Chemotherapy response is an important predictor of local recurrence in ewing sarcoma

Abstract

BACKGROUND.

Local recurrence in Ewing sarcoma is associated with a poor prognosis. The purpose of the study was to determine the factors that predict local recurrence after surgical treatment of the primary tumor.

METHODS.

Between 1990 and 2001, 64 patients underwent surgical resection of Ewing sarcoma. Surgical margins were assessed histologically and radiologically. Response to preoperative chemotherapy was determined by detailed specimen mapping. Local recurrence-free survival (LRFS) was calculated by Kaplan–Meier analysis. Multivariate analysis was performed with the Cox proportional hazards model.

RESULTS.

A number of factors were found to be associated with local recurrence on univariate analysis. Patients with a good response to chemotherapy (≥90% tumor necrosis), had superior LRFS at 5 years (86% vs 51%, P = .015). Central site of disease was associated with an increased rate of recurrence. The LRFS at 5 years was 50% for the chest wall, 74% for pelvic/scapular, and 86% for extremity tumors (P = .083). Positive surgical margin was not a strong predictor of recurrence (P = .72). A critical analysis of minimal surgical margin based on preoperative magnetic resonance imaging (MRI) and computed tomography (CT) scans also failed to reveal an association between margin and local recurrence. In multivariate analysis, the 2 independent predictors of local recurrence were histological response to chemotherapy and central site of disease.

CONCLUSION.

Local recurrence after surgical resection is a complex phenomenon. An important predictive factor is the response to chemotherapy. In the current study, this seems to have the largest impact. Central site of disease may be a second independent predictive factor. Cancer 2007;109:603–611. © 2006 American Cancer Society.

The local treatment of the primary tumor in Ewing sarcoma has been the subject of debate for some time. Both radiation and surgery have shown efficacy in treatment.1–3 Each modality offers distinct advantages and disadvantages. In recent years there has been greater acceptance and use of surgery for local treatment. There is a need to critically analyze the outcome of surgical treatment and determine which patients are most likely to benefit from surgery.

In the past, local recurrence of Ewing sarcoma has been associated with poor outcome.4–7 Local recurrence is usually followed by death, and thus, local control of the disease remains a critical goal of treatment. Factors that could potentially influence the development of local recurrence include the site of disease, the stage, the size of the primary tumor, the response to preoperative chemotherapy, and the surgical margin. The goal of this study was to identify the factors that determine the success of surgical treatment with regard to local control of disease. A detailed study of these factors may help to better define the indications and contraindications for surgery.

MATERIALS AND METHODS

Patients

Sixty-four patients with Ewing sarcoma underwent surgical resection of the primary tumor after preoperative chemotherapy. The study period was from 1990 to 2001. Patients were followed for a minimum of 24 months unless they succumbed to their disease before 24 months (15 patients). The median follow-up was 57 months (range, 7–167 months). The patients' medical records and imaging studies were reviewed retrospectively. Permission to review the records was obtained from the Institutional Review Board.

Patient demographics are shown in Table 1. There were 22 women and 42 men. The median age was 17 years (range, 1–67). There were 39 tumors of the extremities and 25 tumors in central locations. There were 48 patients who presented with localized, nonmetastatic disease and 16 patients who presented with metastatic disease. The presence of metastasis was verified by review of chest computed tomography (CT), bone scans, bone marrow aspirates, and radiographs. Patients with metastasis were included to evaluate whether the advanced stage of disease at presentation may affect the likelihood of local recurrence.

Table 1. Patient Demographics

Total patients		64
Follow-up	Range	7–143 months
	Median	57
Age	Range	1–67 years
	Median	17
Sex	Men	42
	Women	22
Race	Caucasian	40
	Hispanic	16
	African-American	3
	Other	5
Stage	Local	48
	Metastatic	16
Site	Central	25
	Scapula	5
	Clavicle	1
	Chest wall	8
	Pelvis	8
	Sacrum	3
	Extremity	39
	Humerus	6
	Radius/ulna	1
	Femur	10
	Tibia/fibula	9
	Calcaneus	3
Surgical treatment	Chest wall resection	8
	Limb salvage	47
	Amputation	9

Pathology

All surgically resected specimens were processed in standard fashion with analysis of inked surfaces for margins. Tumor necrosis was determined by careful mapping of the specimen, which was divided into a grid and separately analyzed for each cassette in the grid.8 The specimens were bisected and a coronal slice from the central portion of the tumor was divided entirely into smaller blocks that could be processed in standard histology cassettes. Specimen radiographs were obtained in order to document the positions of the smaller sections in relation to the whole cross-section. After the sections were processed, slides from each section containing tumor were assessed as to the extent of necrosis present. The percentage of necrotic cells in each of these sections was then averaged for an estimate of the tumor's response to chemotherapy.

