Fax: (011) 390223902804
Tumor response assessment by modified Choi criteria in localized high-risk soft tissue sarcoma treated with chemotherapy
Article first published online: 17 MAY 2012
Copyright © 2012 American Cancer Society
Volume 118, Issue 23, pages 5857–5866, 1 December 2012
How to Cite
Stacchiotti, S., Verderio, P., Messina, A., Morosi, C., Collini, P., Llombart-Bosch, A., Martin, J., Comandone, A., Cruz, J., Ferraro, A., Grignani, G., Pizzamiglio, S., Quagliuolo, V., Picci, P., Frustaci, S., Dei Tos, A. P., Casali, P. G. and Gronchi, A. (2012), Tumor response assessment by modified Choi criteria in localized high-risk soft tissue sarcoma treated with chemotherapy. Cancer, 118: 5857–5866. doi: 10.1002/cncr.27624
- Issue published online: 19 NOV 2012
- Article first published online: 17 MAY 2012
- Manuscript Accepted: 26 MAR 2012
- Manuscript Revised: 18 MAR 2012
- Manuscript Received: 9 JAN 2012
- response assessment;
- Response Evaluation Criteria in Solid Tumors;
- Choi criteria;
The objective of this study was to compare the prognostic relevance of Response Evaluation Criteria in Solid Tumors (RECIST) versus Choi criteria for the assessment of response in patients with high-risk soft tissue sarcoma of the extremities or trunk wall who received preoperative chemotherapy with or without radiotherapy in a phase 3 trial.
Patients received 3 cycles of preoperative epirubicin + ifosfamide with or without radiotherapy. The diagnostic concordance between RECIST and Choi criteria and their correlation with overall survival (OS) and freedom from progression (FFP) were evaluated in a univariate Cox regression model.
In 243 of 321 eligible patients, RECIST, Choi criteria, and histology were predictive for OS and FFP. In the subgroup of 69 patients who received chemotherapy alone and were evaluable by both RECIST and Choi criteria, Choi criteria were associated significantly with OS and FFP, whereas RECIST predicted only FFP, and the pattern of agreement observed between the 2 criteria was unsatisfactory. On a dichotomous scale, comparing objective response (complete and partial responses) and lack of response (stable and progressive disease) to preoperative chemotherapy according to RECIST and Choi criteria, only Choi criteria were predictive of OS and FFP, and fair agreement between RECIST and Choi criteria was observed. When lack of progression and progression were compared (complete and partial responses + stable disease vs progressive disease), both assessment criteria were significantly predictive of OS and FFP, and there was substantial agreement between the 2 criteria.
Response to chemotherapy with or without radiotherapy was associated with a better outcome in patients with high-risk soft tissue sarcoma. Choi criteria were better predictors than RECIST in patients who received preoperative chemotherapy alone. Cancer 2012. © 2012 American Cancer Society.
Response Evaluation Criteria in Solid Tumors (RECIST) are based on tumor size.1, 2 This is an added value in terms of their reliability (reproducibility), because size is relatively easy to measure. However, the introduction of molecular-targeted therapies has emphasized the limitations of RECIST in terms of their validity, because tumor response may be nondimensional, especially in an early phase.3-21 In these patients, tumor response is marked by changes in the tumor radiologic aspect and in its functional imaging.
Attempts to establish new response criteria have been undertaken. Choi criteria were conceived to assess tumor response in patients with advanced gastrointestinal stromal tumors who received treatment with imatinib, and the criteria were evaluated by computed tomography (CT) scans and were validated based on progression-free survival (PFS).22, 23 It is noteworthy that pathologic responses without any decrease in size were reported even in patients with solid tumors who received cytotoxic chemotherapy. Patients with myxoid liposarcoma who received trabectedin24, 25 and patients with osteosarcoma who received chemotherapy26-28 are examples among sarcomas.
We previously demonstrated that Choi criteria adapted to magnetic resonance imaging (MRI) correlated better with pathologic tumor response than RECIST in a series of 38 patients with high-grade soft tissue sarcomas who received preoperative chemotherapy with or without radiation therapy within a prospective phase 3 Italian Sarcoma Group/Spanish Sarcoma Group trial.29 The results of the trial recently were reported.30 Herein, we evaluate a comparison between RECIST and Choi criteria in eligible patients and correlate response to overall survival (OS) and freedom from progression (FFP).
