Fax: (011) 31-20-5122554
Pretreatment probability model for predicting outcome after intraarterial chemoradiation for advanced head and neck carcinoma
Version of Record online: 9 SEP 2004
Copyright © 2004 American Cancer Society
Volume 101, Issue 8, pages 1809–1817, 15 October 2004
How to Cite
van den Broek, G. B., Rasch, C. R. N., Pameijer, F. A., Peter, E., van den Brekel, M. W. M., Tan, I. B., Schornagel, J. H., de Bois, J. A., Zijp, L. J. and Balm, A. J. M. (2004), Pretreatment probability model for predicting outcome after intraarterial chemoradiation for advanced head and neck carcinoma. Cancer, 101: 1809–1817. doi: 10.1002/cncr.20556
- Issue online: 1 OCT 2004
- Version of Record online: 9 SEP 2004
- Manuscript Accepted: 1 JUL 2004
- Manuscript Revised: 28 JUN 2004
- Manuscript Received: 13 APR 2004
- head and neck neoplasms;
- intraarterial infusions;
- selective-targeted chemoradiation (RADPLAT);
- tumor volume;
- prognostic factors
Concurrent chemoradiation is being used increasingly to treat patients with advanced-stage head and neck carcinoma. In the current study, a clinical nomogram was developed to predict local control and overall survival rates for individual patients who will undergo chemoradiation.
Ninety-two consecutive patients with UICC TNM Stage III/IV squamous cell carcinoma of the oral cavity, oropharynx, hypopharynx, and supraglottic larynx were treated with selective-targeted chemoradiation (acronym: RADPLAT). All living patients had a minimum follow-up of 2 years. In addition to general factors, the following parameters were analyzed in a multivariable analysis: primary tumor volume, lymph node tumor volume, total tumor volume, lowest involved neck level, comorbidity, pretreatment hemoglobin level, pretreatment weight loss, and unilateral/bilateral intraarterial infusion. Relevant factors for local control and survival were analyzed using the Cox proportional hazards model.
At 5 years, the local control and overall survival rates for the whole group were 60% and 38%, respectively. Primary tumor volume (hazard ratio [HR], 1.03; P = 0.01) and unilateral infusion (HR, 5.05; P = 0.004) were found to influence local control significantly. Using tumor volume as a continuous variable, an adjusted risk ratio of 1.026 was found, indicating that each 1-cm3 increase in volume was associated with a 2.6% decrease in probability of local control. Primary tumor volume (HR, 1.01; P = 0.003), comorbidity (American Society of Anesthesiologists [ASA] physical status 1 vs. > 1; HR, 2.47; P = 0.01), lowest involved neck level (HR, 3.45; P = 0.007), and pretreatment weight loss > 10% (HR, 2.04; P = 0.02) were found to be significant predictors of worse overall survival. Variables from the multivariable analysis were used to develop a nomogram capable of predicting local control and overall survival.
Tumor volume was found to play a significant role in predicting local control and overall survival in patients with advanced-stage head and neck carcinoma who were treated with targeted chemoradiation. The nomograms may be useful for pretreatment selection of patients with advanced-stage head and neck carcinoma. Cancer 2004. © 2004 American Cancer Society.
Head and neck squamous cell carcinoma (HNSCC) is the fifth most common cancer in men, with a worldwide incidence of approximately 780,000 new cases per year.1 Greater than 70% of patients with head and neck carcinoma present with advanced-stage disease (i.e., Stage III/IV). Treatment for patients with advanced, inoperable HNSCC is a therapeutic challenge. There is increasing evidence that concomitant chemoradiation, compared with conventional radiotherapy, leads to improved local control (LC) and overall survival (OS) in patients with advanced-stage head and neck carcinoma. Therefore, concomitant chemoradiation is more suitable as a curative treatment option in these patients.2–8 Clinical trials comparing different treatment schedules (e.g., differences in the route of administration route [intraarterial vs. intravenous], chemotherapy dose, and radiation schedules) are ongoing to optimize the concomitant delivery of chemotherapy and radiation. However, chemoradiation is frequently associated with serious toxicity.3, 4 Patients unlikely to be cured with chemoradiation should ideally be recognized before treatment. Therefore, the assessment of factors that significantly influence LC and survival is essential. Prognostic factors for outcome include gender, hemoglobin level,9–11 comorbidity,12 and tumor volume.13 However, to our knowledge, information regarding the prognostic value of these factors in chemoradiation remains scarce. Unfortunately, powerful predictors used in surgical patients (i.e., depth of tumor invasion,14 number of positive lymph nodes, and extracapsular spread)15, 16 are not accessible for patients treated with primary chemoradiation.
