A Nomogram to Predict Postresection 5-Year Overall Survival for Patients With Uterine Leiomyosarcoma

Abstract

BACKGROUND

The clinical course of patients with uterine leiomyosarcoma (LMS) is difficult to predict with the currently available categorical staging systems of the American Joint Committee on Cancer (AJCC) and the International Federation of Gynecology and Obstetrics (FIGO). The objective of the current study was to develop and validate a novel, clinically relevant, individualized prognostic model for patients with uterine LMS.

METHODS

Patients with uterine LMS who presented at the authors’ institution from 1982 to 2008 were analyzed. The nomogram model was chosen based on the clinical evidence and statistical significance of the predictors, including age at diagnosis, tumor size, histologic grade, uterine cervix involvement, extrauterine spread, distant metastases, and mitotic index. Five-year overall survival (OS) was the predicted endpoint. The concordance probability (CP) was used as a predictive accuracy measure and compared with the CP of current staging systems. The model was internally validated using 200 bootstrap samples to correct for over fitting.

RESULTS

One hundred eighty-five of 270 patients were eligible for the nomogram analysis. The median follow-up was 5.4 years, and the median OS was 3.75 years (95% confidence interval, 3-6 years). The CP of the newly developed nomogram was 0.67 (95% confidence interval, 0.63-0.72). This was superior to predictions based on AJCC and FIGO staging. The bootstrap-validated CP was 0.65 with good calibration accuracy.

CONCLUSIONS

The authors developed and internally validated a uterine LMS-specific nomogram to predict 5-year OS. This novel, individualized prognostic model outperforms traditionally used categorical staging systems and may be useful for patient counseling and for better selection of patients for adjuvant therapy trials.

Uterine leiomyosarcoma (LMS) is a rare neoplasm with an annual incidence of 0.64 per 100,000 women.¹ It accounts for <5% of all uterine malignancies and approximately 30% of all uterine sarcomas.² The disease is characterized by a propensity for early hematogenous spread, leading to high local and distant failure rates.^3-5 In contrast to the much more common uterine adenocarcinoma, lymphatic spread is a rare event, and the overall prognosis is poor.⁶ Surgical resection, including hysterectomy and bilateral salpingo-oophorectomy, is the recommended initial treatment strategy. Although most frequently diagnosed while still confined to the uterus, the clinical course of LMS is difficult to predict. To date, no adjuvant treatment strategies have demonstrated a survival benefit. Current staging systems fail to identify which patients are at highest risk for death; thus, it is difficult to select patients in whom to test potentially beneficial adjuvant strategies.^7-10

Until recently, the rarity of this disease hampered the development of a uterine LMS-specific staging system. Physicians used a modification of the 1988 International Federation of Gynecology and Obstetrics (FIGO) staging system,¹¹ an organ-based staging system that was developed primarily for the staging of uterine endometrial adenocarcinomas. Alternatively, the American Joint Committee on Cancer (AJCC) uses a separate staging system specifically for soft tissue sarcomas. AJCC is a compartment-based, mixed clinical-pathologic staging system that, among other variables, includes information on tumor size and grade. However, the AJCC system is used primarily for the staging of soft tissue sarcomas of the extremity or trunk and has limited value for the staging of visceral tumors.

The predictive accuracy of both the FIGO and AJCC staging systems for patients with uterine LMS has been assessed and compared.^5,12 The strength of the FIGO staging system was its ability to identify patients with the poorest prognosis, whereas the strength of AJCC was its ability to identify patients with a better prognosis. However, significant prognostic overlap between stages was observed in both staging systems. Furthermore, because uterine LMS is diagnosed most frequently while it is still confined to the uterus and is predominantly high-grade in nature, the traditional staging systems tend to group most patients within certain stages (ie, stage I in FIGO and stage III in AJCC), falsely suggesting prognostic homogeneity in a highly heterogeneous group of patients. These findings implicate an urgent need for a more robust prognostic model for patients with uterine LMS.

