Improved local control for early T-stage nasopharyngeal carcinoma – a tale of two hospitals

Radiotherapy and Oncology 57 (2000) 155±166

www.elsevier.com/locate/radonline

Improved local control for early T-stage nasopharyngeal carcinoma ± a tale of two hospitals Peter Man Lung Teo a,*, Sing Fai Leung a, Jack Fowler b, To Wai Leung c, Yuk Tung c, Sai Ki O c, Wai Yee Lee a, Benny Zee d a

Department of Clinical Oncology, Prince of Wales Hospital, Shatin, Hong Kong, People's Republic of China b Department of Human Oncology, University of Wisconsin±Madison, Madison, WI, USA c Department of Clinical Oncology, Tuen Mun Hospital, Tuen Mun, Hong Kong, People's Republic of China d National Cancer Institute of Canada, Clinical Trials Group, Kingston, Ontario, Canada Received 28 March 2000; received in revised form 10 July 2000; accepted 19 July 2000

Abstract Purpose: To study the ef®cacy of intracavitary brachytherapy (ICT) in early T-stage nasopharyngeal carcinoma (NPC). Methods and materials: All early T-stage (T1 and T2 nasal cavity tumour) NPC treated with a curative intent up to 1996 were analyzed (n ˆ 743), 163 from the Prince of Wales Hospital (PWH) and 25 from Tuen Mun Hospital (TMH) were given ICT after radical external radiotherapy (ERT; group A). They were compared with 555 patients treated with ERT alone (group B). The radiotherapy techniques were identical between the two hospitals. The ERT delivered the tumoricidal dose (uncorrected biological equivalent dose (BED)-10, $75 Gy) to the primary tumour, and this did not differ in technique or dosage between the two groups. The ICT delivered a dose of 18±24 Gy in three fractions over 15 days to a point 1 cm perpendicular to the midpoint of the plane of the sources. Results: The local failure was signi®cantly less (crude rates, 6.9 vs. 13.0%; 5-year actuarial rates, 5.8 vs. 11.7%) and the disease-speci®c mortality was signi®cantly lower (crude rates, 13.8 vs. 18.9%; 5-year actuarial rates, 12.2 vs. 15.2%) in group A compared with group B. ICT was the only signi®cant independent prognostic factor predictive of fewer local failures. When ICT was excluded from the Cox regression model, the total physical dose or the total BED-10 uncorrected for tumour repopulation became signi®cant in predicting the ultimate local failure rate. The two groups were comparable in the rate of the chronic radiation complications. A signi®cant dose±tumour-control relationship existed, plotting the local failure as a function of the total physical dose or the total BED. Conclusions: Supplementing ERT, which delivered the tumoricidal dose (uncorrected BED-10, $75 Gy), with ICT signi®cantly enhanced ultimate local control in early T-stage (T1/T2 nasal in®ltration) NPC. A signi®cant dose±tumour-control relationship exists above the conventional tumoricidal dose level. q 2000 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Nasopharyngeal carcinoma; Intracavitary brachytherapy; Local control; Dose±response/dose±tumour-control relationship

1. Introduction Brachytherapy, either in the form of intracavitary intubation (ICT) or interstitial implantation, has been used to supplement the radiation dose to the nasopharynx after radical external beam radiotherapy (ERT) for nasopharyngeal carcinoma (NPC) [1,3,13,25,27,29,33]. The routine practice of brachytherapy after ERT is only justi®ed if there is a dose±tumour-control relationship of NPC above the conventional tumoricidal dose and the complications of the additional dose by brachytherapy are mild and do not outweigh its ability to enhance local control. However, hitherto, there has been no large scale, randomized study * Corresponding author

that has established the ef®cacy of dose escalation by brachytherapy after conventional dose ERT. Most studies are retrospective, comparing the recent results with brachytherapy supplementing ERT with the historical results of ERT alone [1,3,4,10,13,15,21,25,27,29,30,32,33]. Previously, we have reported our experience in ICT between 1984 and 1989 for the primary treatment of NPC [22]. We concluded that the ICT was effective in converting the vast majority of local persistences to complete remission and in enhancing their ultimate local control, provided that the initial tumour was of early T-stage (Ho's stage T1 and T2 with tumour extension only to the nasal cavity) which could be adequately treated by the ICT from a dosimetric viewpoint [22]. The adjuvant use of ICT for the complete responders to ERT was, however, not found to be bene®cial [22].

0167-8140/00/$ - see front matter q 2000 Elsevier Science Ireland Ltd. All rights reserved. PII: S 0167-814 0(00)00248-6

156

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

Since that report, the techniques and doses for both ERT and ICT have remained unchanged in our centre. After the commencement of service of the Tuen Mun Radiotherapy and Oncology Department (TMH) in 1990, the technique of ICT practised in the Prince of Wales Hospital (PWH) was adopted with only minor modi®cations [12]. In the PWH, ICT was used to treat local persistence after ERT and also as an adjuvant for the complete responders to ERT. In the Tuen Mun Hospital (TMH), however, ICT was only applied in cases of local persistence after ERT. In both hospitals, the decision to use ICT was made at the ®rst endoscopic examination, 4±6 weeks after ERT. We now present the results pooled from the two hospitals and evaluate the contribution of ICT to the local control of the early T-stage NPC. 2. Methods and materials All non-disseminated (M0) NPC patients with early Tstages (T1 and T2 nasal cavity tumours; n ˆ 509) [7] presenting to the Clinical Oncology Department of the PWH (1984±1996) and TMH (1990±1996) were included in the analysis. More advanced T-stages were not studied because they could not be adequately treated by ICT from the dosimetric viewpoint. Of the early T-stages, 188 (163 PWH, 25 TMH) were treated by ICT in addition to ERT (group A) and 555 (346 PWH, 209 TMH) were treated by ERT alone (group B). All patients treated by ERT alone had a complete primary tumour response diagnosed endoscopically with or without T-site biopsy 4±6 weeks after ERT. On the other hand, ICT was used to treat local persistence after ERT (indication-1; 101 PWH and 25 TMH) or as an adjuvant treatment for complete responders to ERT, at the discretion of an individual radiation oncologist (indication-2; n ˆ 62, all from PWH). For indication-1 within group A, the local persistence was diagnosed 4±6 weeks after ERT with the following criteria: (a) histological proof of residual tumour; and/or (b), endoscopic ®ndings highly suggestive of residual tumour which included overt exophytic tumour or tumour-ulcer in the same location of the pre-treatment cancer. For patients with indication-2 within group A and all the patients in group B, a complete response to ERT was also diagnosed at 4±6 weeks after ERT by nasopharyngoscopic ®ndings compatible with post-irradiation changes (mucosal congestion, minimal contact bleeding, and absence of mucosal/submucosal nodularity/swelling/ulceration/or frank tumour), supported by a negative biopsy. If the nasopharyngoscopic ®ndings were equivocal for local persistence, but the biopsy was negative for tumour, the patient was also considered as a complete responder to ERT, and the ICT, if given, was considered as an adjuvant treatment, thus belonging to indication-2. Ultimate local failure was de®ned as histological tumour persistence for more than 3 months after ERT, or recurrence proven by biopsy after an initial histologically complete remission. All patients had ERT according to the Ho's radiation meth-