In the current study, we investigated the role of the prognostic factors mentioned earlier in predicting LC and OS after the delivery of targeted chemoradiation in patients with advanced-stage head and neck carcinoma. To assess the simultaneous effect of various factors on predicting LC/OS, a logistic regression model was used. By combining the significant prognostic factors, we developed a clinical algorithm for LC and OS to facilitate clinical decision-making.
MATERIALS AND METHODS
Between April 1997 and May 2001, 105 consecutive patients with newly diagnosed inoperable T3–T4 squamous cell carcinoma of the oral cavity, oropharynx, hypopharynx, and supraglottic larynx were enrolled to receive targeted chemoradiation. With the exception of hypopharyngeal carcinoma, all tumors were classified as being surgically or functionally inoperable. Inclusion criteria were defined as follows: 1) oral cavity and base of tongue: no functional reconstruction possible after removal of the tumor, mainly including tumors requiring total glossectomy or resection of both hypoglossal nerves; 2) tonsil and soft palate: extension toward the base of skull as manifested by clinical trismus and apparent on imaging, making it highly unlikely to obtain clear surgical margins at the cranial border or requiring resection of the whole soft palate; 3) posterior pharyngeal wall tumors or hypopharyngeal carcinomas: tumor extensions requiring total laryngectomy and extensive reconstruction, or fixation to the cervical spine; and 4) supraglottic larynx and/or base of tongue: tumor extensions requiring total glossectomy and total laryngectomy for complete removal. All statements regarding unresectability were reviewed by the three head and neck surgeons involved in the current study (A.J.M.B., M.W.M.v.d.B., and I.B.T.).
Three patients were excluded because distant metastases were detected before the start of treatment. Another 10 patients were excluded because good quality pretreatment magnetic resonance imaging (MRI) scans were not available for tumor volume measurements, resulting in a study population of 92 patients. All living patients had ≥ 2-years of follow-up. The study population included 69 men and 23 women with a median age of 53 years (range, 29–78 years). Tumors were staged according to the guidelines of the International Union Against Cancer.17 The T and N classification distribution of the patient cohort was T3 (n = 22), T4 (n = 70), N0 (n = 33), N1 (n = 6), N2a (n = 1), N2b (n = 18), N2c (n = 26), and N3 (n = 8), resulting in 13 patients with Stage III and 79 patients with Stage IV disease. The following site distribution was established: oral cavity (n = 22), oropharynx (n = 58), hypopharynx (n = 9), and supraglottic larynx (n = 3). Patient, tumor, and treatment characteristics are summarized in Tables 1 and 2.
|Stage III||13 (14)|
|Stage IV||79 (86)|
|Oral cavity||22 (24)|
|Supraglottic larynx||3 (3)|
|Neck level involved|
|No, Level II–III||83 (90)|
|Level IV||9 (10)|
|Pretreatment weight loss|
|< 10%||47 (51)|
|Hemoglobin level (mmol/L)||8.5||6.2||10.7|
|Primary tumor volume (cm3)||35.4||6.4||393.0|
|Lymph node tumor volume (cm3)||18.1||2.3||131.7|
|Total tumor volume (cm3)||42.5||6.9||393.0|
Targeted chemoradiation was described in earlier studies.18, 19 Briefly, treatment was comprised of 4 consecutive weekly selective intraarterial infusions of cisplatin (150 mg/m2) simultaneously delivered with intravenous sodium thiosulfate rescue combined with radiation therapy (2 grays [Gy] per day, 5 per week × 7 to a total dose of 70 Gy) according to the RADPLAT protocol.20 In addition to the earlier reported intraarterial administration of cisplatin,21 we performed bilateral infusion in patients whose primary tumors extended across the midline, with equal distribution of cisplatin doses over both sides. Before the initiation of treatment, an informed consent form approved by our institutional protocol review committee was obtained from all patients.
