The significant survival advantage of female sex in nasopharyngeal carcinoma: a propensity-matched analysis.

BACKGROUND: Whether females have better survival than males in nasopharyngeal carcinoma is barely acknowledged and the exact explanations remain unknown.
METHODS: Overall, 5929 patients receiving treatment between January 2005 and December 2010 were separately stratified by stage into early and advanced stage groups, and by age into premenopausal (⩽45 years), menopausal (46-54 years) and postmenopausal (⩾55 years) groups. Matched males and females in each group were identified using the propensity score matching method. Differences in disease-free survival (DSS), overall survival (OS), distant metastasis-free survival (DMFS) and locoregional relapse-free survival (LRFS) were estimated by the Kaplan-Meier method and Cox regression model.
RESULTS: Overall, 398, 923, 744, 319 and 313 pairs of males and females were matched in early stage, advanced stage, premenopausal, menopausal and postmenopausal group, respectively. Females showed significant advantage over males across all end points in both early and advanced stage groups (P⩽0.042). However, this advantage persisted at premenopausal age (P⩽0.042), declined during menopause (DMFS, P=0.021; DSS, P=0.100; OS, P=0.693; LRFS, P=0.330) and totally disappeared at postmenopausal age (P⩾0.344).
CONCLUSIONS: Sex significantly affects NPC survival, with a definite female advantage regardless of tumour stage. Intrinsic biologic traits appear to be the exact explanation according to the declining magnitude of sex effect with age.

Nasopharyngeal carcinoma (NPC) is a malignancy with unique geographic distribution. It is rare in Europe and the United States, with an incidence of 0.5–2 per 100 000 (Ferlay ). By contrast, NPC is endemic in Southern China (Cao ) and Hong Kong (Chang and Adami, 2006) where the incidence can be as high as 20–30 per 100 000. There is a common feature of the incidence rates almost in all populations—the male predominance, with a male-to-female ratio of 2–3 : 1 (Ferlay ).

With regard to the prognostic effect of sex on the treatment outcomes of patients with established NPC, significant female advantage in survival was found in a randomised controlled trial about chemotherapy (Lee ), but null results were reported in the other three trials (Chen , 2012; Fountzilas ). Although female NPC patients were found to have higher survival rates than male counterparts in a retrospective comparison (Xiao ), the causes remain confused. It was previously assumed that behavioural differences across sex, especially diagnostic delays, might contribute to the observed sex differences in survival of caner of oesophagus (Bohanes ) and melanoma (Joosse ). According to the population-based evidence of age-dependent sex ratio in the incidence of NPC with an inflection at menopause ages and a delay of developing NPC in females before menopause (Xie ), we proposed another hypothesis of the intrinsic biologic sex differences, mainly about the protective effect of oestrogen. If the first assumption in other cancers can be applied to NPC, then it is of great necessity to fully balance the interactions of sex and other behavioural prognoses, especially tumour stage. If the second hypothesis holds, then it is important to detect the survival advantage of female sex in all types of disease progression (e.g., both locoregional relapse and distant metastasis) on one side, and no survival superiority of female sex among postmenopausal patients on the other.

To well balance the influence of covariates, we compared the survival outcomes of male and female NPC patients using the propensity score matching method (Baser, 2006; Austin, 2009) and multivariate analysis. To clarify the exact explanations, the influence of behavioural factors and intrinsic biologic trait was further analysed by assessing the magnitude of the prognostic effect of sex in different groups.

Patients and methods

Patients

This retrospective study was approved by the Institutional Review Board at Sun Yat-sen University Cancer Center, and individual informed consent was waived given the anonymous analysis of routine data. Between January 2005 and December 2010, 5929 newly diagnosed, biopsy-proven, non-metastatic and hospitalised NPC patients who were at the age of 20 or above were entered into this study. All patients had complete pretreatment evaluation including patient history, physical examination, haematology and biochemistry profiles, fiberoptic nasopharyngoscopy with biopsy, magnetic resonance imaging (MRI) of the nasopharynx and neck, chest radiography, abdominal sonography and Technetium-99 m-methylene diphosphonate (Tc-99-MDP) whole-body bone scan. Patients were restaged according to the seventh edition of the International Union against Cancer/American Joint Committee on Cancer staging system for NPC (Edge ).

