Fax: (011) 52 55 5606-6947
Multivariate prediction of the probability of recurrence in patients with carcinoma of the parotid gland
Article first published online: 4 APR 2007
Copyright © 2007 American Cancer Society
Volume 109, Issue 10, pages 2043–2051, 15 May 2007
How to Cite
Carrillo, J. F., Vázquez, R., Ramírez-Ortega, M. C., Cano, A., Ochoa-Carrillo, F. J. and Oñate-Ocaña, L. F. (2007), Multivariate prediction of the probability of recurrence in patients with carcinoma of the parotid gland. Cancer, 109: 2043–2051. doi: 10.1002/cncr.22647
- Issue published online: 25 APR 2007
- Article first published online: 4 APR 2007
- Manuscript Accepted: 29 JAN 2007
- Manuscript Revised: 28 DEC 2006
- Manuscript Received: 23 OCT 2006
- parotid gland carcinoma;
- prognostic factors;
Parotid gland carcinoma is an infrequent tumor, and series that report on these neoplasms are relatively scarce in the literature. The objective of the current study was to identify prognostic factors in patients with parotid gland carcinoma and to develop a method for defining the probability of recurrence.
Patients with parotid gland carcinoma who were treated at the authors' institution from January 1981 through December 2004 and who completed treatment constituted the study group. Disease-free survival was calculated by using the Kaplan-Meier method. Logistic regression analysis was employed to define the recurrence-associated prognostic factors.
One hundred twenty-seven patients were included (64 men and 63 women); their mean age was 53 years. Mucoepidermoid carcinoma was diagnosed in 34.6% of patients, adenoid cystic was diagnosed in 15.7% of patients, adenocarcinoma was diagnosed in 14.3% of patients, and acinic cell carcinoma was diagnosed in 9.4% of patients. The median disease-free survival was 8.3 years (95% confidence interval [95% CI], 4.3–12.2 years). Logistic regression analysis confirmed tumor classification, facial nerve palsy, grade of tumor differentiation, patient age, and surgical margins as recurrence-associated factors (P < .00001). Using this model, 3 postoperative risk groups were defined—high-risk, intermediate-risk, and low-risk—that had recurrence frequencies of 71.4%, 43.1%, and 8.8%, respectively (P = .0001). The 5-year disease-free survival rates for these groups were 18.7%, 53.9%, and 99.9%, respectively (P = .00001).
In this study, the authors identified several significant prognostic factors. Consequently, they have proposed a prognostic score categorization that allows for a straightforward calculation of the risk of recurrence for a given patient that may help to define therapeutic strategies, target patient counseling, and design future trials. Cancer 2007. © 2007 American Cancer Society.
Parotid gland carcinoma represents 2% of head and neck malignancies,1 and its annual incidence is 0.6 per 100,000 population.2 Consequently, the biologic behavior of this neoplasm has been difficult to evaluate because of the scarcity of case presentations: Few institutions are able to report the necessary number of patients to define significant prognostic factors.3–5 Main drawbacks of the existing series to date include the frequent mixture of primary sites of origin of salivary gland malignancies and the small number of patients reported, rendering it difficult to obtain analyses from the data presented. In addition, a significant follow-up, preferably conducted at a single institution, is necessary to draw significant conclusions.
Some prognostic factors have been addressed and identified as most likely associated with recurrence and survival; nonetheless, to our knowledge, no prospective studies exist to date because of the low incidence of new cases and the nonexistence of conclusive data regarding the validity of the majority of prognostic factors described. Until now, certain studies have been performed with an objective based on significant prognostic factors for ascribing a prognostic score;5 this aids in making individual patient management uniform. However, validation of these scores is difficult6 because of the complex calculations required and the lack of significance of intrinsic tumor characteristics, such as histology and grade, previously reported as determinant for prognosis in patients with parotid gland carcinoma.7
Based on these concepts, and because our hospital is a tertiary referral cancer center with the ability to administer complete multimodality treatment,7 we performed a retrospective analyses of all patients with parotid gland carcinoma who received definitive treatment. The objective of this study was to identify recurrence-associated prognostic factors8 and to develop a model that could define subgroups of patients with differential risks, with the intention of identifying patients who are candidates for intensive multimodal therapy, thus affording a better design for future prospective and/or multicenter studies.