Radiology

CT and magnetic resonance imaging (MRI) were used to determine the anatomic features of the primary tumors. The imaging studies were obtained before the start of chemotherapy and just before surgery. CT was generally used for chest wall disease and MRI for other sites. The minimal surgical margin at certain areas was defined radiologically by the distance from the tumor to a critical anatomic structure that was preserved during the surgery, such as a major nerve, blood vessel, or internal organ.

Chemotherapy

During the earlier part of the study (1990–1997), chemotherapy utilized the combination of vincristine, doxorubicin, and cyclophosphamide (VAC). Variations of this regimen involved the addition of dactinomycin or ifosfamide with etoposide.

During the latter part of the study (1997–2001), 2 protocols were predominantly used. The high-dose VAC (HD-VAC) protocol included vincristine (2.0 mg/m²) with a maximum dose of 2.0 mg, doxorubicin (90 mg/m²), and cyclophosphamide (4.0 gm/m²) followed by mesna. Patients received up to 6 cycles before surgery (mean, 5.3; range, 4–6 cycles). After surgery, patients were randomized to receive ImmTher, a liposome incorporated disaccharide tripeptide, or no chemotherapy.9, 10 The second regimen (VAI) involved vincristine (2.0 mg), doxorubicin (75 mg/m²), and ifosfamide (10.0 gm/m²). Patients received up to 6 cycles before surgery (mean, 5.3; range, 3–6 cycles). After surgery, postoperative chemotherapy was given at the discretion of the medical oncologist. This typically consisted of either continuation of the same regimen (VAI) or high-dose ifosfamide (12–14 gm/m²). The mean number of postoperative cycles was 4 (range, 2–10).

Surgical Treatment

Fifty-five patients underwent en bloc resection of the primary tumor with preservation of the affected limb. Nine patients underwent amputation as primary local treatment. Amputations were excluded from Kaplan–Meier analysis of local recurrence-free survival (LRFS); however, they were included in the analysis of overall survival and continuous disease-free survival (CDFS). Below-knee amputations were performed in 5 patients for disease in the calcaneus (3), distal tibia (1), and distal fibula (1). Above-knee amputations were performed in patients with disease in the tibia (2) or fibula (1). One shoulder disarticulation was performed for humeral disease.

Statistics

LRFS, CDFS, and overall survival (OS) were calculated by Kaplan–Meier analysis using the SPSS 12.0 (Chicago, IL) statistical program. The log-rank test was used to compare survival curves for univariate analysis. Patients who died of causes unrelated to the disease or its treatment were censored. Associations between categorical variables without time dependence were assessed by the chi-square test or Fisher exact test. Statistical significance was defined as P ≤ .05. Marginal statistical significance was defined as .05 < P ≤ .10.

The Cox proportional hazards model was used to perform multivariate analysis. Factors that were entered into the analysis included age, type of chemotherapy, site of disease, stage of disease, histological response to chemotherapy, surgical margin (both histologically and radiologically defined), race, gender, and size of original tumor. Covariates were selected by the chi-square statistic. A parsimonious model was determined by backward conditional analysis. Only those covariates that contributed significantly to the final model were selected.

RESULTS

Effect of Local Recurrence on Survival

By Kaplan–Meier analysis, the OS of patients at 5 years was 72% for nonmetastatic disease and 25% for metastatic disease (P = .0002; Fig. 1A). The CDFS at 5 years was 58% for nonmetastatic and 20% for metastatic disease (P < .0001).

Figure 1. Kaplan–Meier analysis of overall survival is shown. (A) Stage of disease affected survival. The presence of metastatic disease at the time of diagnosis was associated with worse overall survival (P = .0002). (B) Development of local recurrence was associated with worse overall survival (P = .007). In this analysis patients who presented with metastatic disease were excluded.

Local recurrence was associated with significantly worse survival (Fig. 1B). For patients who presented with localized, nonmetastatic disease, the OS at 5 years was 78% for patients who did not develop local recurrence and 29% for patients who developed local recurrence (P = .007).

Univariate Analysis for Local Recurrence

The effect of individual factors on LRFS was evaluated by Kaplan–Meier survivorship and the log-rank test (Table 2). For certain factors, different cut-off parameters were used to group patients. The results of the log-rank test are shown in Table 2 for the different groupings.