MATERIALS AND METHODS
From January 2002 and March 2007, a prospective, randomized study of preoperative chemoradiation therapy for localized, high-risk (high grade, deep, ≥5 cm in greatest dimension) soft tissue sarcoma of the extremities and trunk wall enrolled 321 eligible patients in Italy (10 sites) and Spain (9 sites).29 In this retrospective analysis, we included all patients who entered the study with measurable disease and completed the study protocol treatment; these patients were evaluated before and after treatment by MRI or CT. The current study was conducted on a total of 256 patients, including 143 patients who received neoadjuvant chemotherapy and radiotherapy and 113 patients who received neoadjuvant chemotherapy and adjuvant radiotherapy. Among the last group of 113 patients treated preoperatively only with chemotherapy, we selected the subgroup of patients who were fully evaluable for both RECIST (version 1)1 and Choi criteria.22 Table 1 reports the frequency distribution of RECIST, Choi criteria, and histology for the whole series of 256 patients as well as in the 113 patients who received only preoperative chemotherapy. Institutional review boards at each institution approved the trial, and signed informed consent was obtained for all patients.
|Variable||No. of Patients||%||No. of Patients||%|
Patients received 3 cycles of epirubicin 120 mg/m2 plus ifosfamide 9000 mg/m2. Then, they underwent surgery within 4 to 6 weeks after the end of chemotherapy. Radiologic response was evaluated according to RECIST (version 1)1 and Choi criteria22 and was centrally reviewed.
Magnetic Resonance Imaging and Computed Tomography Techniques
Before surgery, MRI or CT studies were obtained at baseline and after chemotherapy with or without radiotherapy. The scans were collected from every center, and the images were centrally reviewed at Istituto Nazionale Tumori (Milano, Italy), by 2 expert sarcoma radiologists (A.M. for MRI, C.M. for CT) who were blinded to patient outcomes.
Only patients who were studied using the same technique (MRI or CT) before and after treatment and whose radiologic studies were available on a Digital Imaging and Communications in Medicine (DICOM) file and loaded into a multimodality work place (ie, Picture Archiving and Communication System [PACS]; Syngo [Siemens Medical Solutions, New York, NY]; Leonardo [Siemens Medical Solutions]) at Istituto Nazionale Tumori were considered eligible for response evaluation by Choi criteria. The images had to be available in portal venous phase for CT scans (slice thickness 5 mm). For MRI, turbo-spin-echo (TSE) T2-weighted images and TSE T1-weighted images (slice thickness, 5 mm) followed by a contrast-enhanced TSE T1-weighted image were requested.
To evaluate MRIs, contrast-enhanced TSE T1-weighted images were evaluated before and after digital subtraction (by A.M.). All data sets were evaluated both qualitatively and semiquantitatively. The percentage of contrast enhancement was assessed as follows: on subtracted, contrast-enhanced, T1-weighted sequences, the radiologist (A.M.) measured the contrast enhancement of the tumor by manually drawing a region of interest around the margin of the whole lesion on sections taken every 5 mm, thus encompassing the entire tumor mass. Muscle was used as the reference tissue. The contrast enhancement measurements from all sections were added, and the average contrast enhancement for each lesion was calculated.
To evaluate CT scans, the radiologist (C.M.) measured CT attenuation values of the tumor lesion in Hounsfield units on CT images that were obtained during the portal venous phase. Two-dimensional regions of interest that included the entire lesion were drawn, and all axial slices encompassing the lesion were included. To accurately define the lesion, the radiologist manually edited the contours of the regions of interest on each axial slice. The volume of interest covering the whole lesion was defined as the sum of all 2-dimensional regions of interest. Both cystic and necrotic areas were included. The software semiquantitatively calculated the mean tumor density (in Hounsfield units), which was defined as the average of all pixels enclosed in the volume of interest.
The 2 radiologists retrospectively reviewed all MRI and CT images. The percentage of change in mean tumor density/contrast enhancement before and after treatment was computed for each lesion.