MATERIALS AND METHODS
Patients with a surgical-pathology diagnosis of parotid gland carcinoma who were admitted to the Head and Neck Department of the Instituto Nacional de Cancerología in Mexico City from January 1982 through December 2004 were included in the retrospective database for this study. Patients who were included were required to have a full clinical evaluation, must have received definitive treatment, and had to have a minimum 18-month follow-up. All patient evaluations included clinical history, physical examination, blood cell count and blood chemistry studies, parotid gland ultrasound, and a computed tomography (CT) scan included in their evaluation; during the last 6 years, magnetic resonance imaging (MRI) also was performed. A retrospective search of the patients' charts was performed, and the following variables were obtained: age, sex, tumor size, histology, grade, clinical tumor (T) and lymph node (N) classifications, tumor mobility, facial nerve palsy (FP), skin invasion, vascular invasion, treatment modality, surgical procedure, surgical margin status obtained, and radiotherapy modality.
The 2002 version of the Tumor-Lymph Node-Metastases staging system of the American Joint Committee on Cancer was used for clinical staging.9 Two independent pathologists reviewed all surgical-pathology materials from each patient prospectively and confirmed the diagnosis of carcinoma of the parotid gland, defined the histopathology type, and carried out a classification of the neoplasms using the World Health Organization classification (version 2005).10 Accordingly and with regard to the grade of tumor differentiation, lesions were categorized as well differentiated, moderately differentiated, poorly differentiated, or undifferentiated.
The standard surgical procedure was total parotidectomy with facial nerve preservation unless there was gross neural invasion. Patients who had a tumor that measured ≤2 cm and who, in transoperative studies or previous fine-needle aspiration biopsy (FNAB), had a low-grade tumors underwent superficial parotidectomy and had their surgical marginsevaluated for adjuvant therapy with radiation. Patients who were in poor clinical condition or who had unresectable tumors received radiotherapy as primary treatment.
Radiation therapy was administered with γ photons that were produced by a telecobalt unit or with high-energy x-rays from a 6-megavolt linear accelerator unit. For most patients, a simple, ipsilateral portal was used that encompassed the primary tumor (with a 2-cm safety margin) and ipsilateral neck lymph nodes. The majority of patients were treated in a classic manner with 5 courses of 2 grays (Gy) weekly for a period of 5 to 7 weeks. The technique of direct fields with electrons was used for overdosage.
Associations of independent variables with recurrence events were analyzed with the Student t test and the chi-square test for continuous and categorical variables, respectively. Disease-free survival (DFS) was calculated using the Kaplan-Meier method, and differences were analyzed with the log-rank test.
Prognostic factors associated with significant impact on recurrence or DFS with a probability value ≤.2 were included in the multivariate analyses. Multivariate analysis of recurrence event-associated factors was performed using logistic regression analysis. Odds ratios (ORs) were obtained as a measurement of association with 95% confidence intervals (95% CIs). The final model was selected according to the maximum log-likelihood criterion and was tested for interaction terms.
The predicted probability of recurrence for each patient was obtained by using the final logistic regression model with a specifically designed, Windows-based computer software program. Subsequently, categorization of the predicted recurrence probability was performed by tertiles, and the 3 groups were defined as low-risk, intermediate-risk, and high-risk groups.
Two-tailed statistics were taken into consideration in all patients. Version 10 of the SPSS software program for Windows was used to perform computations (SPSS, Inc., Chicago, Ill).
One hundred twenty-seven patients fulfilled the inclusion criteria and constituted the database for this study. Study participants included 64 women and 63 men (ratio, 1:1). Ages ranged from 15 years to 95 years (mean, 52.98 years; standard deviation, 20.8 years). The median greatest tumor dimension was 5.5 cm (range, 1–15 cm).
