Presented at the 39th Annual Meeting of the American Pediatric Surgical Association, Phoenix, Arizona, May 29-13, 2008.
The objectives of this study were to compare tumor volume and patient weight versus traditional factors of tumor size (greatest dimension) and patient age and to determine which parameters best discriminated outcome among pediatric patients with intermediate-risk rhabdomyosarcoma (RMS).
Complete information was available for 370 patients with nonmetastatic RMS who were enrolled in the Children's Oncology Group (COG) intermediate-risk study D9803 (1999-2005). The Kaplan-Meier method was used to estimate survival distributions. A recursive partitioning model was used to identify prognostic factors that were associated with event-free survival (EFS). Cox proportional hazards regression models were used to estimate the association between patient characteristics and the risk of failure or death.
For all patients with intermediate-risk RMS, a recursive partitioning algorithm for EFS suggested that prognostic groups should be defined optimally by tumor volume (with a transition point at 20 cm3), patient weight (with a transition point at 50 kg), and embryonal histology. Tumor volume and patient weight added significant outcome information to the standard prognostic factors, including greatest tumor dimension and patient age (P = .02). The ability to resect the tumor completely was not associated significantly with the size of the patient, and patient weight did not significantly modify the association between tumor volume and EFS after adjustment for standard risk factors (P = .2).
Rhabdomyosarcoma (RMS) is the most common malignant soft tissue tumor of childhood. Through clinical trials using multimodality therapy, survival has improved steadily over the last 3 decades for these patients.1 Patients are stratified into a low-risk, intermediate-risk, or high-risk category based on the clinical group (completeness of surgical resection at presentation), pretreatment tumor, lymph node, metastasis (TNM) classification (site and size of the primary tumor, tumor invasion into surrounding tissues, lymph node status, and the presence of metastatic disease), and tumor histology.2 These prognostic factors, as well as age, predict outcome for patients with intermediate-risk tumors.3
For sarcomas, 5 cm has been used as the transition point to predict outcome and determine disease risk stratification; tumors that measure >5 cm in greatest dimension are associated with a worse prognosis. In addition, patient ages <1 year and >10 years are associated with a worse prognosis.1, 4-6 Although RMS occurs in patients of varied ages and body sizes, the size of the tumor is characterized by maximum single dimension, and the pretreatment TNM classification does not take into account the relation between the greatest tumor dimension and the patients' age or body size. Ferrari et al recently evaluated the association between the greatest tumor dimension and patient body size, which they quantified according to the body surface area, and correlated these factors with the outcomes of childhood soft tissue sarcomas (STS).7 In patients with localized RMS, for a given greatest tumor dimension, the mortality rate was higher for patients who had a lower body surface area. These data suggest that it is not the absolute greatest tumor dimension but the relation between that dimension and the body surface area that may be a superior discriminator of outcome. In a subsequent study, the same authors demonstrated that both tumor volume and greatest tumor dimension measurements at presentation were equally predictive of overall survival as well as response to induction chemotherapy.8
On the basis of this background, we compared the role of tumor volume and patient weight versus greatest tumor dimension and patient age as prognostic factors in the survival of patients with intermediate-risk RMS. In addition, we determined whether tumor volume relative to patient weight was more predictive of outcome than tumor volume alone.
MATERIALS AND METHODS
Children's Oncology Group (COG) study D9803 enrolled 570 patients with intermediate-risk RMS from 1999 to 2005, and that population was used for the current study. Patients who were eligible for this analysis had nonmetastatic alveolar RMS or nonmetastatic embryonal RMS arising in unfavorable sites. The overall results for COG D9803 have been reported previously.9 Tumor volume measurements or other data values were missing for 200 patients, leaving 370 patients for further evaluation. Measurements of the tumor volume and the greatest tumor dimension were obtained from diagnostic imaging studies in 96% of the patients who we analyzed and from physical examination measurements in 4% of patients. Informed consent was obtained from the study participants. Tumor volume was calculated using software associated with the imaging techniques used at each institution but, in general, was the product of the greatest perpendicular dimensions of the largest primary tumor; whereas the greatest tumor dimension was defined as the greatest dimension of the largest primary tumor.