Tumor response was evaluated using RECIST and Choi criteria according to criteria defined for gastrointestinal stromal tumors and adapted to MRI.1, 22, 28 We applied Choi criteria to MRI, assuming that changes in contrast enhancement on subtracted, contrast enhanced, T1-weighted sequences would parallel the changes in density on CT scans, because both are markers of tumor vascularization. According to Choi criteria, a partial response (PR) was defined as a decrease ≥10% in tumor size or a decrease ≥15% in tumor density/contrast enhancement on CT/MRI studies; progression was defined as new lesions, an increase ≥10% in tumor size without meeting any criteria for a PR according to tumor density/contrast enhancement, or an increase ≥15% in tumor density/contrast enhancement.
A retrospective, centralized pathologic review of pretreatment material was performed by a panel of Italian Sarcoma Group/Spanish Sarcoma Group pathologists with expertise in sarcoma to confirm diagnoses and histologic grading. A pathology review using World Health Organization criteria31 was carried out in 222 patients (71%). In the remaining 89 patients, the diagnosis made at the referral center was considered. For statistical analyses, patients were divided into 4 groups: undifferentiated pleomorphic sarcoma (formerly malignant fibrous histiocytoma) and spindle cell sarcoma not otherwise specified (Group 1), leiomyosarcoma (Group 2), synovial sarcoma (Group 3), and other (Group 4).
In addition to the whole series of patients, we focused on the group of patients who received preoperative chemotherapy alone and were evaluable by both RECIST and Choi criteria. Responses according to RECIST and Choi criteria were investigated in terms of diagnostic concordance as well as the prognostic capability of their original ordinal measurement scale (3 categories). Subsequently, because of the small number of patients in this subgroup, RECIST and Choi criteria variables were dichotomized according to 2 different criteria and were investigated in terms of concordance and prognosis.
Concordance between dichotomous and ordinal variables was assessed by computing (unweighted) Cohen kappa statistics (κ) and weighed kappa statistics (κw), respectively.32 The kappa statistics (both unweighted and weighed) usually lie between 0 (absence of agreement) and 1 (absolute agreement), and negative values may be obtained for situations in which actual agreement is less than chance. Thus, as previously reported33 κw values were considered satisfactory if they were equal to or greater than 0.80, whereas values of κ were interpreted in a qualitative manner based on the report by Landis and Koch.34
To assess the association between the various response criteria and histology and clinical outcomes, we performed a survival analysis in terms of both OS and FFP, which we defined as the time elapsed from randomization to death and to the first evidence of recurrence, respectively. The patterns of OS and FFP were estimated using the Kaplan-Meier method.35 The prognostic role of each variable (univariate analysis) on OS and FFP was investigated using a Cox regression model.36 In this model, each regression coefficient represents the logarithm of the hazard ratio, which is assumed to be constant over time. Under the null hypothesis that a variable has no prognostic role on OS and FFP, the hazard ratio is expected to be 1.00. The hypothesis of a hazard ratio of 1.00 was tested using the Wald statistic. In the Cox regression model, histologic subtype was categorized in 4 dummy variables using undifferentiated pleomorphic sarcoma/spindle cell sarcoma not otherwise specified as the reference category. When RECIST and Choi criteria were considered on their original scale, 2 dummy variables were generated by using the putative better prognosis category (ie, PR) as the reference category.
The risk of local and distant recurrence after surgery was assessed using a competing risks approach.37 All statistical analyses were performed using SAS statistical software (version 9.2; SAS Institute, Inc., Cary, NC).
Figure 1 provides 2 examples of response according to Choi criteria evaluated with MRI and CT studies.
Clinical Outcome and Prognostic Value of RECIST, Choi Criteria, and Histology
Overall series of patients
In total, 243 of 256 patients who were included in this study were evaluable for response (13 were missing radiologic data). Fifty-five were evaluated by CT, and 188 were evaluated by MRI.