Table 2. Univariate Analysis of Selected Factors

No.	Factor	Group	5-y LRFS, %	P
LRFS indicates local recurrence-free survival; HD-VAC, high-dose vincristine, doxorubicin, and cyclophosphamide; VAI, vincristine, doxorubicin, and ifosfamide. * Largest dimension at time of presentation. † Excludes 7 tumors with less than 1 cm extraosseous soft tissue tumor at the time of presentation.
1a	Chemotherapy response	90%–100%	86	.015
		<90%	51
1b	Chemotherapy response	99%–100%	91	.026
		90%–98%	75
		<90%	51
2	Stage	Nonmetastatic	83	.022
		Metastatic	46
3a	Site	Extremity	86	.083
		Pelvis/scapula	74
		Chest wall	50
3b	Site	Extremity	86	.11
		Central	66
4a	Age	0–20 y	83	.12
		>20 y	66
4b	Age	0–20 y	83	.27
		21–30 y	64
		>30 y	67
5	Chemotherapy regimen	HD-VAC or VAI	90	.21
		Other	69
6a	Surgical margin (2 and 10 mm cut-off)	0–2 mm	78	.23
		3–9 mm	83
		≥10 mm	93
6b	Surgical margin (2 mm cut-off)	0–2 mm	78	.49
		>2 mm	88
6c	Surgical margin (inked margin)	Negative	76	.72
		Positive	75
7	Race	Other	68	.38
		Caucasian	80
8	Gender	Male	75	.45
		Female	81
9a	Size (5 and 10 cm cut-off)*	<5 cm	67	.57
		5–10 cm	69
		>10 cm	82
9b	Size (8 cm cut-off)*	<8 cm	76	.96
		≥8 cm	78
10	Volume (at presentation)	<150 cc	74	.81
		≥150 cc	77
11	Volume reduction after chemotherapy†	0%–50%	71	.39
		>50%	81

The degree of tumor necrosis, as quantitatively measured by detailed grid mapping of the resected specimen, demonstrated a significant correlation with local recurrence. A 90% threshold was employed, and this corresponded to traditional designations of “good” and “poor” chemotherapy response. The 5-year LRFS was 88% for good responders versus 50% for poor responders (P = .015). When patients with ≥90% necrosis were further stratified by excellent response (99%-100% necrosis) versus good response (90–98% necrosis), LRFS was found to be 91% versus 82% at 5 years, respectively (P = .026; Fig. 2A).

Figure 2. The effect of various factors on local recurrence-free survival (LRFS) is shown. (A) The response to chemotherapy had a significant effect on LRFS. Patients who had <90% tumor necrosis had an LRFS of 50% at 5 years. Patients who had 90% to 98% necrosis had LRFS of 75% at 5 years. Patients who had 99% to 100% necrosis had LRFS of 92% at 5 years (P = .026). (B) The effect of site of primary tumor on LRFS is shown (P = .083). Patients with disease in the scapula and pelvis are grouped together. These cases do not include disease arising primarily in the proximal humerus or femur. Clavicular disease is grouped together with the scapula. (C) The effect of inferred minimal surgical margin, as defined radiologically, is shown (see Materials and Methods). A trend toward better LRFS is seen for wider margins, but this did not reach statistical significance (P = .23). The 5-year LRFS was 78% for 0–2 mm margins, 83% for 3–9 mm margins, and 93% for margins of at least 10 mm.

The stage of disease showed a significant effect on LRFS. Patients who presented with metastatic disease had a 5-year LRFS of 46% compared with 83% for nonmetastatic disease (P = .022).

The anatomic site was of marginal statistical significance with respect to LRFS. Disease in the chest wall was associated with worse outcome and LRFS of 50% at 5 years. Patients with central disease in the shoulder or pelvic girdles had intermediate LRFS of 74%. Patients with tumors in the extremities had LRFS of 86% (P = .083; Fig. 2B). There was no correlation between central site of disease and metastatic disease at presentation (P = .54, chi square test).

Factors that were not associated with a significant effect on LRFS included age, gender, race, type of chemotherapy regimen, surgical margin, and initial size of tumor (Table 2). The tumor size was measured in several different ways, including total volume and greatest dimension. None of these methods yielded a significant finding for LRFS.

In response to the correlation between chemotherapy response and LRFS, an attempt was made to determine whether imaging response, as measured by reduction in the size of the tumor mass, could also be correlated with LRFS (Table 2). A 50% reduction in the soft tissue tumor, based up volume estimation from CT or MRI, was used as a cut-off point to group patients because tumors often partially ossify and do not altogether disappear (Fig. 3). Patients with greater than 50% volume reduction had a 5-year LRFS of 81% compared with 71% for patients with less than 50% volume reduction (P = .39).