Recognizing the shortcomings of the traditionally used staging systems, FIGO recently developed a new classification specifically for uterine LMS to include variables such as tumor size, extrauterine spread, and invasion of abdominal tissues.¹³ Apart from disease stage, other reported potentially important prognostic factors in patients with uterine LMS include age, mitotic index (MI), and lymphovascular invasion (LVI), none of which are incorporated into the currently used staging systems.^14-17 The purpose of the current study was to combine both the AJCC and FIGO stage-defining variables with other prognostic factors and develop a novel uterine LMS-specific nomogram to predict postresection 5-year overall survival (OS).

MATERIALS AND METHODS

Patients and Nomogram Predictor Variables

After obtaining institutional review board approval, we used the prospectively maintained Department of Surgery sarcoma database at the Memorial Sloan-Kettering Cancer Center (MSKCC) to identify all patients with uterine LMS who were treated at our institution from July 1982 to June 2008. In total, 270 patients met criteria for the diagnosis of uterine LMS as assessed on routine histopathologic evaluation at MSKCC. Data extracted for the analysis included age at diagnosis, date of diagnosis, surgical procedure, intraoperative tumor rupture, tumor size (continuous variable), histologic grade, involvement of the uterine cervix, locoregional metastases (defined as direct extrauterine spread or the presence of locoregional metastases), lymph node metastases, distant metastases, MI, LVI, date and site of first recurrence or progression of disease, and date of death or last follow-up. To meet inclusion criteria for the analysis, patients must have undergone surgical resection of the primary tumor (total abdominal hysterectomy with or without bilateral salpingo-oophorectomy with or without lymph node sampling). Patients with locally advanced or metastatic disease who underwent more extensive surgical procedures were included in the analysis if the primary tumor (hysterectomy specimen) was resected⁵; however, patients who presented with metastatic disease and never underwent surgery were not included. Paraffin blocks from patients with missing or insufficient information on MI (n = 82) and LVI (n = 72) were re-evaluated by 2 pathologists (E.V. and R.A.S.). Patients who presented at MSKCC with second or later recurrences or patients who had missing or insufficient information on any of the nomogram variables were excluded. LMS tumor size was defined as the maximum dimension of the tumor at pathologic analysis. LMS histologic details included: degree of cellularity, grade (high grade or not high grade), presence of cellular atypia, degree of differentiation, number of mitotic figures per 10 high-power fields, presence of necrosis, and degree of vascularity.^18-20

In the Department of Surgery sarcoma database at MSKCC, a 2-tiered grading system based on diagnostic criteria for soft tissue sarcomas of the trunk, retroperitoneum, and extremities was used. However, the diagnosis of “low-grade” uterine LMS is rare, and some departments of pathology do not assign a grade to uterine LMS but consider every uterine LMS “high-grade.” Others use 2-tiered, 3-tiered, or 4-tiered grading systems. The application of uniform pathologic criteria for uterine LMS (Table 1) results in a reclassification of the majority of “low-grade” LMS as “high-grade” LMS, atypical leiomyoma, or smooth muscle tumor of uncertain malignant potential.^19,21 Most tumors classified as low-grade LMS have been associated with a favorable prognosis and an indolent disease course, even in the setting of metastatic disease. Recognizing the challenges of assigning grade in uterine LMS, for the preparation of this report, we performed a pathologic review of all patients (N = 16) who had tumors originally classified as “low-grade” LMS in the Department of Surgery sarcoma database using Stan-ford criteria (Table 2). The purpose of this review was not to change the initial “low-grade” diagnosis but, instead, to confirm that tumors originally classified as “low-grade” uterine LMS represent a heterogeneous group of smooth muscle tumors with a favorable prognosis. By including patients diagnosed with “not high-grade” disease, we provide a prognostic tool for patients who have smooth muscle tumors that frequently are classified as “low-grade” LMS. This ultimately will enhance the nomogram’s applicability and generalizability.

Table 1.

Histologic Criteria for Uterine Leiomyosarcoma^a

Coagulative Tumor Cell Necrosis	MI (Mitoses/10HPF)	Nuclear Atypia
Focal/extensive	>10	None
Focal/extensive	Any	Diffuse moderate or severe
No necrosis/hyalin necrosis	>10	Diffuse moderate or severe

Abbreviations: HPF, high-power fields.