Treatment

All patients were treated by definitive intensity-modulated radiotherapy (IMRT) or conventional radiotherapy (CRT) with or without chemotherapy. Further details of the radiation techniques used at our centre have been described previously (Lai ). Institutional guidelines recommended no chemotherapy for patients in early stage, and induction, concurrent and adjuvant chemotherapy or combined treatment for those in locoregionally advanced stage. Induction or adjuvant chemotherapy consisted of cisplatin with 5-fluorouracil, cisplatin with taxane or triplet of cisplatin and 5-fluorouracil plus taxane every 3 weeks for 2–3 cycles. Concurrent chemotherapy consisted of cisplatin given on weeks 1, 4 and 7 of radiotherapy or cisplatin given weekly. Deviation from the institutional guidelines was result from organ dysfunction, treatment intolerance and/or patient refusal.

Follow-up

Patients were examined every 3–6 months during the first 3 years, with follow-up examinations every 6–12 months thereafter until death. During this period, patients were assessed by history and physical examination and a series of conventional examination equipment at each follow-up visit, to detect the possible relapse or distant metastasis. Local relapses were confirmed by biopsy, MRI scan, or both. Regional relapses were diagnosed by clinical examination and an MRI scan of the neck and, in doubtful cases, by fine needle aspiration of the lymph nodes. Distant metastases were diagnosed by clinical symptoms, physical examinations and imaging methods including chest radiography, bones scan, MRI and abdominal sonography. Patients without recent examination tests in the medical records were followed up by telephone call.

Statistical analysis

We selected male patients who were matched with the female counterparts using propensity score matching. This is a method for creating similar case (female) and control (male) sets from an existing data set on the presumed covariates, to reduce possible biases to a minimum in a retrospective analysis (D'Agostino, 1998). Propensity scores were computed by logistic regression for each patient based on the following covariates, including age, smoking, drinking, histology, titres of immunoglobulin A against viral capsid antigen (VCA-IgA) and early antigen (EA-IgA), body mass index (BMI), T-stage, N-stage, clinical stage, radiation techniques and chemotherapy regimens. Female and male patients were then matched without replacement at the ratio of 1 : 1 on those scores, rather than the individual covariates.

To test the individual hypothesis, 5929 patients were independently stratified by clinical stage and age into five groups: early stage (stage I+II) group and advanced stage (stage III+IVa-b) group according to clinical stage; premenopausal age (⩽45 years) group, menopausal age (46–54 years) group and postmenopausal age (⩾55 years) group according to age. Propensity score matching was utilised to identify the matched female and male patients in each group for subsequent analysis.

Covariates balance between female and male was examined by t test (continuous variable), χ2 test or Fisher's exact test (categorical variable) as appropriate. Disease-specific survival (DSS), overall survival (OS), distant metastasis-free survival (DMFS) and locoregional relapse-free survival (LRFS) were estimated with the Kaplan–Meier method (Kaplan and Meier, 1958). Crude and adjusted hazard ratios with 95% confidence intervals for sex (with male sex as reference) were calculated using Cox regression analysis or Cox regression model with time-dependent covariates if the proportional hazards assumption did not hold (Cox, 1972). DSS, OS, LRFS and DMFS were defined as the time from treatment to death resulting from NPC or treatment complications, to death from any cause, to the first locoregional relapse and to the first distant metastasis, respectively.

All statistical analyses were performed using IBM SPSS Statistics version 22.0 (http://www-01.ibm.com/software/analytics/spss/downloads.html). Two-sided P-values<0.05 were considered to be significant.

Results

From the original 473 female vs 1274 male patients with stage I/II and 1029 female vs 3153 male patients with stage III/IVa-b (Supplementary Table 1), 398 and 923 pairs were respectively selected by propensity score matching (Table 1). The median follow-up time for the included patients in both groups was 61.78 months (4.37–109.23 months) and 53.65 months (3.50–110.67 months), respectively.