Tumor volume assessment was performed by delineation of all visible tumor tissue on pretreatment MRI scans. MRI scans were performed on a 1.5-Tesla scanner (Siemens Magnetom 63 SP4000; Siemens, Erlangen, Germany). Slice thickness was ≤ 4 mm, with an interslice gap of ≤ 1 mm. The field of view for the axial views was 16–18 cm for T1-weighted sequences and 18–20 cm for T2-weighted sequences. T1-weighted images were obtained before and after injection of intravenous gadolinium. Postcontrast images were acquired using fat saturation. An experienced head and neck radiologist (F.A.P.), who was blind to the patients' outcome, performed the primary tumor volume delineations. Delineations as performed by the radiologist then were transferred to a computer with digitized MRI scan images and delineation tools (G.B.v.d.B.). Twenty-nine good-quality MRI scans from referral hospitals were redigitized for tumor volume measurements. Primary tumor volume data are shown in Figure 1. Volumes of lymph node metastases were calculated by the summation of all pathologic lymph node tissue on MRI scans (G.B.v.d.B.). To minimize the risk of measuring reactive lymph nodes, only lymph nodes fulfilling the following criteria were delineated: 1) shortest dimension ≥ 15 mm, 2) signs of central necrosis, or 3) confirmation of malignancy by (ultrasound-guided) fine-needle aspiration cytology.
Patient-related factors were gender, age, pretreatment hemoglobin level, pretreatment weight loss (percentage of body weight), and comorbidity (ASA physical status; always assessed before pretreatment examination under general anesthesia by the attending anesthesist). Tumor-related factors were T classification, N classification, TNM stage, primary tumor volume, lymph node tumor volume, total tumor volume, tumor site, and neck level involvement. The treatment-related factor was unilateral or bilateral intraarterial infusion of cisplatin.
Treatment response was evaluated 6–8 weeks after the completion of radiotherapy by MRI scans, followed by examination of the patient under general anesthesia. If the primary tumor site was macroscopically suspect for residual tumor, a biopsy was performed during examination under general anesthesia. Thereafter, patients were subjected to regular outpatient follow-up and an annual chest X-ray.
Differences in means were compared using the Student t test. Univariate (not shown) and multivariable analyses were performed to assess the effects of various factors on predicting outcome (locoregional control and OS). The Cox proportional hazards model22 was used to perform the multivariable analysis. Six of the 92 patients could not be assessed for LC after completion of treatment due to death caused by pneumonia (n = 3), cervical spondylitis (n = 1), ruptured abdominal aneurysm (n = 1), and arterial (carotid artery) bleeding (n = 1). These patients were not excluded from survival analysis. Continuous variables (e.g., tumor volumes, hemoglobin level, and age) were tested on linearity and time dependency. The final multivariable analysis was adjusted for age, because age demonstrated no linear association with LC and OS. For interactions, the variable with the most influence in the multivariable analysis was included in the final multivariable model. Besides this criterion, all variables were entered at the multivariable phase, regardless of the outcome of the univariate analysis. The forward stepwise selection procedure, in which nonsignificant variables from the univariate analysis are not reanalyzed in the multivariable analysis, may be inferior for maximizing prognostic accuracy,23 and was not preferred. The final multivariable model was generated by a backward elimination method to determine factors with influence on outcome. Because the T and N classifications are generally used as prognostic variables, they were included in all final multivariable analyses. The procedure PROC PHREG (SAS system for Windows, release 8.02; SAS Institute, Inc., Cary, NC) was used to perform the multivariable analyses.