ods [7]. For patients without cervical nodes and whose nasal cavity tumours (if any) were not extending too anteriorly, the primary tumour was treated throughout by a 3-photon-®eld (Clinac 6/100, Varian Associates, Inc, Palo Alto, CA) arrangement with an additional anterior photon ®eld irradiating the neck electively. The majority of these patients received 60 Gy in 24 fractions over 6 weeks. For patients with cervical nodes, the primary tumour was treated en bloc with the upper cervical lymphatics by two lateral opposing facio-cervical photon ®elds to spinal cord radiation tolerance (40 Gy/16±20 fractions/4 weeks) before the radiation plan was changed to that of a 3-photon-®eld arrangement for the delivery of the full tumoricidal dose. The median and mean total ERT doses were 60.0 and 61.2 Gy, respectively (range, 60.0±71.2 Gy), delivered by daily fractions (®ve fractions/week) of 2.0±2.5 Gy each. Common to both hospitals, ICT (Fig. 1a,b) was performed using the same in-house nasopharyngeal applicator which was identical in the two hospitals. In both PWH and TMH, the radiation dose was delivered by a high dose rate (HDR) afterloading machine (Buchler Remote Afterloading, Buchler GmbH, Braunschweig, Germany with Ir192 source (1984±1993) and Microselectron, Nucletron B.V. Veenendaal, The Netherlands) to a reference point at 1 cm above (and below) the midpoint of the plane of sources [19,21,22]. In addition, TMH used a reconstruction box for 3D image reconstruction for dose calculation and prescription when applying the nasopharyngeal applicators invented in PWH for the HDR ICT for NPC [12]. Nevertheless, the similarity between ERT and ICT techniques and dose prescription and the common selection of early T-stage NPC for the ICT justi®ed the pooling of data from the two hospitals for the present analysis. In general, the nasopharyngeal applicators were inserted under local anaesthesia, one on each side, via the middle or inferior meatus to their treatment positions, abutting on the postero-superior wall of the nasopharynx. They were then immobilized externally by a head-frame. The separation between the central axis of the applicators ranged between 1.1 and 1.6 cm. The principal fractionation schemes for the ICT were: (a), 24 Gy/three fractions/15 days (n ˆ 118); and (b), 18 Gy/three fractions/15 days (n ˆ 62). The former was mainly used to treat local persistence after ERT, whereas the latter was mainly used as an adjuvant treatment for the complete responders to ERT. In addition, there were 11 patients, all treated in PWH, whose ICT dose ranged between 8 and 46 Gy, delivered in 1±7 fractions within a period of 1±43 days. Of these 11 patients, ten were given ICT for local persistence and one was given ICT as an adjuvant treatment. The biological equivalent doses (BED-10 and BED-3) were calculated separately for ERT and ICT for each patient from the dose/fraction and the total number of fractions and an a/b ratio of 10 and 3 Gy, respectively, for the early and the late responding tissues [2,6,14,26]. The total uncorrected BED was then calculated by the summation of the BEDs of the ERT and ICT, without corrections for the

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

157

Fig. 1. (a) Lateral skull radiograph showing the ICT applicators in situ with radio-opaque dummies. (b) Isodose distribution of the ICT superimposed on an axial CT-view through the level of the nasopharynx.

158

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

possibility of tumour repopulation during the time course of the radiotherapy (ERT ^ ICT). Finally, the corrected BED10 was calculated by adding together the BED-10 of the ERT and that of the ICT, while correcting for the possibility of tumour repopulation by employing a correction term of 0.5 Gy/day [2,6,14,26] throughout the whole period of radiotherapy (i.e. from the start of ERT to the end of ICT, including the time interval between, with Ir-192 source (1994±1996) ERT and ICT). The radiation dose was considered tumoricidal if the total uncorrected BED-10 was 75 Gy or above. Group A (ERT and ICT) was compared with group B (ERT alone) for T-stage [7,8], N-stage [7,8] and overall stage [7,8], and patient sex and age. Individual nodal parameters, including ®xation to surrounding organs, maximal diameter, laterality (homolateral to the side of the bulk of the primary tumour vs. contralateral vs. bilateral) and number (single vs. multiple) were also compared between the two groups. The rates of mortality from NPC, distant metastasis and ultimate local failure (histological persistence for 3 months or more after ERT and biopsy-proven recurrences) were compared between the two groups. The median times to death from NPC, to development of distant metastasis and development of local failure were also compared. The pre-treatment CT scan was reviewed for all patients to con®rm the T-stage. In addition, all patients who failed locally were reviewed for adequacy of the primary ERT. The possibility of `geographical miss' was explored by studying the pre-treatment CT scan, the radiation plan of the primary ERT and the site of the local failure with respect to the initial tumour bulk and the region of underdosage (if any) of the primary ERT. The actuarial rates of overall survival (OSR), relapse-free survival (RFS), disease-free survival (DFS), freedom from distant metastases (FDM) and freedom from local failure (FLF) were calculated using the Kaplan±Meier method and comparisons between groups A and B were made using the log-rank test. Logistic regression was then performed to identify any confounding factors from the group of potential prognostic parameters [16]. Statistical signi®cance was conventionally taken at a P value of 0.05 or less. Analyses were done separately for the PWH and TMH data before the ®nal analysis for the pooled data from both hospitals was performed in a multivariate fashion using Cox regression [5] for each of the clinical endpoints (OSR, RFS, DFS, FDM and FLF), evaluating the potentially significant prognostic parameters, which included the patient's age and sex, the T-stage [7,8], the N-Stage [7,8], the tumour response status to ERT as assessed by endoscope 4±6 weeks after ERT (complete response vs. others), the administration or not of ICT (group A vs. group B), the total physical radiation dose (ERT dose plus ICT dose, if any), and the administration or not of neoadjuvant chemotherapy. Cox regression was repeated with adjustment for the signi®cant confounding variables obtained by the logistic regression with ICT as the dependent variable. Cox regression was