The main signs and symptoms at presentation were tumor mass in 127 patients (100%), facial nerve dysfunction in 38 patients (29.9%), facial pain in 9 patients (7.08%), trismus in 5 patients (3.93%), and an ulcerated tumor in 12 patients (9.44%). Histopathologic characteristics of the cohort are shown in Table 1.
|Characteristic||No. of patients||%|
|Adenoid cystic carcinoma||20||15.7|
|Adenocarcinoma, not otherwise specified||18||14.2|
|Acinic cell carcinoma||12||9.4|
|Squamous cell carcinoma||11||8.7|
|Carcinoma expleomorphic adenoma||6||4.7|
|Lymph node classification|
Eight patients (6.29%) underwent surgery alone as their main treatment, 84 patients (66.41%) underwent surgery and received postoperative radiation, 20 patients (15.74%) received radiation therapy followed by surgery, and 15 patients (11.81%) received radiation therapy as their only treatment. Fifteen patients (11.81%) underwent superficial parotidectomy, 97 patients (76.3%) underwent total parotidectomy, and 54 patients (42.5%) underwent total parotidectomy with neck dissection. Of 54 patients who underwent neck dissection, 38 patients (70.4%) had clinically positive lymph nodes. Sixteen patients (29.7%) underwent neck dissection on an elective basis.
Facial nerve preservation was accomplished in 78 patients (70%) who underwent parotidectomy at our institution. Surgical treatment complications included transient facial paresis in 45 patients (40.17%), wound infection in 10 patients (8.9%), salivary fistula in 3 patients (2.67%). and flap necrosis in 3 patients (2.7%). Radiotherapy complications comprised mucositis in 84 patients (80.76%), dermatitis in 10 patients (9.61%), and necrosis and ulceration in 9 patients (8.65%).
During follow-up, 53 patients (41.7%) experienced recurrences, including 14 patients (26.4%) with local recurrences, 14 patients with (26.4%) regional recurrences, and 26 patients (49%) with metastases. One patient had a simultaneous local recurrence and metastases. Metastases were observed in the lung in 20 patients (76.9%), in bone in 5 patients (19.23%), and in the liver in 1 patient (7.7%). Among patients who had local and regional recurrences, 22 patients (78.6%) underwent surgical salvage; and, at the time of this report, 6 of those patients (27.27%) remained alive and disease-free. Six patients received radiation therapy for salvage, and 2 of those patients (33%) had a complete response. The median overall survival and DFS for the cohort were 15.3 years (95% CI, 13.2–17.5 years) and 8.3 years (95% CI, 4.3–12.2 years), respectively. The associations of clinical and pathologic variables with DFS and recurrence are shown in Table 2.
|Factor||No.||Median DFS, y||95% CI||P*||Recurrence frequency (%)||P†|
|Young, ≤29||25||NR||5 (20)|
|Middle-aged, 30–55||41||2.0||0.73–3.28||26 (63.4)|
|Older, ≥56||61||8.26||—||22 (36.1)|
|Low, ≤3.9||54||3.88||2.01–5.76||24 (44.4)|
|High, ≥4||73||8.26||5.31–11.21||29 (39.7)|
|Adenoid cystic||20||4.3||—||.539||8 (40)||.864|
|Acinic cell||12||—||—||.167||3 (25)||.217|
|Squamous cell||11||8.26||—||.78||4 (36.4)||.706|
|Well differentiated||22||NR||3 (13.6)|
|Moderately/poorly differentiated||94||4.53||1.17–7.88||41 (43.6)|
|T1 or T2||33||NR||6 (18.2)|
|Lymph node classification||.02||.128|
|N1, N2, or N3||50||3.46||—||25 (50)|
|Facial nerve dysfunction||.0018||.043|
|Tumor size, cm||.0075||.013|
The mean radiotherapy dose was 61.7 ± 9.1 Gy. The radiotherapy dose was not associated with the recurrence event, even when the analysis was adjusted for preoperative or postoperative modality (data not shown).