Figure 3. The difficulty of trying to quantify response to chemotherapy radiographically is exemplified by this case. Following chemotherapy there was reduction of the soft tissue mass, but it had not completely disappeared. Ossification had occurred in the residual mass corresponding to the region of periosteal bone formation. Although ossification appears nearly complete, there are areas of relative lucency that suggest the possibility of residual viable tumor. The final pathological analysis indicated tumor necrosis of 98%.

Surgical Margin

The surgical margin, defined traditionally by tumor at the inked margin on histological sections, was analyzed as a potential predictor of local recurrence. Of the 4 patients who had a positive or contaminated margin, 1 patient developed a local recurrence (P = .72). This patient had a primary tumor in the chest wall. The response to chemotherapy was poor (49% necrosis). All other patients with local recurrence had negative histological margins.

Because few patients had positive surgical margins, a more critical analysis of surgical margin was performed with the aid of imaging studies. The quality of the surgical margin was assessed with preoperative MRI or CT. Although the actual margin could not be predicted completely, the minimal surgical margin at critical, preserved structures was inferred. The distance from the edge of the tumor to major neurological, vascular, and other vital structures was measured. This measurement defined the minimal possible soft tissue margin at such sites. The closest structure was commonly a nerve and/or large blood vessels. If only a plane of fat separated the vessels from the tumor on T1-weighted images, the margin was typically assessed as 1 mm. If no separation of tumor from vessels could be discerned, the margin was defined as 0 mm.

A trend toward better local control with wider as opposed to narrower margins was seen, but this did not reach statistical significance (Fig. 2C). LRFS at 5 years for margins between 0–2 mm was 78% compared with 88% for margins greater than 2 mm (P = .49). When the group of patients with margins greater than 2 mm was further stratified, the patients with 3–9 mm margins were found to have LRFS at 5 years of 83%, whereas patients with ≥10 mm margins were found to have LRFS of 93% (P = .23). A comparison between patients with worst margins (≤2 mm) and best margins (≥10) also did not yield a statistically significant difference (P = .12).

Radiation

Six patients received adjuvant radiation. The range of doses was 40–60 Gy (mean, 52 Gy). The sites of tumor were chest wall (1), sacrum (3), pelvis (1), and proximal femur (1). There was 1 local recurrence. This occurred in the previously mentioned patient with chest wall disease who had a positive surgical margin and poor tumor necrosis (49%) on histology. The patient received 56 Gy external beam radiation.

Multivariate Analysis

Multivariate analysis using a proportional hazards model with Cox regression analysis identified 2 independent factors predictive of local recurrence: response to chemotherapy (P = .007) and site of disease (P = .002). The histological response to chemotherapy was grouped by parameters of 0% to 89%, 90% to 98%, and 99% to 100% tumor necrosis. This grouping was determined in the analysis to be a stronger predictor than the conventional grouping of 0% to 89% and 90% to 100% tumor necrosis.

The site of disease was an independent factor when grouped by disease in the chest wall, pelvic/scapular girdles, and extremities. This grouping was determined to be more predictive than the grouping of extremity versus nonextremity.

Five of 6 patients who had tumor necrosis less than 90% and central site of disease developed local recurrence. The 1 patient without local recurrence died early of metastatic disease 10 months after surgery.

DISCUSSION

Much controversy has centered around the relative merits of surgery and radiation for primary local treatment of Ewing sarcoma.11 In this debate, one must remember that the efficacy of local treatment is not uniform for all patients. The goal of the present study was to identify high-risk patients, who are likely to fail surgical treatment. The results of the study indicate that the histological response to preoperative chemotherapy, as defined by percent tumor necrosis, is a strong predictor of local recurrence. Patients who have a poor response have a 50% risk of local recurrence at 5 years.

It is not surprising that response to chemotherapy is an important factor determining local recurrence. A correlation between chemotherapy response and survival has previously been reported by Picci et al.12 In the more recent study by Grier et al.,13 local recurrence was noted to be significantly less in the group of patients with better survival. These patients received ifosfamide and etoposide in addition to the standard treatment, which consisted of vincristine, doxorubicin, cyclophosphamide, and dactinomycin.

It would be advantageous to predict the percentage of tumor necrosis before surgery, but this is not possible at the present time. Several methods were used in this study to correlate histological response with imaging response, but no reliable method emerged for predicting necrosis. A number of factors complicate the interpretation of tumor response on imaging studies. Most tumors produce intraosseous and extraosseous components that do not shrink equally with chemotherapy. Abnormalities of the bone marrow may persist, whereas soft tissue masses may shrink dramatically. Tumors with little or no extramedullary soft tissue may demonstrate no measurable shrinkage. Therefore, reduction in volume, based on imaging, becomes hard to quantify.