At a median follow-up of 62 months (range, 28-105 months), the probability of OS (Fig. 2A) and FFP for the overall series was 0.65 (95% Confidence Interval (CI), 0.58-0.71) and 0.54 (95% CI, 0.47-0.60), respectively. Recurrent disease developed in 114 patients, whereas 88 patients died after randomization. Among the 114 patients who relapsed after randomization, 17 developed a local recurrence, whereas 97 had distant metastases as the primary event either alone or concurrent with local recurrence. In univariate Cox regression analysis (Table 2), RECIST, Choi criteria, and histology were significantly predictive for both OS (Fig. 2B, 2C and 2D) and FFP.
|HR (95% CI)|
|PD||2.57 (1.33-4.96)||2.84 (1.55-5.18)||1.63 (0.61-4.41)||1.88 (0.78-4.55)|
|SD||0.95 (0.54-1.69)||1.09 (0.65-1.82)||0.56 (0.24-1.34)||0.60 (0.28-1.28)|
|PD||6.89 (3.55-13.33)||6.33 (3.33-12.04)||6.85 (2.51-18.74)||5.24 (2.05-13.39)|
|SD||1.94 (0.87-4.36)||1.35 (0.61-2.99)||3.18 (1.17-8.63)||1.49 (0.55-4.03)|
|Leiomyosarcoma||2.78 (1.56-5.00)||1.72 (1.02-2.93)||2.83 (0.99-5.74)||1.09 (0.51-2.33)|
|Synovial sarcoma||1.63 (0.93-2.86)||1.13 (0.70-1.82)||1.67 (0.69-4.01)||0.94 (0.46-1.91)|
|Other||1.59 (0.88-2.88)||1.10 (0.66-1.82)||1.99 (0.81-4.91)||1.09 (0.52-2.27)|
Patients who received preoperative chemotherapy alone (113 patients)
Among the 113 patients who received chemotherapy alone, the 5-year probability of OS and FFP was 0.65 (95% CI, 0.55-0.74) and 0.51 (95% CI, 0.41-0.60), respectively. Recurrent disease was observed in 53 patients, whereas 39 patients died after randomization. Among the 53 patients who relapsed after randomization, 11 developed a local recurrence, whereas 42 had metastases as the primary event either alone or concurrent with local recurrence. In univariate Cox analysis (Table 2), RECIST were not associated with the 2 clinical outcomes. Conversely, Choi criteria maintained their predictive role for both OS and FFP, whereas histology was weakly associated only with OS.
Patients who received preoperative chemotherapy alone and were evaluable by both RECIST and Choi criteria (69 patients)
An unsatisfactory level of agreement between the 2 diagnostic criteria was observed with a κw statistic value of 0.46 (95% CI, 0.28-0.63). Overall, the probability of OS and FFP in this subset of patients was 0.67 (95% CI, 0.56-0.79) and 0.56 (95% CI, 0.44-0.69), respectively. Recurrent disease developed in 29 patients, and 24 patients died after randomization. Among the 29 patients who relapsed after randomization, 6 developed a local recurrence, whereas 23 had distant metastases as the primary event either alone or concurrent with local recurrence. The frequency distribution of the variable histology in this subset was: 1) undifferentiated pleomorphic sarcoma and spindle cell sarcoma not otherwise specified, n = 29 patients; 2) leiomyosarcoma, n = 14 patients; 3) synovial sarcoma, n = 12 patients; and 4) other histologies, n = 14 patients. The overall cumulative incidence of local recurrence and distant metastasis (±standard error) was 0.09 ± 0.03 and 0.35 ± 0.03, respectively (Fig. 3A).
Prognostic Value of RECIST, Choi Criteria, and Histology
Overall series of patients
In the univariate Cox analysis, with PR used as the reference category (Table 3), RECIST did not have any prognostic role in OS (Fig. 3B); and, for FFP, only the progressive disease (PD) category was associated significantly with outcome compared with the PR category. With Choi criteria, the considered comparisons (PD vs PR and stable disease [SD] vs PR) were associated significantly with OS (Fig. 3C). Conversely, only the comparison between PD and PR resulted significantly associated with FFP. Histology did not appear to be associated with the 2 considered clinical outcomes in this subset of patients (Fig. 3D).