The distributions of categorical patient characteristics were compared among independent groups using the chi-square test or the Fisher exact test when expected cell counts were small. The Kruskal-Wallis test was used to compare the distributions of continuous measures among independent groups. Event-free survival (EFS) was defined as the time from study enrollment to either disease progression or death. EFS for patients who had not experienced an event were censored at the patient's last date of contact. The Kaplan-Meier method was used to estimate the survival distribution for groups of patients defined by patient or disease characteristics, and the distributions were compared using a log-rank test. To identify risk groups, a recursive partitioning method was used.10, 11 Noting that some of the subgroups splits may over fit the data, a graphic approach was used to prune off splits that were not necessary.11 Some of the subgroups in the resulting tree may have had similar survival experiences and were combined by using a recursive partitioning algorithm based on the prognostic group identifiers. The recursive partitioning approach results in more easily interpretable prognostic groups, which are defined by higher order interactions, compared with the Cox proportional hazards regression modeling approach, and has been used in the oncology setting.10-12 Cox proportional hazards regression models were used to estimate the hazard of an event or death associated with risk groups, which were defined using cutoff points that were suggested by the recursive partitioning approach and rounded to the nearest 5 kg for weight, 1 cm for greatest tumor dimension, or 5 cm3 for tumor volume for ease of clinical interpretation. Changes in the likelihood-ratio test statistic were used to determine whether additional patient characteristics were associated with the hazard of an event beyond the identified risk-group factors. Data were current through June 2008, and the median follow-up was 4.4 years (range, 0.1-8.2 years).
Patient characteristics are summarized in Table 1. The patients who were analyzed in the current study (n = 370) differed significantly from the patients who were missing tumor volume measurements (n = 190) or other clinical data (n = 10) in terms of disease stage, clinical group, primary site, tumor size, tumor invasiveness, lymph node involvement, histology, and risk stratification. There were more patients with advanced disease in the group with complete information that was used for the tumor volume analysis. However, even with these differences, the 4-year EFS estimate was 0.71 (95% confidence interval [CI], 0.66-0.75) among the patients who were analyzed in this study (n = 370), which was similar to the estimate for the patients who were excluded from this study because of missing tumor volume measurements or other clinical data (n = 200; 4-year EFS, 0.71; 95% CI, 0.64-0.77) and to the estimate for the entire cohort (n = 570; 4-year EFS, 0.71; 95% CI, 0.67-0.75).
Table 1. Demographic and Disease Characteristics for Patients Who Were Included (n=370) and Excluded (n=200) From the Analysis Set
No. of Patients (%)
Patients With Tumor Volume Measures and Complete Covariate Information, n=370
Patients Missing Tumor Volume or Other Covariate Data, n=200
Prognostic Significance of Tumor Volume and Patient Weight
The recursive partitioning algorithm considered the following variables: patient age, disease stage, clinical group, tumor classification, lymph node status, histology, primary site, greatest tumor dimension, tumor volume, patient height, patient weight, body surface area, and treatment regimens. On the basis of the recursive partitioning algorithm, when considering all variables (including tumor volume and patient weight), the best predictor of outcome was tumor volume (with a transition point of 20 cm3). Among patients who had tumor volumes ≥20 cm3, the model suggested splitting the patients into groups depending on weight (with a transition point of 50 kg). Finally, patients who had tumor volumes ≥20 cm3 and who weighed <50 kg were stratified based on tumor histology (embryonal or nonembryonal). The recursive partitioning model resulted in 4 risk groups: 1) patients with tumor volumes <20 cm3 (n = 58; 4-year EFS, 0.95; 95% CI, 0.84-0.98), 2) patients with tumor volumes ≥20 cm3 and weight <50 kg who had embryonal disease (n = 157; 4-year EFS, 0.77; 95% CI, 0.69-0.83), 3) patients with tumor volumes ≥20 cm3 and weight <50 kg who had nonembryonal disease (n = 99; 4-year EFS, 0.60; 95% CI, 0.49-0.69), and 4) patients with tumor volumes ≥20 cm3 and weight ≥50 kg (n = 56; 4-year EFS, 0.47; 95% CI, 0.33-0.60) (Fig. 1A). Kaplan-Meier EFS curves for these risk groups are presented in Figure 1B and indicate the prognostic utility of tumor volume and patient weight with patients separated into distinct risk groups (overall P<.001; log-rank test). Disease and patient characteristics for each risk-stratification group are described in Table 2. Characteristics related to the factors that we used to define the risk groups—patient age, tumor stage, tumor histology, and tumor size—differed significantly among the risk groups. In addition, other standard prognostic factors also differed significantly among the 4 risk groups.