then repeated with the individual nodal parameters substituting the N-stage, and also with the total BED-10 (corrected or uncorrected for tumour repopulation) substituting the total physical radiation dose. The multivariate analysis was repeated with exclusion of the administration or not of ICT from the Cox regression model. Finally, Cox regression was performed separately for groups A and B to evaluate signi®cant prognostic factors within the two groups. The dose±tumour-control relationship of NPC was studied by plotting the rate of local failure (failure/patient) against the total physical dose (ERT dose plus ICT dose), the total uncorrected BED-10 (BED-10 of ERT plus BED10 of ICT), and the total BED-10 corrected for tumour repopulation during the whole period of radiotherapy from the start of ERT to the last fraction of ICT [2,6,14,26]. Since the ICT dose to the temporo-mandibular joints, parotid, submandibular glands and the auditory apparatus was minimal [19,21], the corresponding radiation complications of trismus, xerostomia, otitis media and hearing loss were not compared between groups A and B. The two groups, however, were compared for the rate of epistaxis, blood-stained nasal discharge, headache, nasopharynxulceration/necrosis, nasal±oral foul smell, cranial nerve(s) palsy, pituitary±endocrine dysfunction and temporal lobe radiation encephalopathy (clinical and radiological). The actuarial cumulative chronic radiation complication rate (all complications/patient) were plotted for groups A and B as a function of time by the Kaplan±Meier method, and a comparison between the 2 groups was made using the logrank test. The rates of all chronic radiation complications (/ patient) were studied by logistical regression as a function of BED-3 [2,6]. The rates of administration of salvage treatments for local failures (nasopharyngectomy or re-irradiation with ERT with/without ICT boost) were compared between the two groups. The outcome of the salvage treatments, in terms of success of local tumour control and the occurrence of severe late treatment complications (trismus with dental gap of #1.5 cm, deafness, severe otitis media, blindness, prolonged hospital-stay of $30 days, malnutrition with wasting and treatment-related mortality) was also compared. 3. Results The sex ratio, patient age (,40 vs. $40 years) and fractionation schemes of ERT were comparable between groups A and B, but there were signi®cantly more T2 (with nasal cavity in®ltration) in group A, and more advanced N-stages and overall stages in group B (Table 1). Neoadjuvant chemotherapy was also more frequently given to patients in group B, as a result of its more advanced N-stages. Compared with group A, group B had signi®cantly more local failures as ®rst failure and disease-speci®c mortality (Table 2). The two groups were comparable in the rates of distant metastasis and regional

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

159

Table 1 Comparison of the patient and tumour characteristics and the ERT BED-10 and use of neoadjuvant chemotherapy between groups A and B a

Age Median ,40 years $40 years Sex Male Female Ho's stage T1 T2 N0 N1 N2 N3 I II III IV UICC stage (1997) T1 T2 N0 N1 N2 N3 I IIa IIb III IV Node ®xation Fixed Mobile Node size (cm) Median Range Node laterality Homolateral b Contra/bilateral Node number Single Multiple ERT BED-10 Median Range Neoadjuvant chemotherapy Not given Given a b

Group A (n ˆ 188)

Group B (n ˆ 555)

Statistical test

P value

42.5 (24±74) 78 (41.5) 110 (58.5)

45 (16±81) 173 (31.2) 382 (68.8)

t-Test Chi-square

0.0029 0.01

135 (71.8) 53 (28.2)

377 (67.9) 178 (32.1)

Chi-square

0.32

86 (45.7) 102 (54.3) 99 (52.7) 41 (21.8) 37 (19.7) 11 (5.8) 41 (21.8) 99 (52.7) 37 (19.7) 11 (5.8)

375 (67.6) 180 (32.4) 203 (36.6) 128 (23.1) 158 (28.4) 66 (11.9) 130 (23.4) 201 (36.2) 158 (28.5) 66 (11.9)

Chi-square

0.001

Chi-square

0.001

Chi-square

0.001

86 (45.7) 102 (54.3) 99 (52.7) 65 (34.6) 19 (10.1) 5 (2.7) 42 (22.3) 57 (30.3) 65 (34.6) 19 (10.1) 5 (2.7)

375 (67.6) 180 (32.4) 203 (36.6) 218 (39.3) 93 (16.7) 41 (7.4) 129 (23.2) 74 (13.3) 218 (39.3) 93 (16.8) 41 (7.4)

Chi-square

0.001

Chi-square

0.001

Chi-square

0.001

8 (4.3%) 81 (43.1)

40 (7.2%) 312 (56.2)

Chi-square

0.52

3 0.5±6.5

3 0.5±10

Wilcoxon

0.104

65 (34.6%) 24 (12.8)

221 (39.8%) 131 (23.6)

Chi-square

0.07

45 (23.9) 44 (23.4)

216 (38.9) 136 (24.5)

Chi-square

0.064

75 75.0±84.4

76.125 75.0±84.4

Wilcoxon

0.009

176 (93.6) 12 (6.4)

473 (85.2) 82 (14.8)

Chi-square

0.003

Figures in parentheses represent ranges or percentages. Cervical nodes homolateral to bulk of the tumour in the nasopharynx.

failure as ®rst failure (Table 2). The median duration of follow-up, rate of patients defaulting follow-up and the follow-up duration of the patients prior to their defaulting follow-up were comparable between groups A and B. The application of ICT was associated with signi®cantly fewer ultimate local failures for PWH patients (Fig. 2a), TMH patients (Fig. 2b), and all patients from both hospitals (Fig. 2c). The 5-year actuarial free of local failure rate was

signi®cantly higher in group A (with ICT) than in group B (without ICT), 94.2 vs. 88.3%, respectively; P ˆ 0:013. The 5-year actuarial disease-speci®c mortality was lower in group A than in group B, 12.2 vs. 15.2%, respectively. In addition, the crude ultimate local failure rates of T1 tumours (4.65 vs. 12%, respectively; P ˆ 0:046) and the ultimate failure rates of T2 tumours (8.8 vs. 15%, respectively; P ˆ 0:135) were also lower in group A than in group B.