DFS curves by T classification, degree of tumor differentiation, surgical margin status, and histopathology are depicted in Figure 1. In Figure 2, DFS curves from patients with negative lymph nodes who underwent neck dissection and did not undergo neck dissection are shown.
Table 3 displays the final logistic regression model (P = .00001). Age, T classification (including FP), surgical margin status, histopathology, and degree of tumor differentiation were the most significant recurrent disease-related prognostic factors.
|Factor||β||SE||P||Exp (β)||95% CI|
|T1 or T2†||—||—||—||1||—|
|T4 with facial palsy||1.3||0.699||.063||3.67||0.93–14.4|
|T4 without facial palsy||2.264||0.836||.007||9.62||1.87–49.5|
|Moderately or poorly differentiated||1||0.741||.177||2.72||0.636–11.6|
Using this model, and depending on the predicted probability of recurrence for each patient, we defined 3 postsurgical risk groups: high-risk (estimated P values, ≥.57), intermediate-risk (estimated P values, .18–.56), and low-risk (estimated probability value, <.18), with observed recurrence frequencies of 71.4%, 43.1%, and 8.8%, respectively (chi-square test; P = .0001). The 5-year DFS rates for each of these groups were 18.7%, 53.9%, and 99.9%, respectively (log-rank test; P = .00001), as shown in Figure 3.
For the current study, we explored the prognostic factors related to nonmetastatic parotid gland carcinoma and associated with disease recurrence based on data that were obtained from a single tertiary cancer referral center. Recurrence event-associated factors were identified by using logistic regression analysis. A model was fitted that included age, grade, clinical T classification (including FP), and surgical margin status. We decided to calculate the expected probability of recurrent disease for each patient using this model; and, accordingly, we defined 3 risk groups. Finally, we tested the observed DFS for these 3 groups, and the results suggested that the performance of this method could be highly efficient in terms of prognostic discrimination.
An important difference regarding this case series compared with others is that a substantial number of patients presented with intermediate or advanced disease. Approximately 75% of our patients had T3 or T4 disease. In addition, our database included a considerable number of patients who had a long follow-up, who proceeded from a single institution with prospective histologic revision by 2 independent pathologists, and who had a homogeneous treatment design and consistent follow-up protocol. However, a major drawback of our study, in common with other previous reports,11 was its retrospective nature.
Our 10-year overall survival and DFS rates of 74% and 60% compare favorably with results reported in the literature,3 which range from 50% to 70% for 10-year overall survival and from 35% to 55% for 10-year DFS. Multiple reports have indicated prognostic significance for age.5, 7, 11 Our study results indicated that younger patients had a better prognosis than middle-aged and older patients in univariate analyses, an observation that was confirmed by logistic regression analysis. This significance was specifically strong when we compared young and middle-aged patients (OR, 5.35) with the lesser impact in older patients (OR, 1.05). Certain studies that have highlighted the importance of this factor included salivary gland malignancies in general.11, 12 However, our analyses have the advantage of referring exclusively to parotid gland carcinoma.
T classification was of utmost prognostic importance, as reported previously,5–7 in both univariate and logistic regression analyses. We also categorized tumor size into 4 groups, as shown in Figure 1D, and this factor remained significant in univariate analyses. It is noteworthy that there are studies, such as that by Terhaard et al.,12 that ascribe better predictive value to tumor size rather than to clinical T classification for distant metastases; nevertheless, T classification is reported consistently as a stronger prognostic factor for parotid gland carcinoma,13 especially because the most recent review of this classification9 uses the T3 category for major salivary gland carcinomas that measure ≥4 cm or that have extraparenchymal invasion into soft tissues, removing the upper 6-cm limit.