Table 2. Characteristics for Risk Groups According to Tumor Volume and Patient Weight
Distributions were compared among groups using a chi-square test or Fisher exact test. No formal comparison was made for factors that involved histology, which was used directly in the risk group definitions.
Greatest tumor dimension, cm
Regional lymph node status
Embryonal, stage 2/3, Group III
ALV/UDS, stage 1 or Group I
PM with extension
Prognostic Significance of Greatest Tumor Dimension and Patient Age
To determine whether the greatest tumor dimension and patient age, which are known predictors of outcome, were equivalent to tumor volume and patient weight for prognostic stratification, the recursive partitioning method was repeated after tumor volume, patient weight, and patient body surface area were excluded (Fig. 2A). In this model, the most important predictor of outcome was the greatest tumor dimension. Patients who had small tumors were stratified further based on greatest tumor dimension and primary tumor site. Patients who had large tumors were stratified further based on age and primary tumor site. The analysis that evaluated the traditional prognostic factors of greatest tumor dimension and patient age resulted in 6 risk groups that, based on EFS, could be combined into 3 distinct risk-group categories (Table 3). EFS Kaplan-Meier curves for these risk groups are presented in Figure 2B. Disease and patient characteristics for each risk group are described in Table 4. There were differences in patient age, tumor stage, tumor primary site, histology, lymph node status, tumor invasiveness, and tumor size among the risk groups.
Table 3. Risk Stratification for Event-Free Survival Based on Traditional Criteria of Greatest Tumor Dimension and Patient Age
Risk-stratification groups are defined in Table 3.
Distributions were compared among groups using a chi-square test or Fisher exact test. No formal comparison was made for factors that involved age, greatest tumor dimension, or primary site, which were used directly in the risk group definitions.
Tumor size, cm
Regional lymph node status
Embryonal, stage 2/3, Group III
ALV/UDS, stage 1 or Group I
PM with extension
Comparison of Volume/Weight Versus Greatest Tumor Dimension/Patient Age
The correlation between the risk categories using the 2 approaches (tumor volume/patient weight vs greatest tumor dimension/patient age) was better for the highest and lowest risk categories and was worse for the middle risk categories. Also, there were some patients who were classified in the lowest risk category based on greatest tumor dimension, age, and primary tumor site but were classified in the highest risk category based on tumor volume, weight, and histology (Table 5). Thus, although the correlation was good, it is possible that 1 strategy was better at predicting outcome. To determine whether greatest tumor dimension plus patient age were superior to tumor volume plus patient weight for predicting outcome, a Cox proportional hazards regression model was fit with the 3 risk categories defined by greatest tumor dimension, patient age, and primary tumor site. Then, the risk grouping based on volume and weight was entered into the model. This methodology can determine whether the addition of volume and weight risk categories adds significant information about the risk of failure. The difference in the likelihood-ratio statistic for each of these models was determined, and a P value for the loss of information was determined using a chi-square distribution with 3 degrees of freedom. In the first model (which included risk categories identified by greatest tumor dimension, patient age, and primary tumor site as well as risk categories defined by tumor volume, patient weight, and tumor histology), the likelihood-ratio test statistic was 63.22 with 5 degrees of freedom. For the second model (which included risk categories identified by greatest tumor dimension, patient age, and primary tumor site), the likelihood-ratio test statistic was only 53.36 with 2 degrees of freedom. The addition of tumor volume, patient weight, and tumor histology was important, and the difference between the likelihood-ratio test statistic for the 2 models was 9.86 (P = .02). Therefore, the addition of tumor volume and patient weight provided significant prognostic information to a model that included risk categories defined by the traditional prognostic factors, including greatest tumor dimension and patient age.
Table 5. Correlation of Risk Stratification Based on Tumor Volume and Patient Weight Versus Greatest Tumor Dimension and Patient Age
Group 4: Tumor volume ≥20 cm3, weight ≥50 kg (n=56)
In summary, the factors that had the strongest association with the risk of failure were tumor volume, patient weight, and tumor histology. When volume and weight were not included as potential factors, greatest tumor dimension, patient age, and primary tumor site had the strongest association with the risk of failure. Tumor volume and patient weight were superior predictors of outcome and may be better measures of tumor burden and patient size. In addition, patient size, relative to tumor size, is important when considering larger tumors. Among children with larger tumors, patients who are older have poorer outcomes.