^aFor the purposes of the pathologic review of low-grade tumors, tumors that met these criteria were considered high-grade. Tumors that were classified as uterine leiomyosarcoma but failed to meet these criteria on pathologic review were considered “not high-grade.”

Table 2.

Pathologic Review of 16 Patients Originally Classified With Low-Grade Uterine Leiomyosarcoma

Patient No.	Soft Tissue Sarcoma Database Diagnosis	MI	Atypia	Necrosis	Pathologic Review Diagnosisa
1	Low-grade leiomyosarcoma	1	Present	Absent	Myxoid leiomyosarcoma
2	Low-grade epitheloid leiomyosarcoma	4	Present (diffuse)	Absent	Mesenchymal epitheloid neoplasm
3	Low-grade leiomyosarcoma	1	Present (minimal)	Absent	Endometrial stromal sarcoma with smooth muscle differentiation
4	Low-grade leiomyosarcoma	1	Present (diffuse)	Absent	Smooth muscle tumor of uncertain malignant potential
5	Low-grade leiomyosarcoma/atypical myxoid smooth muscle tumor	1	Present (focal)	Absent	Endometrial stromal sarcoma with smooth muscle differentiation
6	Low-grade leiomyosarcoma with epitheloid and myxoid features	8	Absent	Absent	Smooth muscle tumor of uncertain malignant potential
7	Low-grade leiomyosarcoma	1	Present (moderate/diffuse)	Absent	Smooth muscle tumor of uncertain malignant potential
8	Low-grade leiomyosarcoma	2	Present (mild/diffuse)	Absent	Smooth muscle tumor of uncertain malignant potential
9	Low-grade leiomyosarcoma	>20	Present (marked/diffuse)	Absent	High-grade leiomyosarcoma
10	Low-grade leiomyosarcoma	>10	Present (marked/diffuse)	Absent	High-grade leiomyosarcoma
11	Low-grade leiomyosarcoma	>10	Present (marked/diffuse)	Absent	High-grade leiomyosarcoma
12	Low-grade leiomyosarcoma	1	Present (moderate/diffuse)	Absent	Endometrial stromal sarcoma with smooth muscle differentiation
13	Low-grade leiomyosarcoma	2	Present (mild/diffuse)	Absent	Smooth muscle tumor of uncertain malignant potential
14	Low-grade leiomyosarcoma/atypical cellular leiomyoma	5	Present (mild/diffuse)	Absent	Smooth muscle tumor of uncertain malignant potential
15	Low-grade leiomyosarcoma	8	Present (mild/diffuse)	Absent	Smooth muscle tumor of uncertain malignant potential
16	Low-grade leiomyosarcoma	1	Present (marked/diffuse)	Present	High-grade leiomyosarcoma

Abbreviations: HPF, high-power fields.

^aPathologic review using Stanford leiomyosarcoma criteria (see Bell 1994¹⁶). For the purposes of the current study, it was assumed that all patients who met Stanford diagnostic criteria for leiomyosarcoma had high-grade uterine leiomyosarcomas.

After resection, some patients received additional treatment in the form of chemotherapy and/or radiation therapy at the discretion of the multidisciplinary disease management team or as part of clinical trial participation.^22-26 The objective of our nomogram was to provide prognostic information at the time of diagnosis to enable more informed clinical decision making for selecting treatment options after surgery. Hence, postoperative treatment variables were not included in the building of the nomogram.

Statistical Considerations

Five-year OS was chosen as the predicted endpoint. OS was defined as the time interval from the date of initial diagnosis to the date of death or last follow-up for patients who were alive and censored. Multivariable analysis was conducted using the Cox proportional hazards model as described previously.^27,28 After modeling continuous variables with transformations, such as restricted cubic splines, linear modeling was considered appropriate for further analysis.²⁷ Categorical variables were grouped based on clinical judgment. Decisions regarding groupings were made before modeling. Colinearity was assessed, and the association between variables was calculated using the Wilcoxon rank-sum test for continuous variables and the chi-square test for categorical variables.