Tumor extension, represented by surgical margins and FP, has been reported in the literature as a significant factor for prognosis.5, 14 Surgical margins in our study group had significance in both univariate and multivariate analyses, reflecting the need to perform a wide excision of malignancies, especially in series such as ours in which intermediate and advanced-stage tumors comprised a significant number of cases. These findings most likely support the indication for total parotidectomy and for even more advanced surgical procedures (temporal bone resection, mandibulectomy, parapharyngeal space tumorectomy), especially in patients who have advanced-stage tumors in which surgical margins may be compromised. Moreover, the determination of this factor with a transoperative study and other preoperative factors, such as tumor grade and T classification, also may determine the need for neck dissection in patients with N0 necks, as mentioned (although not with sound evidence) previously.7 The importance of surgical margins as a significant prognostic postoperative factor is major, because it may define the need for adjuvant treatment, such as radiation therapy, as discussed below.
With regard to facial paralysis, our results indicated a better prognosis for patients with T4 tumors who had facial dysfunction, a finding that differs from what has been reported previously in several series on parotid gland carcinoma.3, 5 However, the significance of this factor was determined previously by comparing patients who had facial dysfunction in all disease stages, whereas, in our series, the results indicated a better prognosis for patients who had T4 tumors with facial nerve dysfunction. This point has been addressed in previous studies that, by including all disease stages in multivariate prognostic analyses, removed FP as a significant prognostic factor.14 In addition, facial paralysis in our group of patients with advanced tumors most likely signified an alarm signal that it was time to seek specialized medical care, as pointed out previously for patients who had salivary gland neoplasms with other histologies.15 It is worth noting that these findings may provide support for the recent T4-classification subdivision, in which T4 lesions are reorganized into T4a and T4b categories and in which FP indicates a resectable tumor along with other prognostic factors (skin, ear canal, mandible invasion), in contraposition to tumors that invade the infratemporal fossa or carotid artery.
Specific histopathology exerted no significant influence on prognosis for recurrence, as reported in previous studies,5 Nonetheless, Table 2 shows that the disease-free interval was shorter for patients with adenocarcinoma16 and undifferentiated neoplasms, but longer for patients with other neoplasms, such as mucoepidermoid and adenoid cystic cell carcinoma. We believe, as other authors have pointed out,7 that this may be associated with the rare presentation of these malignancies and the consequent small numbers of patients ascribed to each histopathology in the current report and in the reports of others.
Contrary to other published series5, 6 that attempted to propose a prognostic score, we were able to determine prognostic significance for the grade of tumor differentiation, and this significance was supported in the logistic regression analyses. Thus, patients who had low-grade tumors had better DFS and fewer recurrences. Certain studies12 have pointed out the possible effect of subjectivity on tumor grade. Nevertheless, 2 different pathologists evaluated each pathologic diagnoses in our series, and a final consensus was ascribed accordingly to each case.
Several series have postulated that there is a high frequency of lymph node disease in patients with salivary gland carcinomas,3, 17 and some have addressed this point by stating that neck dissection should be performed in all patients with parotid gland carcinoma.3 Despite this, and although the frequency of lymph node metastasis was 39% in our series, when we analyzed the impact on prognosis of performing an elective neck dissection, our results indicated that the DFS was shorter in patients who underwent elective neck dissection than in patients who went without the procedure. This may be explained by the retrospective nature of our analyses, in which there is a selection bias for administering more extensive treatment to patients with a greater risk of recurrence. Neck dissection most likely should be performed in those patients who have proven risk factors for locoregional failure, a point that should be addressed in future prospective and retrospective trials.14
Patients who received postoperative radiation therapy displayed a tendency toward shorter DFS, most likely reflecting the therapy-selection bias that exists in this retrospective series.14 However, there are studies18, 19 that confer a benefit to adjuvant radiation, especially in patients with advanced-stage disease and high tumor grade, factors that, in our study, were associated with decreased DFS.
Considering the factors that had significance in multivariate analyses, a logistic regression equation was used to define 3 groups of patients with high, intermediate, and low risk, and that has been reported in previous trials on gastric carcinoma.20 Applying logistic regression analyses, a software program has been developed to include the individuality of each patient by inserting the corresponding values and in which the output is the risk of recurrence for each patient.