Association Between Tumor Size, Patient Weight, and Completeness of Tumor Resection and Outcome
Completeness of tumor resection, defined by clinical group, is a component of disease risk stratification. It is possible that tumor size in relation to patient size may influence the clinical group and, thus, the outcome. Completeness of surgical resection was quantified for each patient (Clinical Group I, complete resection; Clinical Group II, microscopic residual disease; and Clinical Group III, gross residual disease). A nonparametric Kruskal-Wallis test was used to evaluate the relation between patient size and clinical group. There was no significant difference in median age (P > .9) or median weight (P = .8) among the 3 clinical group categories (Table 6). This was further supported using a Cox proportional hazard regression model to evaluate the relation between the distribution of EFS and tumor and patient sizes after adjustment for histology, primary site, completeness of surgical resection (clinical group), disease stage, lymph node status, and tumor classification. When considering cutoff points based on median weight and tumor volume, there was no significant indication that patient weight (<20 kg vs ≥20 kg) significantly modified the association between tumor volume (<90 cm3 vs ≥90 cm3) and EFS after adjustment for standard risk factors (P = .2). Patients were split into 4 roughly equally sized groups defined by cutoff points of 90 cm3 for tumor volume and 20 kg for patient weight. By using large children (≥20 kg) with small tumors (<90 cm3) as the reference group (n = 91), suggestive of lower relative tumor burden, the risk of failure increased by 33% for small children with small tumors (n = 92; P = .4), increased by 63% for small children with large tumors (n = 87; P = .2), and increased by 112% for large children with large tumors (n = 100, P = .02). Although there seemed to be a trend, the difference between large patients with small tumors (suggestive of lower relative tumor burden) and small patients with large tumors (suggestive of higher relative tumor burden) was not statistically significant (Table 7). To further investigate this issue, the ratio of tumor volume (cm3) to patient weight (kg) was calculated, and patients were grouped according to quartiles of this ratio. After adjustment for histology, primary site, completeness of surgical resection (clinical group), disease stage, lymph node status, and tumor classification, the increased hazard of failure for patients who had a higher tumor-volume-to-weight ratio values, relative to the those with tumor-volume-to-weight ratio values <1.25 cm3/kg (n = 91), was not statistically significant (ratio values from 1.25 cm3/kg to 3.99 cm3/kg [n = 95]: hazard ratio [HR], 1.71; 95% CI, 0.91-3.20; P = .09; ratio values from 4.00 cm3/kg to 9.99 cm3/kg [n = 93]: HR, 1.92; 95% CI, 0.97-3.79; P = .06; and ratio values ≥10 cm3/kg [n = 91]: HR, 1.67; 95% CI, 0.73-3.83; P = .2).
Table 6. Association Between Patient Weight and Completeness of Resection by Surgical (Clinical) Group
SD indicates standard deviation.
Group I (n=12)
Group II (n=32)
Group III (n=326)
Table 7. Impact of Tumor Volume (Split at <90 cm3) and Patient Weight (Split at <20 kg) on Event-Free Survival Outcome
In this study, we demonstrated that the factors that had the strongest association with the risk of failure among patients with intermediate-risk RMS were tumor volume, patient weight, and tumor histology. When volume and weight were not used, greatest tumor dimension, patient age, and primary tumor site had the strongest association with the risk of failure. Although, overall, tumor volume and patient weight are mathematically superior predictors of outcome, modeling based on the greatest tumor dimension and patient age also can be used to identify the risk of disease recurrence for subgroups within the intermediate-risk population. In addition, our study demonstrated that the ability to resect the tumor completely was not associated significantly with the size of the patient, and patient weight did not significantly modify the association between tumor volume and EFS after adjustment for standard risk factors.