160

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

Table 2 Comparison of disease-speci®c mortality, ®rst failures, and follow-up between groups A and B Crude rate

Disease-speci®c mortality 5-year survival 95% con®dence interval Local failure as the ®rst failure 5-year actuarial free from local failure rate 95% con®dence interval Distant metastasis as the ®rst failure 5-year actuarial free from distant metastasis rate 95% con®dence interval Regional failure as the ®rst failure 5-year actuarial free from regional failure rate 95% con®dence interval Follow-up duration Median (months) Range (months)

Group A (n ˆ 188)

Group B (n ˆ 555)

Frequency

%

Frequency

%

26/188

13.83 87.8 82.9±92.7 6.91 94.2 90.7±97.7 13.30 87.4 82.4±92.4 9.04 91.8 87.5±96.1

105/555

18.92 84.8 81.7±87.9 12.97 88.3 85.4±91.2 12.79 87 84.1±90.0 7.57 91.9 89.4±94.4

13/188 25/188 17/188

90.9 10.8±152

The two indications for ICT (indication-1 vs. indication-2) did not differ signi®cantly in the ultimate local failure or survival rates. The 5-year-FLFs of patients given ICT for local persistence (indication-1), patients given adjuvant ICT (indication-2), and patients given ERT only (group B) were 94.3 (95% CI, 90.2±98.4%), 94.5 (95% CI, 88.3±100%) and 88.3 (95% CI, 85.4±91.2%), respectively (Fig. 3). Multivariate analysis showed that the administration of ICT was the only signi®cant factor predictive of fewer local failures (Cox regression, P ˆ 0:0133; risk ratio, 0.48 (95% CI, 0.27±0.87)). In addition, logistic regression analysis was performed with ICT as the dependent variable against the other potential prognostic parameters to look for any confounding factors in the same group. The T-Ho, N-Ho, sex, age and the tumour response status to ERT were found to be signi®cant in the logistic model. However, when all these factors were ®tted into the Cox model together with ICT, the parameter estimate of ICT was changed from its original value by only 5%, suggesting that the potential confounding factors had no substantial prognostic impact and that the ICT was still the only signi®cant prognostic factor determining the ultimate local failure rate. In addition, since TMH gave ICT only for residual/persistent tumours, while PWH gave ICT as an adjuvant treatment for some patients, we performed a strati®ed log-rank test to cater for possible differences between the two hospitals. In the strati®ed log-rank test, the signi®cance of ICT on enhancing local control was maintained (P ˆ 0:0288). In the presence of ICT, the T-stage (T1 vs. T2), response status to ERT (complete response vs. others), and the total physical dose or biological dose were not prognostically signi®cant. However, on excluding the use of ICT from the Cox regression model, the total physical radiation dose became signi®cant in predicting the local failure rate (Cox regression, P ˆ 0:0131). Similarly, when the Cox regression was repeated with the total BED-10 uncorrected for tumour

72/555 71/555 42/555

78.8 5±170

P value (log-rank test)

0.0811 0.0128 0.986 0.72 0.183 (Wilcoxon test)

repopulation substituting the total physical dose, the total uncorrected BED-10 was also signi®cant in predicting the local failure rate (Cox regression, P ˆ 0:0094). However, when a correction term of 0.5 Gy/day [14] was used to cater for tumour repopulation during the time course of radiotherapy for the calculation of the total BED-10, the corrected BED-10 did not correlate signi®cantly with the ultimate local failure rate. Nevertheless, an apparent dose± tumour-control relationship (less failures with higher dose) existed when the local failure rate was plotted against the physical dose (Fig. 4a), the uncorrected BED-10 (Fig. 4b), and the corrected BED-10 (Fig. 4c). Group A patients experienced signi®cantly higher rates of chronic nasopharyngeal radiation-ulceration (5.3%; ten patients) than group B patients (0.2%; one patient; Table 3). However, the rates of epistaxis and bloody nasal discharge were comparable between the two groups. The epistaxis was always self-limiting and required no blood transfusion, hospitalization or surgical intervention. In group A, the radiation-ulceration/necrosis in the nasopharynx caused a foul-smelling crust/exudate in ®ve patients, headache in four patients and both symptoms in one patient. The patient in group B with nasopharyngeal ulceration/ necrosis had a foul smell only. There was no increase in temporal lobe radiation encephalopathy (clinical or radiological), pituitary±endocrine dysfunction, optic nerve/ chiasma radiation injury or radiation-induced cranial nerve(s) palsy (without evidence of local recurrence) in group A compared with group B (Table 3). The actuarial cumulative incidence rates of all the chronic radiation complications (/patient) were comparable between groups A and B (Fig. 5). Nine patients with local failures in group A (4.8%; two nasopharyngectomies, seven re-irradiations to $60 Gy) were subject to radical salvage treatments, compared with 34 patients with local failures in group B (6.13%; eight

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

161

nasopharyngectomies, 26 re-irradiations to $60 Gy; P ˆ 0:48). Despite aggressive salvage treatments, complete remission was achieved in only around two-thirds of the cases (six in group A and 25 in group B) and long-term disease-free survivors were uncommon (three in group A and 18 in group B).

4. Discussion The selection of patients for ICT in TMH was based on

Fig. 3. Comparison of the FLF between the two indications of the ICT and no ICT.

