnosis, tumor size, tumor grade, involvement of the cervix, locoregional metastases (direct extrauterine spread, locoregional lymph node spread), distant metastases, and the mitotic index.

Figure 1.

This is the Kaplan-Meier overall survival curve for the validation cohort (N = 187) along with 95% confidence intervals.

Figure 2.

This is the Memorial Sloan-Kettering Cancer Center uterine leiomyosarcoma nomogram, which uses 7 readily available clinical characteristics to predict 5-year overall survival. The mitotic index (measured as the number of mitoses per 10 high-power fields) was modeled using logarithmic transformation; the values here are displayed in the original scale. Instructions: Locate patient's variable on the corresponding axis. Draw a line to the Points axis. Sum the points. Draw a line from the Total Points axis to the 5-year overall survival probability axis. (Reprinted with permission from Zivanovic O, Jacks LM, Iasonos A et al. A nomogram to predict postresection 5-year overall survival for patients with uterine leiomyosarcoma. Cancer. 2012;118:660-66913.)

The nomogram concordance index was 0.67 (95% CI, 0.62-0.72), which was as high as the concordance index from the model-development cohort (0.67; 95% CI, 0.63-0.72). The concordance between actual OS and nomogram predictions suggests excellent calibration of the nomogram in the validation cohort. The calibration plot illustrates how the nomogram predictions compare with the actual outcomes of the validation cohort (Fig. 3). Predictions were within 1% of actual 5-year OS rates, except for patients who had a predicted 5-year OS probability >0.68. For this favorable prognostic group, on average, the nomogram prediction was 0.69, whereas the actual proportion of patients who remained alive at 5 years was 0.80.

Figure 3.

This is a calibration curve for predicted (using the Memorial Sloan-Kettering Cancer Center nomogram) and observed 5-year overall survival (OS). The x-axis indicates the prediction obtained from the nomogram, and the y-axis indicates the observed 5-year OS rate. Patients were divided into 4 equal groups based on their predicted risk. The line of unity, at 45 degrees, represents ideal agreement between actual and predicted probabilities. The line with error bars represents the performance of the nomogram when applied to the validation cohort. The vertical bars represent the 95% confidence interval of the observed 5-year OS rate.

Nomogram Versus FIGO Staging

Figure 4 illustrates the nomogram-predicted 5-year OS probability presented separately within each FIGO stage. Across all 4 FIGO stages, there is a wide range of predictions, as illustrated from the histograms, indicating heterogeneity within each stage. Patients with FIGO stage I ULMS have a nomogram-predicted 5-year OS probability ranging from 0.2 to 0.9 when taking into account the nomogram characteristics; whereas patients with FIGO stage IV ULMS have a nomogram-predicted 5-year OS probability ranging from 0 to 0.75.

Figure 4.

Nomogram-predicted 5-year survival probability is illustrated for each 1988 International Federation of Gynecology and Obstetrics (FIGO) stage group. Colors indicate different nomogram-predicted probabilities.

DISCUSSION

Nomograms have received increased attention within clinical oncology, and many nomograms have been developed and validated for various malignancies.17-20 Recently, several nomograms specific for gynecologic malignancies, such as endometrial,11 ovarian,12 and uterine cancers,21, 22 have been developed that have superior accuracy compared with staging systems alone. Raut et al.7 and Zivanovic et al.8 illustrated the poor performance of the FIGO and AJCC staging classifications for predicting OS in patients with ULMS. ULMS is a mesenchymal malignancy that has biology and a natural history distinct from epithelial malignancies. Therefore, it is not surprising that staging systems developed for epithelial malignancies, such as endometrial cancer or other nonleiomyosarcoma uterine sarcomas, were invalid in patients with ULMS. In addition, those studies demonstrated that patients who had ULMS classified as stage I and II according to FIGO or AJCC had quite disparate clinical outcomes among patients within each stage. This is in contrast to other disease settings; for example, in endometrial cancer, patients who have early stage disease have similar clinical outcomes, and patients with advanced-stage disease have widely differing outcomes.21 Hence, there was a need for an individualized prognostic tool that could be tailored specifically to patients with ULMS.