Calculated coefficients from the final multivariable model were converted into a 0–100 scale. The maximum score (100) was based on the maximum coefficient. All coefficients for each prognostic group then were plotted relative to this maximum. A summation of each variable score resulted in an overall score (total points). The total points were finally converted into a probability of LC/OS. This allowed summation of the risks for any combination of prognostic variables.
Primary tumor volumes ranged from 6.4 to 393.0 cm3 (Table 2, Fig. 1). The median primary tumor volumes stratified by site were oral cavity, 39 cm3; oropharynx, 37 cm3; hypopharynx, 32 cm3; and supraglottic larynx, 19 cm3. Patients with clinical evidence of lymph node disease had lymph node tumor volumes ranging from 2.3 to 131.7 cm3. Lymph node tumor volume was associated with N classification. Patients with N3 disease (median, 75.3 cm3) were found to have larger lymph node tumor volumes than patients with N1 (median, 0.7 cm3) and N2 disease (median, 11.1 cm3). Total tumor volumes ranged from 6.9–393.0 cm3 (mean, 57.2 cm3; median, 42.5 cm3). The mean primary tumor volumes of patients who received unilateral and bilateral intraarterial infusion were similar and not significantly different (P = 0.94), being 40.5 cm3 and 39.8 cm3, respectively.
A complete local response after 6–8 weeks was achieved in 79 patients (92%) and a partial response was achieved in 7 patients (8%). After 2 years, 25 patients had local disease recurrence, including all partial responders. Six of these patients underwent salvage surgery with 3 incomplete resections and only 2 of these 6 patients survived > 7 months. Sixty-one patients achieved LC at the primary site. For all patients, the estimated LC rate at 5 years was 60%. Seven patients (8%) developed regional disease recurrence. Regional disease recurrence was defined as persistent disease after chemoradiation or as regional disease recurrence without local disease recurrence. Because of the low number of regional disease recurrences, analysis of prediction for regional control was unreliable and was not performed.
Multivariable analysis was used to determine the association between factors and LC. Besides the T and N classifications, primary tumor volume, age, tumor site, and unilateral/bilateral intraarterial infusion were included in the final multivariable analysis for LC. After this final analysis, two factors were found to have significant predictive value for LC: primary tumor volume (P = 0.01) and unilateral intraarterial infusion (P = 0.004) (Table 3). Using tumor volume as a continuous variable, an adjusted risk ratio of 1.026 was found, indicating that each 1-cm3 increase in volume was associated with a 2.6% decrease in probability of LC.
|Variables||P value||Hazard ratio||95% Hazard ratio|
|Site (oral cavity vs. other sites)||0.71||0.76||0.17||3.42|
|Site (other sites vs. oropharynx)||0.13||0.33||0.08||1.39|
|T classification (T3 vs. T4)||0.36||0.58||0.18||1.87|
|N classification (N0–1 vs. N2–3)||0.07||0.42||0.16||1.07|
|Primary tumor volume||0.01||1.03||1.01||1.05|
|Infusion mode (unilateral vs. bilateral)||0.004||5.05||1.69||15.08|
A nomogram (Fig. 2A) was constructed based on factors used in the final multivariable analysis. The nomogram provides a total score that can be translated to a probability of LC (Fig. 2B). Patients with a score of 60 will have the worst prognosis. An example is shown in Figure 2.
After a median follow-up of 35 months (range, 25–78 months), 37 patients remained alive. Thirty-two patients died of tumor-related causes, 8 patients of a second primary tumor, 7 patients of other causes (cerebro-vascular accident [CVA; n = 2], pneumonia [n = 3], lung embolus [n = 1], myocardial infarction [n = 1]), and 1 died of an unknown cause. This patient died 18 months after the initiation of the treatment and was physically examined 6 weeks before death in the outpatient clinic and had no evidence of recurrent disease. An autopsy was not performed. Only one patient was lost to follow-up. The estimated 5-year OS rate for the entire study group, including the 6 patients who died during treatment, was 38%.