Fig. 2. (a) Comparison of the FLF between patients given ICT and patients not given ICT in the PWH. (b) Comparison of the FLF between patients given ICT and patients not given ICT in the TMH. (c) Comparison of the FLF between all the patients given ICT and all the patients not given ICT.

histological local residual cancer 4±6 weeks after completion of ERT. On the other hand, the indications for ICT in the PWH included both clinical (endoscopical) and histological local residual disease at 4±6 weeks after ERT. In addition, a signi®cant number of patients (n ˆ 62) were given adjuvant ICT after complete response to ERT. We believe that local persistence, especially histological local persistence at 4±6 weeks after ERT, predisposed patients to ultimate local failure for the following reasons: (a), in 1990, Yan et al. [31] showed in a randomized study, that histological local persistence diagnosed soon after primary ERT runs a risk of ultimate local failure signi®cantly higher than that of the histological complete responders to ERT or that of the local persistences given additional radiation; (b), although spontaneous, complete histological regression of local persistences diagnosed within 3 months after ERT occurred, eventually, in half to two-thirds of the cases [18,30], the remainder will persist inde®nitely and contribute to an increased ultimate local failure rate; and (c), not only will at least one-third of the histological residual tumours diagnosed at various intervals within 3 months after primary ERT persist inde®nitely [9,10,18,21,30,31], but those requiring a longer than usual time to regress completely have been shown, by multivariate analysis, to carry a signi®cantly higher risk of ultimate local failure compared with the immediate complete responders to ERT if the initial T-stage is T3 or T4 [22]. If local persistence at 4±6 weeks after ERT is a genuine poor prognostic factor, by extrapolating clinical experience from the advanced T-stage to the early T-stage, we have chosen a substantial number of patients predisposed to ultimate local failure by virtue of this very character for ICT in group A. Moreover, group A had signi®cantly more T2 tumours proportionately and less frequent exposure to neoadjuvant chemotherapy (because of fewer patients with advanced N-stage; Table 1). Therefore, the paradoxical signi®cant reduction of local failures in group A warrants

162

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

Fig. 4. (a) The crude local failure rate (with 95% con®dence interval) of all patients as a function of the total physical radiation dose (ERT dose 1 ICT dose). (b) The crude local failure rate (with 95% con®dence interval) of all patients as a function of the total BED-10, uncorrected for tumour repopulation. (c) The crude local failure rate (with 95% con®dence interval) of all patients as a function of the total BED-10, corrected for tumour repopulation during the whole period of radiotherapy (start of ERT to end of ICT).

an explanation. The data of the PWH had been separately analyzed and published with the conclusion that the ICT signi®cantly enhanced local tumour control over that achievable by ERT alone for early T-stage NPC [24]. The present pooled data from both PWH and TMH further reinforced the statistical power of the analysis and the strength of the conclusions previously drawn based solely on the PWH data. From the present pooled data, the administration of ICT was shown by both univariate and multivariate

analyses to be associated with signi®cantly less local failures. In addition, groups A and B had comparable rates of distant metastasis as ®rst failure and comparable times after radiotherapy to the development of the distant metastasis (Table 2). With comparable distant metastasis rates and times after radiotherapy to the development of distant metastasis between groups A and B, the signi®cantly higher rate of ultimate local failure in group B could not be accounted for by excessive `competitive' mortality from distant metastasis in either group. Thus, the present data support the role of ICT in enhancing ultimate local control for the early T-stage NPC that persist after ERT to conventional cancericidal dose. It is particularly noteworthy that our 5-year actuarial local control rate of 94.6% is nearly identical to the 94% reported by Levendag et al. [14] for the T1±T3 tumours according to the UICC/AJCC 1992 classi®cation. However, T1±T3 according to the UICC/AJCC classi®cation 1992 is equivalent to T1±T2 according to the UICC classi®cation 1997 [8]. Notwithstanding the minor differences in the manner that the doses of ERT and ICT were prescribed by Levendag et al. [14] and us, the ERT dose was identical (both being 60 Gy) and the ICT dose was very similar (18 Gy for Levendag et al. vs. 18±24 Gy for the present study) between the two series for the T1±T2a tumours (UICC/AJCC 1992) [8]. Therefore, both this study and Levendag et al. [14] achieved a local control in the order of 94% for early T-stage NPC with the addition of ICT after ERT (with comparable dose levels of both ICT and ERT between the two series). To our best knowledge, this level of local control has never been reported previously with ERT alone for the T1 1 T2 tumours. Reviewing the literature, in the absence of prospective documentation of the local tumour status at the completion of primary ERT, it is often not possible to differentiate between reports on brachytherapy used as an adjuvant to ERT and reports on brachytherapy used to treat local persistence after ERT. Nonetheless, only a minority (8±13%) of tumours [10,18,21,30,31] persist locally after primary ERT, so the series employing brachytherapy to supplement ERT without mentioning the tumour response to ERT should be regarded as essentially reporting on the adjuvant use of brachytherapy [1,3,14,25,27,29,33]. Common to most of these series, the follow-up intervals were too short and the patient numbers were too small to draw de®nite conclusions. Nonetheless, 5-year local controls in excess of 90% were often reported [1,3,14,33] after using brachytherapy to supplement ERT, suggesting that supplementing ERT by brachytherapy could enhance local control and that a dose±tumour-control relationship of NPC exists above the conventional tumoricidal dose using ERT alone. Previously, NPC was considered a very radiosensitive tumour which could be readily controlled by external beam radiation to a modest dose and there was no convincing data to support the use of a total radiation dose above 60 Gy. Indeed, our ERT dose of 60 Gy (2.5 Gy/fraction, four fractions/week), equivalent to 75 Gy of BED-10, was at the

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

163

Table 3 Comparison of the number of chronic radiation complications and the time after radiotherapy to their occurrence between groups A and B a

Epistaxis BND Epistaxis and/or BND Median (months) Range (months) CN palsy II III V VI XII Total Median (months) Range (months) NP ulceration/necrosis Median (months) Range (months) Neuro-endocrine Median (months) Range (months) Re-treatment Median (months) Range (months) a

Group A (n ˆ 188)

Group B (n ˆ 555)

Statistical test

P value

13 3 14 36.5 10.6±107

25 14 36 40.8 8.0±146

Chi-square

0.521

Wilcoxon

0.689

2 0 0 1 1 4 33.3 19.4±79.7 10 49.4 6.80±95.7 17 50.4 31.1±145 9 22.3 12.5±123

3 1 1 4 14 23 58.2 8.50±137 1 63.5 63.5 33 64.0 22.4±135 34 30.3 8.70±134

Chi-square

0.56

Wilcoxon

0.242

Fisher's NA

0.0001 NA

Chi-square Wilcoxon

0.22 0.870

Chi-square Wilcoxon

0.103 0.612

BNS, bloody nasal discharge; CN, cranial nerve(s); NP, nasopharynx; NA, statistical test not applicable.

