Clinic characteristics were given in Table 1. The median age was 50 years (range, 16–77 years). The male-female ratio was 3.3:1. Eighty-eight (32.6%) patients underwent grade 1 mucositis, 102 (37.7%) patients underwent grade 2 mucositis and 80 (29.6%) patients underwent grade 3 mucositis.
Table 1
Clinic characteristics of 270 nasopharyngeal carcinoma patients
Characteristics | Mucositis |
Grade 0–2 (N = 190) | Grade 3 (N = 80) |
Age (year) Mean (SD) | 51.32 (11.06) | 49.68 (10.98) |
Gender (n, %) | | |
Male | 147(77.4) | 60(75.0) |
Female | 43(22.6) | 20(25.0) |
Diabetes (n, %) | | |
No | 178(93.7) | 73(91.2) |
Yes | 12(6.3) | 7(8.8) |
Hypertension (n, %) | | |
No | 137(72.1) | 58(72.5) |
Yes | 53(27.9) | 22(27.5) |
Smoking (n, %) | | |
No | 111(58.4) | 39(48.8) |
Yes | 79(41.6) | 41(51.2) |
Drinking (n, %) | | |
No | 131(68.9) | 53(66.3) |
Yes | 59(31.1) | 27(33.7) |
BMI before RT (n, %) | | |
23.62 kg/m2 | 115(60.5) | 31(38.8) |
> 23.62 kg/m2 | 75(39.5) | 49(61.2) |
RT duration M (Q25, Q75) | 45(44–47) | 45(44–48) |
T stage (n, %) | | |
T1 − 2 | 41(21.6) | 19(23.7) |
T3 − 4 | 149(78.4) | 61(76.3) |
N stage (n, %) | | |
N0 − 1 | 83(43.7) | 26(32.5) |
N2 − 3 | 107(56.3) | 54(67.5) |
Clinic stage (n, %) | | |
I- II | 21(11.1) | 5(6.3) |
III- IV | 169(88.9) | 75(93.7) |
C-Chemotherapy (n, %) | | |
No | 24(12.6) | 10(12.5) |
Yes | 166(87.4) | 70(87.5) |
C-Nimotuzumab (n, %) | | |
No | 89(46.8) | 27(33.8) |
Yes | 101(53.2) | 53(66.2) |
SGI (n, %) | | |
No | 41(21.6) | 23(28.8) |
Yes | 149(78.4) | 57(71.2) |
Amifostin (n, %) | | |
No | 91(47.9) | 36(45.0) |
Yes | 99(52.1) | 44(55.0) |
IL-11 (n, %) | | |
No | 87(45.8) | 31(38.8) |
Yes | 103(54.2) | 49(61.2) |
RT technique (n, %) | | |
Traditional IMRT | 80(42.1) | 22(27.5) |
TOMO | 110(57.9) | 58(72.5) |
RLN (n, %) | | |
None | 27(14.2) | 3(3.8) |
Unilateral | 55(28.9) | 18(22.5) |
Bilateral | 108(56.8) | 59(73.8) |
I b (n, %) | | |
None | 134(70.5) | 51(63.8) |
Unilateral | 46(24.2) | 26(32.5) |
Bilateral | 10(5.3) | 3(3.8) |
Abbreviation: RT = radiation, BMI = body mass index, C-Chemotherapy = concurrent chemotherapy, C-Nimotuzumab = concurrent nimotuzumab, SGI = sodium giycididaole for injection, IL-11 = recombinant human interleukin-11, IMRT = intensity-modulated radiotherapy, TOMO = helical tomotherapy, RLN = retropharyngeal lymph node region irradiation, I b = I b region irradiation, SD = standard deviation, M = median, Q = quartile. |
Dosimetric parameters
The distribution of each dose-volume objectives from severe oral mucositis group (grade 3) and non-severe oral mucositis group (grade = 1, 2) patients were compared as shown in Supplementary Data (Fig. 1). Distinctively smaller values of non-severe oral mucositis group patients can be observed for mean dose, maximum dose, minimum dose, V10-V65 (%) directly from plots. Then, all the dosimetric parameters performed logistic univariate analysis to screen potential predictors for the development of severe oral mucositis. Mean dose, maximum dose, minimum dose, V15 (%) and V40-V70 (%) under OCC method were risk factors (all p), while mean dose, maximum dose, V10-V15 (%) and V25-V70 (%) were risk factors by using MSC method (all p), p values and odd ratios are listed in Supplementary Data (Table 1). It can be observed that most of intermediate to high dose of volume percentage (V40-V70 (%)) in both MSC and OCC showed statistically significant difference between two groups. In order to further analyze the predictive power of the dose-volume objectives for severe oral mucositis, we performed univariate (receiver operating characteristic) ROC analysis for all above objectives. The predictive power was quantified as area under curve (AUC). Figure 1 shows AUCs of all the dose-volume objectives under both OCC and MSC methods. Better performance can be observed for objectives under MSC method in general. Besides, we considered correlation between dosimetric parameters. Thus, we performed non-parametric Spearman’s test, except Dmin and V15(%) in OCC, V10-15 (%) in MSC have no correlation with some other variables, almost all other variables have significant correlations with each other in each method, coefficients are shown in Supplementary Data (Tables 2 and 3). On the basis of these results, the best performing objective MSC-V55 (%) and OCC-V55 (%) (with the highest AUC) was selected as the representative dosimetric predictor in final multivariate analysis.