After preliminary screening of data, based on clinical evidence and as previously suggested by the literature, 8 prognostic variables were incorporated into the model (Table 3).^5,15,29,30 Five stage-defining variables (FIGO and AJCC) were considered for the nomogram: tumor size, histologic grade, involvement of the uterine cervix, locoregional metastases/direct extrauterine spread (including regional lymph node metastases), and distant metastases. In addition, age, LVI, and MI were included in the model. MI was modeled using a log transformation.

Table 3.

Univariate and Multivariable Cox Regression Analysis

Variable	No. of Patients	Unadjusted Analysis		Adjusted Analysis
		HR (95% CI)	P	HR (95% CI)	P
Age, y	270	1.02 (1.01-1.04)	.005	1.00 (0.98-1.02)	.79
Tumor size, cm	256	1.06 (1.03-1.09)	<.0001	1.02 (0.98-1.07)	.36
Tumor grade	270
Not high		1.00		1.00	.04
High		6.79 (3.13-14.75)	<.0001	2.70 (1.04-7.02)
Cervical involvement	255
No		1.00		1.00	.66
Yes		1.41 (0.86-2.31)	.18	1.13 (0.65-1.96)
Locoregional metastasesa	257
No		1.00	<.0001	1.00	.0001
Yes		2.43 (1.77-3.34)		2.38 (1.52-3.73)
Distant metastases	269
No		1.00		1.00	.14
Yes		3.45 (2.45-4.84)	<.0001	1.47 (0.89-2.42)
LVI	198
No		1.00		1.00	.92
Yes		1.64 (1.15-2.36)	.007	1.02 (0.66-1.59)
Mitotic indexb	188	1.36 (1.13-1.64)	.001	1.31 (1.05-1.63)	.02

Abbreviations: CI, confidence interval; HR, hazard ratio; LVI, lymphovascular invasion.

^aThe variables positive lymph nodes, direct extrauterine spread, and locally advanced disease were combined as 2 variables (locoregional metastases).

^bLog transformed. Mitoses/10HPF.

Subsequently, 4 different multivariate models that included the pathologic factors (LVI and MI) 1 at a time, both, or none were examined (data not shown). LVI was excluded from the final model, because LVI was associated with MI, grade, distant metastases, and tumor size. The final regression model (7 variables) was chosen based on the clinical relevance of all evaluated prognostic factors (Table 4). Different nomograms were modeled based on the variables that were included in the final regression model (eg, only AJCC and FIGO stage-defining variables; only statistically significant variables with or without age, cervical involvement, and tumor size). The final nomogram model demonstrated the best fit and included all 7 variables. Univariate Cox regression analysis was used to assess the predictive ability of each of the staging systems.

Table 4.

Multivariable Cox Regression Analysis (N = 185)

Variable	Adjusted HR	95% CI	P
Age, y	1.00	0.98-1.02	.86
Tumor size, cm	1.01	0.97-1.06	.49
Tumor grade
Not high	1.00		.05
High	2.57	0.99-6.67
Cervical involvement
No	1.00		.45
Yes	1.23	0.72-2.13
Locoregional metastasesa
No	1.00		.0002
Yes	2.25	1.47-3.45
Distant metastases
No	1.00		.02
Yes	1.74	1.08-2.79
Mitotic indexb	1.30	1.06-1.60	.01

Abbreviations: CI, confidence interval; HR, hazard ratio.

^aLocal metastases/direct extrauterine spread (including regional lymph node metastases.

^bLog transformed.

The discriminative ability of the nomogram and the AJCC and FIGO staging systems was measured with concordance estimation.³¹ The concordance probability (CP) is the chance that, of 2 randomly selected patients, the patient who survives longer has a longer predicted survival probability. The CP can range from perfect concordance (1.0) to perfect discordance (0.0). A CP value of 0.5 indicates that, of 2 randomly selected patients, there is a 50% chance that the patient with the higher predicted survival will survive longer (in other words, the model’s predictive capability is no better than a coin flip).