lower level of the conventional tumoricidal dose range for NPC (60±70 Gy) [1,7,10,29,30]. Despite the relatively low ERT dose used in Hong Kong[7,10,18,24], the local control rate of NPC with ERT alone has been apparently comparable with those reported by series using similar or higher ERT doses [1,29,30]. Therefore, prior to the PWH report [24], the report by Levendag et al. [14] and the present report, we did not realize the presence of a dose±tumourcontrol relationship for NPC above 60 Gy (BED-10 of 75 Gy), and hence, routine boosting of the total dose by ICT (or other means) was not considered necessary. In the present series, pooling patient data for two hospitals with similar radiation techniques and indications, the administration of ICT was shown to be the only signi®cant factor in determining local control, independent of the other potential prognostic variables. When the administration or not of ICT was excluded from the Cox regression model, the local failure rate correlated signi®cantly with the total physical dose or the total BED-10 without correcting for tumour repopulation during the time course of radiotherapy (from the start of ERT to the end of ICT). However, when correction for the possibility of tumour repopulation with a correction term of 0.5 Gy/day was made for the whole period of radiotherapy, which included the time gap between ERT and ICT, the correlation between the corrected total BED-10 and the ultimate local failure rate in the Cox regression model became weaker and fell short of statistical signi®cance (P ˆ 0:07). Nevertheless, there was a consistently strong trend towards less local failures with escalating doses when the ultimate local failure rate was plotted against the

total physical dose (Fig. 4a), the uncorrected BED-10 (Fig. 4b) or the corrected BED-10 (Fig. 4c). Assuming that a signi®cant dose±tumour-control relationship is present, it would be prudent to query the appropriateness of employing a correction term of the value of 0.5 Gy/day or correcting for the effect of tumour repopulation throughout the whole time course of the radiotherapy. Indeed, Withers et al. [28] showed that the effect of tumour repopulation was only prominent towards the latter portion (21±28 days after start of radiotherapy) of a course of radical radiotherapy. This implies that it is not necessary to correct for tumour repopulation for the initial 3±4 weeks of the ERT. Therefore, instead of applying the correction term of 0.5 Gy/day

Fig. 5. Comparison of the actuarial cumulative chronic radiation complication rate (/patient) between groups A and B.

164

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

from the start of ERT to the end of ICT, we repeated the calculation of BED-10 for each patient sparing the correction for the initial 4 weeks of the ERT. However, despite a strong trend towards less local failures with higher values of BED-10, the correlation between the BED-10 corrected for tumour repopulation in this manner and the ultimate local failure rate in the Cox regression model still fell short of statistical signi®cance (P ˆ 0:07), similar to the case when the BED-10 was corrected for the whole period of radiotherapy. Also, the shape of the graph plotting the local failure rate against the BED-10 was nearly identical to that of Fig. 4c, with simply a horizontal shift of the BED-10 values. This is because the omission of correction for tumour repopulation for the initial 4 weeks of ERT, when applied to all patients, resulted in an increase of BED-10 of an equal amount for each patient. The weaker correlation between local failure rate and the time-corrected BED-10s (compared with total physical dose or uncorrected BED10) could be due to the fact that our range of overall radiotherapy times was relatively narrow, without enough variation to reach statistical signi®cance in spite of the obvious trend (Fig. 4c). On the other hand, it appears more likely that over-correction for tumour repopulation during the signi®cant time gap between ERT and ICT (typically 5±7 weeks), which concerned only those patients subject to ICT, constituted the greatest problem and inaccuracy in the calculation of the corrected BED-10s, and contributed substantially to the lack of statistically signi®cant correlation between the ultimate local failure rate and the corrected BED-10s. In fact, the value of 0.5 Gy/day [2,6,14,26] for the correction of tumour repopulation was originally derived from tumour models which were subject to interruptions or delay of planned continuous delivery of the tumoricidal dose. In our case, the planned continuous delivery of the conventional tumoricidal dose by ERT was completed (without mid-course interruption) before the not insigni®cant time gap (typically 5±7 weeks) between ERT and ICT had occurred. The correction term of 0.5 Gy/day is not likely to be applicable after a planned and uninterrupted course of ERT delivering the conventional tumoricidal dose (uncorrected BED-10, $75 Gy) when the tumour is very likely to have fallen back towards its pre-treatment slow rate of proliferation. In addition, we used a daily fractional dose of 2.5 rather than 2.0 Gy during ERT for most patients, leading to nearly a 5% difference in the value of Gy/day to subtract, since the daily subtraction should be smaller for larger fraction sizes. Altogether, this could explain why a signi®cant dose±tumour-control relationship existed for the total BED-10 uncorrected for tumour repopulation, but not for the BED-10 over-corrected for tumour repopulation by applying a correction term of 0.5 Gy/day. The former must have more accurately represented the actual biological dose relevant for tumour control. Hitherto, support for the existence of a dose±tumourcontrol relationship using ERT alone for NPC has come only from three sources. Firstly, Yan et al. [31] in a prospec-

tive randomized study for 182 tumours persisting soon after 70 Gy of primary ERT, compared the local control between an experimental arm given additional external beam radiation (20±50 Gy, 3±4 Gy/fraction, two fractions/week) and a control arm without additional irradiation. They demonstrated signi®cant enhancement in local control for patients randomized to receive additional ERT (5-year local control rate, 83.3 vs. 54.5% for T1 lesions). Secondly, Lee et al. [11] indicated that there was a 9% compromise in the local control rate for NPC for each Gy of total radiation dose reduction below the conventional tumoricidal dose levels equivalent to 66±70 Gy (2 Gy/fraction, ®ve fractions/week). However, they did not have enough data to de®ne the dose±tumourcontrol relationship above the conventional tumoricidal dose levels. Moreover, the different T-stages were not separately analyzed, so that the dose±tumour-control relationship speci®c for each T-stage (or, at least, speci®c for the early T-stages in contrast to the advanced T-stages) could not be studied. Thirdly, even though both Sham et al. [17] and our group [20] reported signi®cant enhancement in the local control of NPC (T2) that in®ltrated the parapharyngeal (or para-nasopharyngeal) region by the addition of booster ERT via an ipsilateral postero-lateral oblique photon ®eld, one could only barely suggest the existence of a dose±tumour-control relationship because the potential bene®cial effect of recti®cation of initial geographical misses (by the booster ERT) cannot be separated from the effect of dose escalation (by the booster ERT). Neither Sham et al. [17] nor our group [20] mentioned the number of initial geographical misses (during primary ERT) which could have been recti®ed by the booster radiotherapy, and consequently, the dose±tumour-control relationship of the T2-NPC in®ltrating the parapharynx, although suggested, could not be substantiated. The present study is, to the best of our knowledge, the largest multi-institute study that has reported on the dose± tumour-control relationship of early T-stage NPC above the conventional tumoricidal dose levels (uncorrected BED-10, $75 Gy), by supplementing ERT with ICT. We acknowledge the intrinsic assumptions and possible errors, both biological and physical, during the calculation of BEDs for the ICT, the correction for tumour repopulation during radiotherapy, and the summation of the BEDs of the ERT and the BEDs of the ICT [2,6,14,26]. Notwithstanding such limitations, we conclude that a de®nite dose±tumour-control relationship for early T-stage NPC exists above the conventional tumoricidal dose (Fig. 4a±c). This conclusion has been drawn after the virtual exclusion of errors resulting from incorrect T-staging and geographical misses during the primary ERT. 5. Complications With the exception of chronic radiation nasopharyngeal ulceration/necrosis, group A did not differ signi®cantly from group B in the incidence rates of each individual chronic radiation complication (Table 3) and the actuarial