Table 2
Logistic univariate analysis of clinic factors for the development of severe mucositis
| OR (95%CI) | P value |
Age | 0.99(0.96–1.01) | 0.265 |
Gender (male vs. female) | 0.88(0.48–1.61) | 0.674 |
Smoking (Yes vs. no) | 1.48(0.87–2.50) | 0.145 |
Drinking (Yes vs. no) | 1.13 (0.65–1.97) | 0.664 |
Hypertension (Yes vs. no) | 0.98(0.55–1.76) | 0.947 |
Diabetes (Yes vs. no) | 1.42(0.54–3.76) | 0.477 |
BMI before RT (kg/m2) (> 23.62 vs. 23.62) | 1.91(1.11–3.31) | 0.020* |
RT technique (TOMO vs. IMRT) | 1.92(1.09–3.39) | 0.025* |
T stage (T3 − 4 vs. T1 − 2) | 1.13(0.61–2.11) | 0.695 |
N stage (N2 − 3 vs. N0 − 1) | 1.61(0.93–2.79) | 0.088 |
Clinic stage (III-IV vs. I-II) | 1.77(0.64–4.88) | 0.274 |
C-Chemotherapy (Yes vs. no) | 1.01(0.46–2.23) | 0.976 |
C-Nimotuzumab (Yes vs. no) | 1.73(1.00-2.98) | 0.048* |
SGI (Yes vs. no) | 0.68(0.38–1.24) | 0.207 |
Amifostin (Yes vs. no) | 1.12 (0.67–1.90) | 0.664 |
IL-11 (Yes vs. no) | 1.34(0.78–2.27) | 0.288 |
RLN | | |
Unilateral (Unilateral vs. none) | 2.95(0.80-10.87) | 0.495 |
Bilateral (Bilateral vs. none) | 4.92(1.43–16.89) | 0.005* |
I b | | |
Unilateral (Unilateral vs. none) | 1.49(0.83–2.65) | 0.219 |
Bilateral (Bilateral vs. none) | 0.79(0.21–2.98) | 0.519 |
Abbreviation: SGI = sodium giycididaole for injection, RT = radiation, RLN = retropharyngeal lymph node region irradiation, I b = I b region irradiation, IL-11 = recombinant human interleukin-11, C-Chemotherapy = concurrent chemotherapy, C-Nimotuzumab = concurrent nimotuzumab, BMI = body mass index, TOMO = helical tomotheapy, IMRT = intensity-modulated radiation therapy, OR = odd ratio, CI = confidence interval. Note: *Statistically significant at p = 0 .05 level. |
Table 3
Predictive modesl for severe mucositis by using OCC and MSC method
Method | Variable | Regression coefficient | OR (95%CI) | P value |
MSC | MSC-V55 (%) (X1) | 0.132 | 1.141 (1.079–1.206) | 0.001 |
BMI before RT (X2) | 0.993 | 2.699 (1.345–5.417) | 0.005 |
Unilateral RLN (X3) | 1.905 | 6.722 (1.084–41.677) | 0.041 |
Bilateral RLN (X4) | 2.592 | 13.360 (2.280-78.283) | 0.004 |
Constant | -4.580 | - | 0.001 |
OCC | OCC-V55 (%) (X1) | 0.206 | 1.228 (1.122–1.344) | 0.001 |
BMI before RT (X2) | 1.002 | 2.724 (1.356–5.472) | 0.005 |
Unilateral RLN (X3) | 2.135 | 8.454 (1.126–63.489) | 0.038 |
Bilateral RLN (X4) | 2.487 | 12.031 (1.764–82.055) | 0.011 |
C-Nimotuzumab (X5) | 0.953 | 2.592 (1.231–5.458) | 0.012 |
Constant | -5.427 | - | 0.001 |
Abbreviation: OCC = oral mucosa contour, MSC = mucosa surface contour, BMI = body mass index, RT = radiation, RLN = retropharyngeal lymph node region irradiation, C-Nimotuzumab = concurrent nimotuzumab, OR = odd ratio, CI = confidence interval. Variable assignment: X1 = OCC-V55 (%), or = MSC-V55 (%) in OCC and MSC model respectively, X2 (BMI 23.62 kg/m2: 0, BMI 23.62 kg/m2: 1), X3 (none RLN region irradiation: 0, unilateral RLN region irradiation: 1), X4 (none RLN region irradiation: 0, bilateral RLN region irradiation: 1), X5 (none concurrent nimotuzumab: 0, concurrent nimotuzumab: 1). Note: *Statistically significant at p = 0 .05 level. |