In addition to the T and N classifications, primary tumor volume, gender, age, pretreatment weight loss, neck lymph node level involvement, and comorbidity were included in the final multivariable analysis for OS. The analysis demonstrated that primary tumor volume (P = 0.003), neck lymph node level involvement (P = 0.007), comorbidity (P = 0.01), and pretreatment weight loss (P = 0.02) were independent prognostic factors for OS (Table 4). Based on variables in the final multivariable analysis, a nomogram for predicting the probability of the 2-year OS was constructed (Fig. 3). The nomogram visualizes the importance of all prognostic factors for OS. In the nomogram, age < 40 years was not shown, because only 4 patients were < 40 years and all had a bad outcome and most likely do not provide a good representation of this subgroup. However, to exclude biased data, we did include these four patients in the multivariable analysis.
|Variables||P value||Hazard ratio||95% Hazard ratio|
|N classification (N0–1 vs. N2–3)||0.66||1.16||0.61||2.21|
|T classification (T3 vs. T4)||0.40||0.72||0.34||1.55|
|Gender (male vs. female)||0.07||0.49||0.23||1.05|
|Weight loss (< 10% vs. ≥ 10%)||0.02||2.04||1.10||3.78|
|Comorbidity (ASA 1 vs. ASA 2–3)||0.01||2.47||1.21||5.04|
|Neck level (no, Level II–III vs. IV)||0.007||3.45||1.40||8.48|
|Primary tumor volume||0.003||1.01||1.004||1.017|
Our estimated 5-year LC and OS rates were 60% and 38%, respectively, which are in agreement with the LC and OS data reported by Robbins et al.19 In developing a prognostic model, we were able to confirm several previously described prognostic factors such as primary tumor volume24 and comorbidity.12 In addition, we found a predictive factor (unilateral vs. bilateral intraarterial infusion of chemotherapy) for LC, which has not been described before.
Primary tumor volume emerged as an independent significant factor for predicting LC/OS. This confirms the earlier published results of patients treated with radiotherapy alone or with chemoradiation.13, 24–36 An overview of literature (Table 5) demonstrates a variety of mean tumor volumes and site-dependent cutoff volumes. Patients with comparable primary tumor volumes at different sites appear to have different LC probabilities. For a certain volume of a laryngeal tumor, the probability of LC is lower than for a hypopharyngeal tumor with the same volume.13 These differences were also found when we compared hypopharyngeal and oropharyngeal carcinomas. Nathu et al.26 presented a 5-year LC rate of 86% in a group of 35 patients with oropharyngeal carcinoma with a mean tumor volume of 14.8 cm3, whereas Hermans et al.27 presented a 5-year LC rate of 75% in a group of 119 patients with laryngeal carcinoma with a mean tumor volume of 2.3 cm3. Some authors used cutoff volumes to separate patients into favorable and unfavorable groups. Cutoff volumes have limited value for clinical use and linear correlations as found in glottic and supraglottic laryngeal carcinoma27, 28, 32 appear to be more practical for individual use. We demonstrated a near linear correlation in the current study using volume as a continuous variable in the multivariable analysis. For example, a 1-cm3 increase in tumor volume resulted in a 2.6% decrease in LC (P = 0.01). This finding enables us to implement volume measurement as a tool for clinical decision-making.