Although a soft tissue sarcoma nomogram that predicts 12-year sarcoma-specific death has been developed,23 that nomogram attempts to encompass many histologies and anatomic sites of sarcoma. In addition to being both histology-specific and location-specific, the ULMS-specific nomogram validated here includes the additional prognostic information of tumor size as a continuous (rather than categorical) variable, extrauterine spread, and mitotic index. The current nomogram predicts 5-year OS rather than 12-year disease-specific death, which is a more commonly used endpoint in assessing prognosis both in patient counseling and in clinical trial development.

The goal of this nomogram is not to determine response to therapy. Forthcoming molecular diagnostic data hopefully will serve both as a diagnostic tool and as a predictor of response to therapy, which are important to individualized patient care. Instead, the goal of the nomogram is to allow the clinician evaluating a particular patient to determine that individual's prognosis more accurately than existing staging systems. In turn, this allows clinicians to more accurately distinguish low-risk patients with a good prognosis who may be followed more conservatively compared with uniformly high-risk patients with a poor prognosis who may be included in future clinical trials.

The predictions from the nomogram were within 1% of actual OS outcomes, except for the most favorable prognosis group; in the latter group, patients who had a predicted 5-year OS >68% had greater variability in the prediction. In evaluating this group further, we noted that patients with low-grade ULMS had a nomogram-predicted 5-year OS of 0.68 to 0.80, whereas patients with high-grade ULMS had a nomogram-predicted 5-year OS of 0.55 to 0.66. Thus, the nomogram-predicted 5-year OS for patients with high-grade ULMS potentially may be lower than their actual outcome. The mitotic index and tumor grade had the first and second most significant effect on the nomogram predictions, respectively. However, there is subjectivity in assessing both of these covariates, and a comparison of patients' characteristics indicated differences between the centers both in the proportion of patients with high-grade tumors and in the range of values for mitotic index. In general, patients in the favorable prognosis group were younger, with low-stage disease, a lower mitotic index, smaller tumors, no cervical involvement, and no locoregional or distant metastasis; some of these patients lived only 1 year, and others lived as long as 10 years. The nomogram could not distinguish these outliers, but it combined several covariates to give them individual predictions. If the calibration plot consisted of 3 groups instead of 4 (Fig. 5), then the sample size within each group would become larger, and predictions would be within 5% of actual outcomes for all groups, which is reasonable and comparable to published nomograms.24

Figure 5.

This is a calibration curve similar to that shown in Figure 3; here, patients are grouped into 3 equal groups based on their predicted risk. The x-axis indicates the prediction obtained from the uterine leiomyosarcoma nomogram, and the y-axis indicates the observed 5-year overall survival (OS) rate. KM indicates Kaplan-Meier.

In the current study, we externally validated the MSKCC nomogram for patients with ULMS using an international, multicenter cohort of patients with ULMS. When applied to this external patient cohort, the nomogram predictions were comparable to the actual observed OS, demonstrating that the nomogram reliably predicts outcome for patients with ULMS. These findings support the international use of this nomogram for predicting OS in patients with ULMS.

A potential concern in the development and validation of postoperative nomograms is that they do not account for the different treatments individual patients may have received. Some patients who were included in the model-development cohort received postoperative treatment as part of either standard or protocol therapy. At primary presentation, all patients in the validation cohort underwent surgery, 61 patients (33%) received chemotherapy, 15 patients (8%) received radiation therapy, 3 patients (1.6%) received neoadjuvant chemotherapy, and 6 patients (3%) received hormone therapy. However, postresection treatment was not included as a covariate in the nomogram development, because the goal of this nomogram is to provide prognostic information for patients and physicians at the time of diagnosis, when treatment details are not known. In addition, OS did not differ significantly between patients who did or did not receive postresection treatment in the model-development cohort.13 Treatment modalities may vary, depending on hospital type, location, and different referral patterns. Our observation that the nomogram accurately predicted outcomes in a validation cohort, which included patients from 2 independent academic centers over a 16-year period, further supports the strength and generalizability of this nomogram to patients with resected ULMS.

FUNDING SOURCES

No specific funding was disclosed.

CONFLICT OF INTEREST DISCLOSURES

The authors made no disclosures.

Ancillary