Ossification also confounds one's ability to judge tumor response through imaging studies. Periosteal ossification elicited by the tumor does not resolve, leaving a “mass” regardless of the completeness of tumor necrosis. A large degree of ossification in a mass is considered a favorable sign qualitatively, but is difficult to quantify.

Newer tests such as positron emission tomography (PET) may provide a better reflection of tumor response.14 PET offers a potential surrogate measure of tumor necrosis, but efforts in this area are still investigational.15 Dynamic MRI may be another method to assess chemotherapy response.16

The histological response to chemotherapy alone is not entirely predictive of local recurrence. Two patients had 100% tumor necrosis but developed local recurrence. Both patients had other unfavorable characteristics. One had chest wall disease and the other had a scapular tumor with metastases in the lungs at presentation.

Aside from chemotherapy response, other factors may be important in determining the risk of local recurrence, including central site of disease and stage of disease (metastasis at presentation). The stage of disease was significant in univariate analysis, whereas the site of disease was significant in multivariate analysis. The reason that these factors did not reach statistical significance on both univariate and multivariate analysis may have been due in part to the relatively small numbers of patients in the study. Central location of the primary tumor, such as in the pelvis, shoulder girdle, spine, or chest, has previously been shown to have a negative prognostic effect.7, 17–19 In the present study, disease in the chest wall had a worse prognosis than at other sites, both in terms of local failure and overall survival. This may be due to a number of unfavorable anatomical factors, including invasion of the pleura, tumor extension into the lung parenchyma, pleural effusions in direct contact with the tumor, and regional metastasis to surrounding lung tissue.20–22

Counterintuitively, our study found that surgical margins were not predictive of local control. One reason for this is that only 5% of cases were noted to have a positive margin as traditionally defined by tumor at the inked margin. Cases with negative margins could potentially be associated with very narrow margins. Previous studies have reported an association between margins and local recurrence.23–26 In these studies, a distinction was usually made between marginal excision and wide excision, as defined by Enneking et al.27 In the present study no attempt was made to make this distinction, because there were no objective criteria for designating marginal excision in a retrospective manner.

To address the possibility that some resections may have had close margins, the quality of surgical margin was assessed by MRI. In this analysis, the thickness of the margin at critical sites such as major nerves and blood vessels was measured. Although a certain trend toward better local control with wider margins was detected, it did not reach statistical significance. It is quite possible that with a larger sample size the trend would become statistically significant.

An important consideration with regard to margins is that the soft tissue tumor mass often regresses after preoperative chemotherapy. A previously involved muscle may appear to be normal on MRI scan after chemotherapy but may harbor microscopic disease. If this tissue is not resected, it could potentially be a source of subsequent local recurrence.

The results of this study should not be interpreted as proof that wide margins are unimportant. Quite possibly, what it demonstrates is our inability to judge true margins accurately and quantitatively. Therefore, although it is important that wide negative margins be achieved, this alone is not a strong predictor of local recurrence and may not be helpful in guiding subsequent treatment.24, 25

Adjuvant radiation can be considered for patients with unfavorable characteristics, such as poor chemotherapy response and central location.7, 24, 28 Chest wall lesions associated with pleural effusions may benefit from adjuvant radiation.29 During the study period, there were only 6 patients who received adjuvant radiation and, therefore, no definite conclusions could be drawn with regard to this treatment strategy. Only 1 patient receiving radiation has developed local recurrence thus far. These results, although preliminary, are encouraging. Anecdotally, another patient in our study received adjuvant radiation after resection of a second local recurrence after a poor chemotherapy response (20% necrosis). She is currently free of disease 35 months after treatment.

Further studies will be necessary to determine the merit of adjuvant radiation for high-risk patients. The toxicity and morbidity of combined surgery and radiation is greater than either alone and must be considered. The results of this study suggest that patients with both central site of disease and poor tumor necrosis may need to be considered for adjuvant radiation. Five of 6 patients with both unfavorable characteristics developed local recurrence. The sixth patient died at 10 months, possibly before local recurrence could be detected.

In conclusion, local recurrence after surgical resection is a complex phenomenon. The strongest predictive factor of recurrence is response to chemotherapy. Central site of disease and metastasis at presentation may also be associated with increased recurrence rates. These factors may be more predictive of recurrence than surgical margin as traditionally defined.