The prognosis of patients with RMS depends on many factors. Favorable prognostic factors include embryonal/botryoid histology, favorable primary tumor sites, no distant metastases at diagnosis, complete gross removal of tumor at the time of diagnosis, tumor size ≤5 cm, and age <10 years at the time of diagnosis.6, 12-14 The extent of surgical resection (ie, clinical group) was identified as 1 of the most important predictors of treatment failure and tumor recurrence.6, 13-15
The study by Meza et al evaluated a total of 1258 patients without distant metastases enrolled into either Intergroup Rhabdomyosarcoma Study (IRS) III or IRS IV between 1984 and 1997.3 In that study, prognostic factors for alveolar RMS included tumor classification and clinical group, age <1 year, and locoregional lymph node disease. For patients with embryonal RMS, both tumor classification and clinical group were associated significantly with failure-free survival (FFS). Other predictive factors included age <1 year or >10 years, unfavorable primary tumor site, tumor size >5 cm, and tumor invasiveness. The FFS of those patients ranged from 90% for patients in Clinical Group I who had stage 1 embryonal RMS down to 45% for patients in Clinical Group III who had alveolar RMS at unfavorable sites. This broad range of FFS demonstrates the importance of prognostic factors to allow allocation of patients to the most appropriate intensity of therapy. In this current study, patients with intermediate-risk RMS from the COG D9803 study were analyzed to compare traditional prognostic factors with the potential prognostic factors tumor volume and patient weight. Our results support the conclusion that tumor volume and patient weight are important prognostic markers and may be more predictive than the traditional factors—greatest tumor dimension and patient age. These results differ from those reported by Ferrari et al.8 In their single-institution, retrospective study of 205 patients from 1982 to 2008, those authors demonstrated that greatest tumor dimension and tumor volume were equally predictive of survival.
It has been postulated that the size of the tumor relative to the size of the patient may have an impact on the outcomes for children with RMS. Ferrari et al demonstrated that, for patients with non-RMS STS, increased tumor size was a stronger unfavorable prognostic factor in children than in adults.16 In a subsequent study, the same group evaluated 553 patients aged <21 years who were treated for nonmetastatic STS between 1977 and 2005.7 The 5-year and 10-year overall survival estimates among the 304 patients with RMS were 64% and 63%, respectively. In that study, the greatest tumor dimension was identified as a statistically significant prognostic marker of overall survival, and the interaction between tumor size and body surface area was significant for patients with RMS (P = .03) but not for patients with other types of STS. An increasing greatest tumor dimension was identified as an unfavorable prognostic factor regardless of body surface area but particularly for patients with small body surface area. Those authors also demonstrated that, in tumors <2 cm, the effect of body surface area and mortality tended to be diminished. This finding was supported by our study, in which patients who had small-volume tumors had a very good EFS regardless of their weight. However, our study demonstrated no significant differences in the age or weight distribution among the clinical group categories, suggesting that patient weight did not have an impact on successful tumor resection. Although large children who had large tumors had a significantly poorer prognosis relative to large children who had small tumors, there was no significant indication that the outcomes were worse for small patients who had large tumors compared with large patients who had small tumors.
The differences in observations between the results reported by Ferrari et al7, 8 and our current study may reside in the different treatment paradigms used by European and North American studies, specifically regarding the use of radiotherapy. It is possible that the more aggressive local control philosophy used in COG studies minimizes the impact of tumor size in relation to patient size. In addition, the study by Ferrari et al included all patients with nonmetastatic RMS (low-risk and intermediate-risk patients), whereas our study only evaluated intermediate-risk patients. Another possible explanation is that weight may be a more reliable prognostic indicator compared with body surface area when assessing the contribution of body proportion to outcome. In our analysis, patient weight, but not body surface area, was selected as an important prognostic factor, thus indicating that weight may be a better indicator of patient size. In addition, our observations held true regardless of whether we used tumor volume or greatest tumor dimension as the indicator of tumor size.
The current study has several limitations. Tumor measurements (greatest tumor dimension and tumor volume) were submitted by institutions after a local review of imaging studies. Variations in the techniques that were used for tumor measurement may have led to inaccurate reporting of tumor size. Tumor volume was missing for 33% of the D9803 study participants. The patients who were analyzed in our report differed significantly from the patients who were missing tumor volume measurements or other clinical data. These differences may have had an uncontrolled impact on our current results and conclusions. Finally, our analysis included only patients with intermediate risk RMS, and our findings cannot be extended to either low-risk or high-risk patients without further evaluation. Validation of these results will be required to extend the observations to all patients with RMS.
In conclusion, tumor volume and patient weight are strongly associated with the risk of failure and may have better predictive value than traditional prognostic factors, may improve disease risk stratification, and subsequently may allow more accurate tailoring of therapy to disease risk. To apply these findings more broadly to all patients with RMS, they will need to be confirmed in patients with low-risk and high-risk RMS.
CONFLICT OF INTEREST DISCLOSURES
This research was supported by grants U10 CA24507 (to the Intergroup Rhabdomyosarcoma Study Group), U10 CA98413 (to the Statistics and Data Center, and U10 CA98543 (to the Children's Oncology Group Chair) from the National Cancer Institute.