Model Validation

The nomogram was internally validated using 200 bootstrap samples to prevent over fitting. We used 200 bootstrap samples, and each sample was equal to the original sample (185), because we sampled with replacement. Specifically, a model was built on a bootstrap sample (training set) and then evaluated on the original data (test set) without modification. Two concordance indices were calculated based on the training and test data sets. The difference between the 2 indices describes the optimism of the fit. The process was repeated 200 times. The final optimism estimate was calculated as the average of the 200 differences. The difference between the original CP based on all data and the optimism estimate is the unbiased measure of the CP, which describes the ability to discriminate among individual patients if the nomogram was to be used in a new cohort.

Finally, the nomogram was calibrated by plotting the nomogram’s predicted 5-year survival against the actually observed survival as calculated by the Kaplan-Meier method. Patients were divided into 5 groups using the quartiles of predicted risk as cutoff points. The calibration plot describes how far predictions are from the actual outcomes. Again, 200 bootstrap samples were used to reduce the overfit bias, which would overstate the accuracy of the nomogram. All tests were performed with SAS software (version 9.2; SAS Institute Inc., Cary, NC) and the Design, Hmisc, and CPE libraries in R 2.8.1 (R Project for Statistical Computing, Vienna, Austria).

RESULTS

During the study period, 270 patients presented with pathologically confirmed uterine LMS. Patients with 1 or more missing nomogram values (n = 85) were excluded, leaving 185 patients for the nomogram analysis. There were no significant differences between the whole cohort (n = 270) and the nomogram cohort (n = 185) with regard to follow-up, median OS, and clinical characteristics. There was no statistically significant difference in OS between patients who received no further treatment versus those who received any type of postresection treatment. In addition, there was no statistically significant difference in stage-specific OS (by either FIGO or AJCC) between patients who received no further treatment versus those who received any type of postresection treatment (data not shown).

Patient demographics and clinical characteristics of the nomogram cohort are provided in Table 5. The median OS (Fig. 1) for all patients was 3.75 years (95% confidence interval [CI], 3-6 years). The median follow-up for the surviving patients was 5.4 years (range, 0.3-22.8 years). The 5-year OS rate was 44.1% (95% CI, 38.9%-51.5%). Of the 185 patients, 136 (73%) progressed, and 111 (60%) died. Of the 111 patients who died, 106 died of disease. One patient died of an unrelated cause, and 4 died of unknown causes. The observed 5-year OS by 1988 FIGO stage is provided in Table 6. Fifty-five percent of patients with FIGO stage I uterine LMS were alive at 5 years.

Table 5.

Demographics and Clinical Characteristics of Patients

Characteristic	No. of Patients (N = 185)	%
Age, y
Median	51
Range	23-81
High tumor grade
Yes	169	91
No	16	9
Tumor size, cm
Median	9
Range	0.9-28.0
Cervical involvement
Yes	21	11
No	164	89
Locoregional metastasesa
Yes	52	28
No	133	72
Distant metastases
Yes	34	18
No	151	82
Mitotic index, mitoses/HPF
Median	15
Range	1-200
Postresection treatment
Any treatment	64	35
Chemotherapy	43	23
Chemoradiation therapy	7	4
Radiation therapy	14	8
No treatment	73	39
Not specified	48	26
FIGO stageb
I	107	58
II	14	8
III	31	17
IV	33	18
Follow-up
No. of survivors	74
Survival, y
Median	5.40
Range	0.33-22.83
Patient status
Alive without disease	44	24
Alive with disease	30	16
Dead of disease	106	57
Dead, not disease related	1	1
Dead of unknown cause	4	2

Abbreviations: CI, confidence interval; FIGO, International Federation of Gynecology and Obstetrics; HPF, high-power fields.

^aThe category locoregional metastases indicates local metastases/direct extrauterine spread (including regional lymph node metastases).

^bAccording to the 1988 FIGO staging system (see Berchuck 1988¹¹).

Figure 1.

This chart illustrates the median overall survival of patients with uterine leiomyosarcoma who were included in the nomogram analysis (n = 185). CI indicates confidence interval.

Table 6.