|PD||2.26 (0.77-6.88)||2.85 (1.05-7.70)|
|SD||0.56 (0.22-1.45)||0.48 (0.20-1.12)|
|PD||6.85 (2.51-18.74)||5.24 (2.05-13.39)|
|SD||3.18 (1.17-8.63)||1.49 (0.55-4.03)|
|Leiomyosarcoma||1.68 (0.53-5.30)||1.05 (0.40-2.77)|
|Synovial sarcoma||2.04 (0.64-6.45)||0.99 (0.35-2.78)|
|Other||2.35 (0.82-6.70)||0.81 (0.29-2.27)|
RECIST and Choi Criteria on Dichotomous Scale
In evaluating RECIST and Choi criteria variables on a dichotomous scale, we compared responses to preoperative chemotherapy (ie, a PR according to RECIST and Choi criteria) with a lack of response (ie, SD + PD according to RECIST and Choi criteria) and observed fair agreement between the 2 diagnostic criteria with a κ statistic of 0.22 (95% CI, 0.10-0.34). Furthermore, when these variables were considered based on complementary dichotomization, that is, with versus without PD (ie, PR + SD vs PD according to RECIST and Choi criteria), we observed substantial agreement between the 2 diagnostic criteria with a κ statistic of 0.79 (95% CI, 0.16-1.00).
The prognostic role of both RECIST and Choi criteria also was explored by considering the 2 types of dichotomization mentioned above. For Choi criteria, both dichotomizations were significantly predictive for OS and FFP in the univariate Cox model; whereas, for RECIST, only the variable dichotomized as PR + SD versus PD was associated significantly with both clinical outcomes (data not shown).
In this study on neoadjuvant chemotherapy for high-risk soft tissue sarcoma (with or without preoperative radiotherapy and with or without 2 additional cycles of adjuvant chemotherapy), tumor response to preoperative chemotherapy with or without radiotherapy was associated with a better outcome. However, Choi criteria, extended to MRI, exhibited a better prognostic role than RECIST (Table 2). In fact, the outcome of patients who had SD according to RECIST was identical to the outcome of patients who had a PR. Conversely, in the subgroup of 69 patients who received preoperative chemotherapy alone and were evaluable by both RECIST and Choi criteria, Choi criteria were better prognosticators than RECIST (Table 3), with most patients who had SD according to RECIST (72%) were classified with PRs according to Choi criteria; furthermore in 2 patients, PD according to RECIST was classified as a PR according to Choi criteria. A trend toward a better outcomes was observed in patients who had a PR according to RECIST compared with those who had PD; however, paradoxically, those who had SD according to RECIST fared better than those who had a PR. Conversely, patients who had a PR according to Choi criteria had a trend toward better survival compared with patients who had PD and SD. The comparison between response to preoperative chemotherapy (ie, complete response + PR, according to both RECIST and Choi criteria) and lack of response (ie, RECIST/Choi criteria SD + PD) indicated that responders according to Choi criteria had better outcomes than nonresponders, whereas responses according to RECIST were not correlated significantly either with OS or FFP compared with nonresponses. Conversely, lack of progression (CR + PR + SD) according to RECIST was correlated with better outcomes compared with PD according to RECIST, similar to what was observed with Choi criteria. In brief, several patients with SD according to RECIST had a PR according to Choi criteria, and this correlated favorably with the outcome. Overall, histology had a significant prognostic role for both OS and FFP: undifferentiated pleomorphic sarcoma was the histotype with the most favorable outcome, and leiomyosarcoma had the worst. Consistently, undifferentiated pleomorphic sarcoma had a higher response rate than leiomyosarcoma according to Choi criteria.
This analysis has some limitations. First, it is an unplanned subgroup analysis. Second, the number of patients was small, because they had to be evaluable by both RECIST and Choi criteria, and those who also received preoperative radiotherapy are at least less meaningful, because the local effect of radiation therapy does not necessarily have any major systemic impact. The main limitation is that this trial did not have a control arm of patients who did not receive adjuvant therapy. Therefore, it cannot settle the long-lasting issue of the efficacy of adjuvant therapy in soft tissue sarcoma.38-41 If the efficacy were nil, then there could not be any correlation in principle between the tumor response and the outcome. Conversely, detecting some correlations, as we did, does not mean that adjuvant chemotherapy is effective (because of the potential selection bias, which theoretically could lead independently to both a tumor response and a better outcome).