Clinic factors
Logistic univariate analysis was performed in potential clinical predictors. Basic information: age, gender, diabetes, hypertension, smoking and drinking history, smoking index, T and N stage, clinic stage (7th Edition of American Joint Committee on Cancer stage), BMI (body mass index) before radiation. Treatment information: radiation technique, irradiation status of retropharyngeal lymph node region, irradiation status of I b region lymph node region, concurrent chemotherapy, concurrent nimotuzumab, sodium giycididaole for injection, amifostin and recombinant human interleukin-11, as shown in Table 2. BMI was analysis as a category variable, the cutoff 23.62 kg/m2 was carried out by ROC curve. Only variables with a p-value less than 0.05 could been included in the final multivariate analysis. BMI before radiation, radiation technique, concurrent nimotuzumab and irradiation status of retropharyngeal lymph node (RLN) region were selected as relevant factors to severe oral mucositis and included into the logistic multivariate analysis.
Final model
In consideration of latent correlation between RLN region irradiation and V55 (%), we performed non-parametric spearman’s test, finally no significant correlation was found between them, as shown in Supplementary Data (Table 4). In total, 5 predictors including V55, BMI before radiation, radiation technique, concurrent nimotuzumab, and RLN region irradiation were included into multivariate analysis. All patients were randomly divided into training set and validation
set at the ratio of 8:2. Each model was trained using logistic regression through SPSS Modeler 18.0. Based on OCC, OCC-V55 (%), BMI before radiation, concurrent nimotuzumab and RLN region irradiation were independent factors. Based on MSC, MSC-V55 (%), BMI before radiation and RLN region irradiation were independent factors. In training set, mean AUC of OCC-based model and MSC-based model in training set were 0.786 both. In validation set, the AUCs were 0.721 and 0.622 in MSC and OCC based model respectively, higher AUC level can be observed for MSC-based predictors after comparing the AUC distributions (Fig. 2). Two prediction models were generated, MSC-based function was logit(P) = ln () =-4.580 + 0.132 X1 + 0.993 X2 + 1.905 X3 + 2.592 X4 and OCC-based function was logit(P) = ln () =-5.427 + 0.206 X1 + 1.002X2 + 2.135X3 + 2.487 × 4 + 0.953 X5, regression coefficients were shown in Table 3. The dosimetric parameter of V55(%) was the most important predictive factor in both discriminant function, the predictive importance of each variable was shown in Fig. 3. P represents the probability of severe oral mucositis under the influence of the above factors in functions. When P 0.5, the closer the P value is to 1, the greater the probability of severe oral mucositis is than that of non-severe oral mucositis (1-P). The performance evaluated parameters were shown in Supplementary Data (Table 5).