Five-Year Overall Survival by 1988 International Federation of Gynecology and Obstetrics Stage (n = 185)

FIGO Stage	No. of Patients		5-Year OS, %
	Total	Failed	Estimate	95% CI
I	107	55	55.2	44.6-64.5
II	14	10	28.6	8.8-52.4
III	31	22	36.2	19.5-53.2
IV	33	24	21.7	8.2-39.4

Abbreviations: CI, confidence interval; FIGO, International Federation of Gynecology and Obstetrics; OS, overall survival.

A nomogram for predicting 5-year OS that included all covariates was developed (Fig. 2). The CP of the nomogram was 0.67 (95% CI, 0.63-0.72). This was superior to survival predictions based on the original 1988 FIGO staging system (CP, 0.60; 95% CI, 0.56-0.65), the recently published 2009 FIGO staging system (CP, 0.58; 95% CI, 0.54-0.62), and the sixth edition of the AJCC staging system for soft tissue sarcomas (CP, 0.59; 95% CI, 0.55-0.64). The bootstrap-validated CP was 0.65. The absolute error of the nomogram’s prediction was assessed with a calibration plot (Fig. 3). The bootstrap-corrected nomogram estimates approximate the ideal nomogram performance (as indicated by the 45-degree reference line in Fig. 3) and illustrate good calibration, with predictions within 5% from actual outcomes.

Figure 2.

This is the uterine leiomyosarcoma nomogram for 5-year overall survival. The mitotic index (asterisk) was modeled using log transformation; for display purposes, values were converted back to original scale (exponential; concordance probability [CP], 0.671; bootstrap-validated CP, 0.651).

Figure 3.

This plot illustrates calibration curves for the nomogram. The x-axis indicates the nomogram-predicted probability. Patients were grouped by quartiles of predicted risk. The y-axis indicates the actual 5-year probability of survival as estimated by the Kaplan-Meier method. The solid line represents the actual nomogram; dashed line, ideal agreement between the actual and predicted probabilities of 5-year survival. Vertical bars represent 95% confidence intervals, dots correspond to apparent predictive accuracy, and crosses mark the bootstrap-corrected estimates. The measures underlying the model calibration are as follows: maximum error, 0.067; average error, 0.029; intercept, 0.038; and slope, 0.915.

For an example of the clinical utility of the nomogram, a woman aged 45 years (3 points) with a 10-cm (10 points), high-grade LMS (65 points), without cervical involvement (0 points), with no locoregional spread (0 points), with no distant metastases (0 points), and with an MI of 3 (18 points) would have an estimated 67% chance of being alive at 5 years after diagnosis. Conversely, a woman aged 70 years (6 points) with a 20-cm (20 points), high-grade LMS (65 points), without cervical involvement (0 points), with no locoregional spread (0 points), with no distant metastases (0 points), and with an MI of 55 (73 points) would have only a 34% chance of being alive at 5 years after diagnosis. Both patients would have been classified with FIGO stage I disease according to categorical staging criteria, with an estimated 5-year OS probability of 55% for both patients.

DISCUSSION

There is increasing evidence that alternative staging platforms, such as nomograms, have advantages over generic staging systems in predicting survival outcomes for various solid tumors, primarily because of their ability to reduce statistical predictive models into continuous numerical estimates tailored to the profile of individual patients.^32-35 User-friendly graphic interfaces for generating these estimates facilitate the use of nomograms during clinical encounters to inform clinical decision making.^27,33

We have developed and internally validated a novel, uterine LMS-specific nomogram for predicting 5-year OS. Rather than assigning patients to 4 stages based on the “all-or-none” principle of categorical stage-defining variables of the traditionally used FIGO and AJCC staging systems, the nomogram considers multiple commonly available prognostic variables simultaneously, including continuous values, thus allowing for the ability to take into account more complex relations between prognostic factors for uterine LMS and outcome.

This novel, uterine LMS-specific nomogram outperforms the traditionally used staging systems, because it improves prognostic accuracy and permits a superior individualized prediction of overall survival. Apart from the improved prognostic accuracy of correctly ranking the risk of death for individual patients, the nomogram demonstrates good calibration accuracy—the nomogram’s survival predictions are within 5% of the actual survival probabilities.