In this way, we support surgical therapy as primary treatment except in patients who have unresectable lesions or prohibitive comorbidities,21 in whom radiotherapy may be used upfront. Moreover, because the significant prognostic factors described above may be determined in the majority of patients prior to the initiation of treatment or during surgery (when FNAB is obtained), it is feasible to design preoperative and intraoperative strategies of treatment, as long as frozen histologic sections are obtained and surgical margin status is determined.22
Rational decisions regarding the choice of adjuvant or complementary treatment type and whether or not patients with N0 necks should undergo neck dissection may be made according to the patient's given risk of failure in future prospective studies. In addition, professional counseling may be administered based on significant recurrence factors.
The prediction model yields a conditional probability for each individual patient. This is a number between 0 and 1 that expresses the estimated likelihood of the recurrence outcome of parotid gland carcinoma based on the independent effect for each of the predictive variables defined in Table 3.
The regression coefficients (β) listed in Table 3 are entered and multiplied by the value of each variable, depending on its presence or absence. Regression coefficients are the same for all patients, whereas variable values are specific for each patient.
If a given patient has a T3 parotid gland carcinoma, then T3 = 1, T4 without FP = 0, T4 with FP = 0, and so on. If a patient has a T1 or T2 parotid gland carcinoma, then all of the terms regarding T classification disappear from the equation. If a given patient is age 53 years, then middle aged = 1, and old age = 0. If a patient has an undifferentiated carcinoma, then moderately or poorly differentiated = 0, and undifferentiated = 1. If a patient has a well-differentiated carcinoma, then all terms related to differentiation disappear from the equation. If a patient has positive surgical margins, then margin = 1. If a patient has negative surgical margins, then the term disappears from the equation.
An example of the calculations for 3 hypothetical patients is provided in Table 4. Calculations can be performed by using the logistic regression equation in the Excel program or with a regular scientific calculator. Otherwise, the corresponding software can be downloaded online (http://www.incan.edu.mx).
|Patient||Tumor classification||Surgical margin||Age, y||Tumor grade||Ŷ||Classification|
|Z||T4 with facial palsy||Positive||60||Undifferentiated||0.87||High|
Therefore, patients in the low-risk group and with good comorbidity factors are candidates for surgical treatment only with objective of obtaining negative surgical margins; the extent of surgical procedure is determined by this objective with the aid of transoperative frozen-section analyses and previous FNAB. Patients with intermediate risk and acceptable comorbidity are candidates for upfront surgical treatment with negative margins (total parotidectomy) and elective neck dissection (if the neck is negative) and, depending on the final score, postoperative radiation therapy. If patients in the high-risk group have tumors that are resectable technically, then they should receive combined treatment (the extent of surgery, even if it is radical, is determined by the need to obtain negative margins), as described above for patients in the intermediate-risk group, with the additional possibility of exploring for new adjuvant treatments (ie, chemotherapy and molecular target therapies),23 and they should receive special patient counseling with regard to the high probability of disease recurrence. Patients with unresectable disease and high comorbidity should be treated initially with radiation therapy and, if a partial response is obtained, then surgical treatment may be considered. If a total response ensues with radiation, then further adjuvant treatment with new agents or conservative follow-up may be offered. The aforementioned software may be employed to calculate a patient's given risk in an expeditious fashion. Consequently, the development of multicentric studies with reporting of cases for external validation is projected. Finally, this algorithm may represent the basic design of future prospective trials.
- 9AJCC Cancer Staging Manual. 6th ed. New York: Springer-Verlag; 2002., , , et al.
- 10BarnesL, EvesonJW, ReichartP, SidranskyD, eds. Pathology and genetics of head and neck tumors. World Health Organization Classification of Tumours. Lyon: IARC Press; 2005: 209–281., , , et al. In:
- 12Salivary gland carcinoma: independent prognostic factors for locoregional control, distant metastases and overall survival: results of the Dutch Head and Neck Oncology Cooperative Group. Head Neck. 2004; 26: 681–693., , , et al.