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

cumulative rates of the chronic radiation complications (Fig. 5). Consequential to chronic nasopharyngeal ulceration/necrosis, the main symptoms were headache and/or foul smell. The incidence of headache (®ve in group A vs. none in group B) did not differ signi®cantly between the two groups and the symptom was mild, requiring either only intermittent use of non-narcotic analgesic (n ˆ 3) or no analgesic (n ˆ 2). Foul smell occurred signi®cantly more frequently in group A (n ˆ 6; 3.2%) than in group B (n ˆ 1; 0.18%). However, this symptom was often ameliorated by regular saline lavage of the nasopharynx which the patients could readily be taught to perform. There were no complication-related mortalities or complications requiring hospitalization or surgery. This underlines the safety of the ICT. In our previous reports [21,22,24], ICT was shown to deliver a signi®cant radiation dose to the nasopharynx, the ¯oor of the sphenoid sinus and the medial portion of the parapharyngeal region, with little dose delivery to the lateral portion of the parapharyngeal region, the pituitary fossa, the optic chiasma, the temporal lobes and the temporo-mandibular joints. Therefore, it is perhaps not unexpected that the incidence rates of neuro-endocrine dysfunction and optic nerve (CNII)/chiasma injury were similar between groups A and B (Table 3). Only ®ve of the 17 in group A and one of the 33 in group B with neuro-endocrine dysfunction had symptoms due to the dysfunction; there was no signi®cant difference in the rate between the two groups. In addition, the rates of radiation-induced cranial nerve(s) palsy, which more commonly affected the sixth and the twelfth nerves, were also comparable between the two groups, re¯ecting a similar radiation dose delivery to the para-cavernous-sinus region, the hypoglossal canals and the retrostyloid compartment of the parapharyngeal region with or without ICT. The lack of a signi®cant correlation between the total BED-3 and the incidence rate of the major chronic radiation complications is explainable by the fact that the total BED-3 merely re¯ected the biological dose to the structures proximal to the brachytherapy sources (nasopharynx, sphenoid sinus ¯oor and the medial portion of the parapharynx), while the biological dose to the structures distal to the sources (pituitary fossa, middle cranial fossae, paracavernous-sinus, lateral/retrostyloid compartment of the parapharyngeal region) should be more relevant in determining the incidence of a signi®cant number of major complications. Finally, although short of statistical signi®cance when compared with group B, group A patients required less salvage treatments with nasopharyngectomy and/or re-irradiation to high dose, both of which were associated with signi®cant morbidities [23]. 6. Conclusions and recommendations Notwithstanding the intrinsic limitations of a retrospective study, we conclude that:

165

1. ICT after ERT signi®cantly enhanced local control of early T-stage NPC (T1 and T2 nasal in®ltration). 2. Above the conventional tumoricidal dose (uncorrected BED-10 of $75 Gy), a signi®cant dose±tumour-control relationship existed for the early T-stage NPC. 3. The addition of ICT did not lead to increased chronic radiation complications other than a slight increase in radiation nasopharyngeal ulceration/necrosis. 4. Thus, the signi®cant morbidities associated with salvage treatments (re-irradiation to high dose or nasopharyngectomy with/without postoperative radiotherapy) were largely avoided by supplementing ERT with ICT. The requirement for salvage treatments appeared to be reduced by supplementing ERT with ICT. 5. In view of the acceptable toxicity of the ICT, the statistically signi®cant and clinically important reduction in the ultimate local failure rate by a factor of 2 by the ICT, which was independent of the tumour response to ERT, and the existence of a signi®cant dose±tumourcontrol relationship above the conventional tumoricidal ERT dose, we recommend the liberal application of ICT as a boost after ERT in early T-stage NPC, while awaiting the de®nitive, conclusive evidence from a randomized study to substantiate such a practice.

Acknowledgements The authors would like to thank Ms Joyce Cheng for preparing the manuscript and Mr H.P. Lam for preparing the glossy prints. The authors would also like to acknowledge Dr S.Y. Tsao for co-designing the brachytherapy applicators and head-frame, Dr C.L. Ting and Dr V. Tsang for performing some of the ICT, and Mr Peter Yu for the dosimetric calculations and quality control of the brachytherapy. The authors are grateful to Professor P.J. Johnson for his advice during the preparation of the manuscript. Funding: Clinical Oncology Department Fund, Prince of Wales Hospital; Clinical Oncology Department Fund, Tuen Mun Hospital. References [1] Amornmarn R, Prempree T, Sewchand W, et al. Radiation management of advanced nasopharyngeal cancer. Cancer 1983;52:802±807. [2] Brenner DJ, Hall EJ. Conditions for the equivalence of continuous to pulsed low dose rate brachytherapy. Int J Radiat Oncol Biol Phys 1991;20(1):181±190. [3] Chang JTC, See LC, Hong JH, et al. Intracavitary brachytherapy for early stage nasopharyngeal carcinoma. Chapter 36 in International Brachytherapy, program and abstract of the 8th International Brachytherapy Conference, Nice, France 25±28 November, 1995. [4] Choy D, Sham JST, Wei WI, et al. Transpalatal insertion of radioactive gold grain for the treatment of persistent and recurrent nasopharyngeal carcinoma. Int J Radiat Oncol Biol Phys 1993;25:505±512.