|Authors||RT/CCRT||Site||No.||Volume (mean, cm3)||Local control (5-yr)||Comment|
|Hermans et al.25||RT||Oropharynx||9||3 (T1)||80b||Only tonsilcarcinoma|
|Nathu et al.26||RT||Oropharynx||49||7 (T2)||92|
|Mendenhall et al.13||RT||Oropharynx (ts)||69||18||86|
|Hermans et al.27||RT||Glottic larynx||119||2.3||75|
|Hermans et al.28||RT||Supraglottic larynx||103||10.9||62|
|Castelijns et al.29||RT||Larynx||80||2.93||62||Supraglottic (n = 21), glottic (n = 67), subglottic (n = 2)|
|Mendenhall et al.30||RT||Glottic larynx||37||≤ 3.5c||87|
|Glottic larynx||37||> 3.5c||29|
|Freeman et al.31||RT||Supraglottic larynx||31||< 6c||83|
|Supraglottic larynx||31||≥ 6c||46|
|Mancuso et al.32||RT||Supraglottic larynx||63||< 6c||89|
|Supraglottic larynx||63||≥ 6c||52|
|Pameijer et al.33||RT||Glottic larynx||42||< 3.5c||85||Only T3|
|Glottic larynx||42||≥ 3.5c||25|
|Pameijer et al.34||RT||Hypopharynx||19||< 6.5c||89b|
|Doweck et al.24||CCRT||Oropharynx||23||18||83|
The effectiveness of concomitant chemoradiotherapy is also suggested by a comparison of tumor volumes between treatment modalities (Table 5). Hermans et al.25 presented a 5-year LC rate of 47% for patients with oropharyngeal carcinoma with a mean tumor volume of 15 cm3 treated with radiotherapy alone, whereas Doweck et al.24 demonstrated a 5-year LC rate of 83% for patients with oropharyngeal carcinoma with a mean tumor volume of 18 cm3 treated with chemoradiotherapy.
Patients with tumor extensions across the midline were treated with bilateral intraarterial infusions of cisplatin. However, bilateral intraarterial infusions resulted in significantly more disease recurrences than unilateral intraarterial infusions (P = 0.004). This is not explained by differences in tumor volumes between groups. A possible explanation is that patients with tumor extensions across the midline have a better response to unilateral intraarterial infusions because of the neovasculature of the tumor than to bilateral intraarterial infusions that divide halved dosages over both sides. Future validation of this finding is needed with emphasis on arteriographic studies and intratumoral distribution of cisplatin.
After multivariable analysis, other clinical factors were found to have no influence on LC. Even the often-found predictor, pretreatment hemoglobin level,9–11 did not emerge as a significant factor for LC after concomitant chemoradiotherapy. Many of the patients in the current study received blood transfusions during treatment, which might explain this outcome.
Primary tumor volume, lowest neck level involvement, pretreatment weight loss, and comorbidity were identified as independent prognostic factors for OS. The inverse relation between comorbidity and survival already has been established in a number of studies.12, 37–39 These studies demonstrated significant correlations between comorbidity and survival in young12 patients and in patients with advanced37 laryngeal carcinoma undergoing radiation therapy. The data from the current study confirm that comorbidity influences OS in patients receiving chemoradiation and that the pretreatment assessment of comorbid conditions is a prerequisite. To our knowledge, nutritional status has not been described often in prognostic studies. It has been reported that preoperative weight loss is a prognostic factor for worse survival40 in male patients. The current data demonstrate that pretreatment weight loss also has prognostic value in patients receiving chemoradiation. We could not confirm the earlier described association between lower neck level involvement and distant metastasis41, 42 because in only 2 of 9 (22%) patients with lower neck level (Level IV) involvement were distant metastases detected, compared with 17 of 83 (20%) patients without lower neck level involvement.
Using all factors from the multivariable analysis, we constructed a nomogram. The strength of a nomogram is the ability to illustrate the wide range of outcomes in a heterogeneous group of patients with head and neck carcinoma. The nomograms for LC and OS may serve as a basis for the more appropriate selection of patients who will benefit most from targeted chemoradiation. It is also helpful in the selection of patients with a low probability of LC who may then become eligible for more aggressive or alternative treatment schedules. However, this nomogram should preferably be validated in an independent patient cohort before it can be used in clinical practice.
In the current study, we found primary tumor volume to be the most important independent prognostic factor for patients with advanced, unresectable head and neck carcinoma treated with targeted chemoradiation. With the use of clinical algorithms, the pretreatment selection of patients with advanced-stage head and neck carcinoma can be improved.
The authors thank F.J.M. Hilgers, M.D., Ph.D., for his critical review of the article and R. Kroger, M.D., for performing excellent interventional radiology.
- 17SobinLH, WittekindCH, editors. TNM classification of malignant tumours, 5th edition. New York: John Wiley & Sons, Inc., 1997.
- 22Regression models and life-tables. J R Stat Soc. 1972; 34: 187–220..