We previously demonstrated that Choi criteria were better predictors of pathologic response than RECIST in a subgroup of patients who were treated on this study.28 At the time, we still lacked data on patient outcomes. Currently, although we have data only in a limited number of patients, we can state that Choi criteria also are better than RECIST in predicting survival. Unfortunately, we could not study pathologic responses in all patients because of differences in tumor sampling between participating centers.
In the current study, we made an effort to refine the way we assess nondimensional responses. Choi criteria are complex for use in daily routine practice, because they require postprocessing assessment by dedicated software with the help of a radiologist in the segmentation of lesions. In addition, they allow for only semiquantitative evaluation. Thus, perfusion and diffusion MRI and CT studies, dynamic contrast-enhanced ultrasound studies, and PET scans also are planned in our new ongoing trial to better describe tumor tissue changes and for the evaluation of quantitative parameters and radiologic-pathologic correlations.
In our series, patients with SD according to RECIST fared better than those who had a PR according to RECIST. It is noteworthy that most SD according to RECIST corresponded to a PR according to Choi criteria. This indicates that nondimensional, or minor, responses brought about a better outcome than dimensional responses, which we also were able to confirm by combining RECIST and Choi criteria. This is intriguing, although the small number of patients, of course, is a limiting factor for any further speculation. However, the patients received 3 cycles of preoperative chemotherapy; thus, major responses actually are fast responses, and they may correspond to the most aggressive tumors. In these patients, chemotherapy may have exerted a limited prognostic impact overall. The value of such a hypothesis will require further assessment.
It is noteworthy that, in terms of the shortcomings of tumor response with targeted agents, PFS often is advocated as a suitable indicator of antitumor activity. However, in our series of 166 patients, 7 patients who were classified with PD according to RECIST had a PR according to Choi criteria, and 2 patients who were classified with SD according to RECIST had PD according to Choi criteria. In other words, using Choi criteria would result in a different assessment of PD and, thus, would change the estimated PFS.
In our study, patients with leiomyosarcoma fared worse than patients who had other histotypes. This result correlated with the responses defined according to Choi criteria, in which leiomyosarcoma is the histotype with the lowest percentage of responses and the highest progression rate. It is noteworthy that Choi criteria again were better predictors than RECIST. These results suggest that chemosensitivity may have an impact on patient outcomes, possibly in combination with the natural history of each histotype. In particular, it is believed that leiomyosarcomas are less sensitive to ifosfamide if they are sensitive at all to this drug.
In conclusion, currently, there is much discussion about tumor response criteria in solid tumors. There is a trend toward nondimensional criteria, especially when molecularly targeted agents are used. Our current analysis suggests that tumor response to chemotherapy may benefit from newer criteria. Further work is needed to refine criteria for nondimensional assessment of tumor responses.
No specific funding was disclosed.
CONFLICT OF INTEREST DISCLOSURES
The authors made no disclosures.
- 1New guidelines to evaluate the response to treatment in solid tumors. European Organization for Research and Treatment of Cancer, National Cancer Institute of the United States, National Cancer Institute of Canada. J Natl Cancer Inst. 2000; 92: 205-216., , , et al.
- 22Correlation of computed tomography and positron emission tomography in patients with metastatic gastrointestinal stromal tumor treated at a single institution with imatinib mesylate: proposal of new computed tomography response criteria. J Clin Oncol. 2007; 25: 1753-1759., , , at al.
- 31Pathology and Genetics. Tumors of Soft Tissue and Bone. Lyon, France: IARC Press; 2002., , , eds.
- 32Statistical Methods for Rates and Proportions. 2nd ed. New York: John Wiley and Sons; 1981..
- 35Nonparametric estimation from incomplete observations. J Am Stat Assoc. 1958; 53: 457-481., .
- 36Regression models and life tables (with discussion). J R Stat Soc B. 1972; 34: 187-220..
- 37Analysing Survival Data from Clinical Trials and Observational Studies. Chichester, United Kingdom: Wiley; 1995., .
- 40EORTC Soft Tissue and Bone Sarcoma Group. Adjuvant chemotherapy with doxorubicin and ifosfamide in resected soft tissue sarcoma: Interim analysis of a randomized phase III trial [abstract]. J Clin Oncol. 2007; 25( 18S). Abstract 10008., , , et al;