166

P.M.L. Teo et al. / Radiotherapy and Oncology 57 (2000) 155±166

[5] Cox DR. Regression models and life tables. J R Stat Soc B 1972;34:187±220. [6] Fowler JF. The linear±quadratic formula and progress in fractionated radiotherapy. Br J Radiol 1989;62:679±694. [7] Ho JHC. An epidemiologic and clinical study of nasopharyngeal carcinoma. Int J Radiat Oncol Biol Phys 1978;4:183±198. [8] International Union Against Cancer (UICC). TNM classi®cation of malignant tumours, Geneva. 5th ed. 1997. [9] Kwong D, Nicholls J, Sham J, et al. Correlation between endoscopic and histological examination of nasopharynx. Abstract in the 5th Hong Kong International Cancer Congress, 8±11 November, 1998. No. 12.3. [10] Lee AWM, Law SCK, Foo W, et al. Nasopharyngeal carcinoma: local control by megavoltage irradiation. Br J Radiol 1993;66:528±536. [11] Lee AWM, Chan DKK, Fowler JF, et al. Effect of time, dose and fractionation on local control of nasopharyngeal carcinoma. Radiother Oncol 1995;36:24±31. [12] Leung TW, Wong VYW, Tung SY, et al. The importance of threedimensional brachytherapy treatment planning for nasopharyngeal carcinoma. Clin Oncol 1997;9:35±40. [13] Levendag P, Visser A. Microselectron-HDR brachytherapy for head and neck cancer with special reference to the nasopharynx and oropharynx. Chapter 14 in International Brachytherapy, program and abstract of the 7th International Brachytherapy Working Conference, Baltimore, Washington, USA, 6±8 September, 1992. [14] Levendag PC, Schmitz PI, Jansen PP, et al. Fractionated high-doserate brachytherapy in primary carcinoma of the nasopharynx. J Clin Oncol 1998;16(6):2213±2220. [15] Lin ZX, Li DR. Preliminary experience of HDR brachytherapy for nasopharyngeal carcinoma. Chapter 81 in International Brachytherapy, program and abstract of the 8th International Brachytherapy Conference, Nice, France, 25±28 November, 1995. [16] Rosenbaum P, Rubin D. The central role of the propensity score in observational studies for causal effects. Biometrika 1983;70:41±55. [17] Sham JST, Choy D. Prognostic value of paranasopharyngeal extension of nasopharyngeal carcinoma on local control and short-term survival. Head Neck 1991;13(4):298±310. [18] Sham JST, Wei WI, Kwan WH, et al. Fiberoptic endoscopic examination and biopsy in determining the extent of nasopharyngeal carcinoma. Cancer 1989;64(9):1838±1842. [19] Teo PML, Tsao SY, Leung SF. Afterloading intracavitary radiation treatment of nasopharyngeal carcinoma. Acta Oncol 1989;28:525± 527. [20] Teo PML, Tsao SY, Shiu W, et al. A clinical study of 407 cases of NPC in Hong Kong. Int J Radiat Oncol Biol Phys 1989;17:515±530. [21] Teo PML, Leung SF, Choi PHK, et al. Afterloading radiotherapy for

[22]

[23]

[24]

[25]

[26]

[27]

[28]

[29]

[30]

[31]

[32]

[33]

local persistence of nasopharyngeal carcinoma. Br J Radiol 1994;67:181±185. Teo PML, Kwan WH, Yu P, et al. A retrospective study of the role of intracavitary brachytherapy and the prognosticators determining local tumour control after primary radical radiotherapy in 903 non-disseminated nasopharyngeal carcinoma. Clin Oncol 1996;8:160±166. Teo PML, Kwan WH, Chan ATC, et al. How successful is high-dose ($60 Gy) reirradiation using mainly external beams in salvaging local failures of nasopharyngeal carcinoma? Int J Radiat Oncol Biol Phys 1998;40:897±913. Teo PML, Leung SF, Lee WY, Zee B. Intracavitary brachytherapy signi®cantly enhances local control of early T-stage nasopharyngeal carcinoma: the existence of a dose±tumor-control relationship above conventional tumoricidal dose. Int J Radiat Oncol Biol Phys 2000;46(2):445±458. Vikram B, Mishra S. Permanent iodine-125 (I-125) boost implants after external radiation therapy in nasopharyngeal cancer. Int J Radiat Oncol Biol Phys 1994;28(3):699±701. Visser AG, van den Aardweg GJMJ, Levendag PC. Pulsed dose rate and fractionated high dose rate brachytherapy: choice of brachytherapy schedules to replace low dose rate treatments. Int J Radiat Oncol Biol Phys 1996;34(2):497±505. Wang CC. Accelerated hyperfractionation radiation therapy for carcinoma of the nasopharynx techniques and results. Cancer 1989;63:2461±2467. Withers HR, Taylor JMG, Maciejewski B. The hazard of accelerated tumour clonogen repopulation during radiotherapy. Acta Oncol 1988;27:131±146. Yamashita S, Kondo M, Inuyama Y, et al. Improved survival of patients with nasopharyngeal squamous cell carcinoma. Int J Radiat Oncol Biol Phys 1986;12:307±312. Yan JH, Qin DX, Hu YH, et al. Management of local residual primary lesion of nasopharyngeal carcinoma (NPC): are higher doses bene®cial? Int J Radiat Oncol Biol Phys 1989;16:1466±1469. Yan JH, Xu GZ, Hu YH, et al. Management of local residual primary lesion of nasopharyngeal carcinoma: IL results of prospective randomized trial on booster dose. Int J Radiat Oncol Biol Phys 1990;18:295±298. Yin WB, Gao L, Xu GZ. Brachytherapy of nasopharyngeal carcinoma. Chapter 106 in International Brachytherapy. Programme and abstracts of the 8th International Brachytherapy Conference, Nice, France, 25±28 November, 1995. Zhang YW, Liu TF, Fi CX. Intracavitary radiation treatment of nasopharyngeal carcinoma by the high dose rate afterloading technique. Int J Radiat Oncol Biol Phys 1989;16:315±318.