In 2002, a soft tissue sarcoma nomogram (for all histology subtypes) was developed.³⁶ Although the soft tissue sarcoma nomogram predicts 12-year sarcoma-specific death with excellent accuracy and calibration, it was not designed to predict 5-year OS specifically for patients with uterine LMS. The uterine LMS-specific nomogram reported here permits the additional prognostic information of tumor size as a continuous (rather than categorical) variable, extrauterine spread, and MI. It predicts 5-year OS rather than 12-year OS, an endpoint of potential utility for the design of trials testing adjuvant treatment strategies.

To date, no adjuvant postoperative treatment strategies have been able to demonstrate a survival benefit for patients with uterine LMS. One possibility for the failure to demonstrate improvement in outcomes is that the group of patients in which adjuvant strategies are tested is too heterogeneous with respect to survival probability. With small sample sizes, minor imbalances in adverse risk prognostic factors may mask potentially meaningful differences in outcomes. The nomogram has the potential to separate the population with respect to outcomes for stratification such that adjuvant strategies can be tested among more homogenous prognostic LMS populations. This is particularly important in uterine LMS, in which a benefit from adjuvant therapy for high-grade disease is difficult to prove.²⁴ Currently, such approaches are being used successfully in other solid tumors, such as in the Cancer and Leukemia Group B 90203 trial.³⁷

There are several limitations to this study. This study was performed using retrospective data, and treatment was not assigned in a randomized fashion. Some patients who were included in this cohort received postoperative treatment as part of either standard or experimental therapy.^22-26 However, our analyses confirm that OS did not differ significantly between patients who did or did not receive postresection treatment. Postresection treatment details were not included as variables in the nomogram development, because the goal of this nomogram is to provide prognostic information for patients and physicians at the time of diagnosis.

The nomogram is limited to patients who underwent at least a hysterectomy to assess for tumor size, extra-uterine spread, or MI and, thus, does not account for patients who presented with unresectable metastatic disease or those who were medically not fit to undergo a surgical procedure. However, the nomogram includes patients with locally advanced or resectable metastatic disease who underwent more extensive surgical procedures in addition to a hysterectomy. Because patients with meta-static disease who did not undergo surgery are not considered in this nomogram, the prognostic impact of distant metastases appears less pronounced in the nomogram. Excluding patients with distant metastases and developing a new nomogram for patients with earlier stage disease did not result in an improved prognostic performance (data not shown). By including patients with metastatic disease, we provide a more comprehensive prognostic tool for a patient cohort with resected uterine LMS that probably is observed in clinical practice.

Because of the rarity of the disease, the number of patients included in this study is relatively small compared with the numbers in other studies that evaluated novel prognostic models for solid tumors. However, internal validation confirmed the robustness of the model (the CP decreased from 0.67 to 0.65 after bootstrapping). More patients and longer follow-up most likely will improve nomogram accuracy. To put the CP of the uterine LMS nomogram into perspective, the majority of CPs for other solid tumor prediction models range between 0.58 and 0.80 (eg, breast, pancreas, colorectal, gastric, prostate).^38-42

Although this uterine LMS nomogram demonstrates better prognostic accuracy than traditionally used categorical staging systems, there is room for improved prediction performance. On the basis of the final multivariable regression model, age, tumor size, and cervical involvement remained nonsignificantly associated with survival in our analysis. We decided to leave the variables in the model, because the nonsignificant associations may have been a result of the relatively small sample size in our series, because the variables were associated significantly with impaired disease-specific survival in other series, and because the addition of the variables improved the predictive accuracy of the nomogram slightly. Thus, as new LMS-specific prognostic markers and molecular profiles are identified, the nomogram can be modified if it is demonstrated that such variables further increase prognostic accuracy.⁴³

In conclusion, we have developed a new, individualized prognostic model that predicts 5-year OS for patients with uterine LMS. This nomogram is expected to be posted on the MSKCC web site at www.mskcc.org/mskcc/html/5794.cfm. The proposed individualized risk-prediction model has the potential to improve the management of women who are diagnosed with uterine LMS by allowing physicians to more precisely identify patients who are at a low risk of death and those who may be candidates for clinical trials testing the efficacy of adjuvant treatment strategies.