Table 1

Baseline characteristics in early and advanced stage groups

	Early stage (I+II)					Advanced stage (III+IVa-b)
	Male (n=398)		Female (n=398)			Male (n=923)		Female (n=923)
	No.	%	No.	%	P	No.	%	No.	%	P
Age					0.501					0.493
Mean	44.6		45.1			45.1		44.7
SD	10.7		11.1			11.9		11.0
Median	44.0		44.0			44.0		44.0
Smoking					1.000					0.880
Ever	5	1.3	5	1.3		23	2.5	22	2.4
Never	393	98.7	393	98.7		900	97.5	901	97.6
Drinking					1.000a					0.283
Ever	1	0.3	0	0		9	1.0	5	0.5
Never	397	99.7	398	100.0		914	99.0	918	99.5
Histologyb					0.785					0.404
I+II	28	7.0	30	7.5		58	6.3	67	7.3
III	370	93.0	368	92.5		865	93.7	856	92.7
VCA-IgAc					0.506					0.175
<80	136	34.2	144	36.2		216	23.4	234	25.4
80–320	194	48.7	178	44.7		487	52.8	447	48.4
⩾320	68	17.1	76	19.1		220	23.8	242	26.2
EA-IgAc					0.994					0.594
<10	220	55.3	220	55.3		392	42.5	395	42.8
10–40	116	29.1	115	28.9		311	33.7	293	31.7
⩾40	62	15.6	63	15.8		220	23.8	235	25.5
BMI					0.606					0.916
Mean	23.0		22.9			22.4		22.4
SD	2.9		3.2			3.0		3.8
Median	22.9		22.8			22.3		22.0
T-stage					0.722					0.763
T1	181	45.5	186	46.7		33	3.6	41	4.4
T2	217	54.5	212	53.3		65	7.0	59	6.4
T3	—					489	53.0	487	52.8
T4	—					336	36.4	336	36.4
N-stage					0.594					0.518
N0	123	30.9	130	32.7		128	13.9	119	12.9
N1	275	69.1	268	67.3		500	54.2	513	55.6
N2	—					236	25.6	245	26.5
N3	—					59	6.4	46	5.0
Clinical stage					0.653					0.705
I	74	18.6	79	19.8		—		—
II	324	81.4	319	80.2		—		—
III	—					541	58.6	549	59.5
IV	—					382	41.4	374	40.5
RT					0.877					0.842
CRT	279	70.1	277	69.6		630	68.3	626	67.8
IMRT	119	29.9	121	30.4		293	31.7	297	32.2
CT					0.914					0.160
No	178	44.7	172	43.2		102	11.1	73	7.9
IC	48	12.1	43	10.8		231	25.0	236	25.6
CC	116	29.1	125	31.4		273	29.6	312	33.8
IC+CC	45	11.3	50	12.6		277	30.0	266	28.8
CC+AC	9	2.3	6	1.5		27	2.9	25	2.7
IC+CC+AC	2	0.5	2	0.5		13	1.4	11	1.2

Abbreviations: AC=adjuvant chemotherapy; BMI=body mass index; CC=concurrent chemotherapy; CRT=conventional radiotherapy; CT=chemotherapy; EA=early antigen; IC=induction chemotherapy; IgA=immunoglobulin A; IMRT=intensity-modulated radiotherapy; RT=radiotherapy; SD=standard deviation; VCA=viral capsid antigen.

Fisher's exact test.

Based on the criteria of WHO histological type (1991): I—Squamous-cell carcinomas, II—Differentiated non-keratinising carcinoma, III—Undifferentiated non-keratinising carcinoma.

In accordance with the criteria adopted in previous studies.

From the original 821 female vs 2194 male patients at premenopausal age, 359 female vs 1181 male patients at menopausal age and 322 female vs 1052 male patients at postmenopausal age (Supplementary Table 2), 744, 319 and 313 pairs were respectively selected by propensity score matching (Table 2). The median follow-up time for the included patients in the three groups was 58.27 months (3.50–110.67 months), 54.67 months (3.73–109.23 months) and 50.97 months (3.67–111.07 months), respectively.

Table 2

Baseline characteristics in three age groups

	Premenopausal age (⩽45 years)			Menopausal age (46–54 years)			Postmenopausal age (⩾55 years)
	Male (n=744)	Female (n=744)		Male (n=319)	Female (n=319)		Male (n=313)	Female (n=313)
	No. (%)	No. (%)	P	No. (%)	No. (%)	P	No. (%)	No. (%)	P
Age			0.747			0.631			0.717
Mean	36.719	36.622		49.709	49.809		61.211	61.051
SD	5.833	5.748		2.729	2.546		5.369	5.636
Median	37.000	37.000		49.000	50.000		60.000	60.000
Smoking			0.363a			1.000			1.000
Ever	5 (0.7)	3 (0.4)		12 (3.8)	12 (3.8)		12 (3.8)	12 (3.8)
Never	739 (99.3)	741 (99.6)		307 (96.2)	307 (96.2)		301 (96.2)	301 (96.2)
Drinking			1.000a			1.000a			1.000a
Ever	2 (0.3)	2 (0.3)		3 (0.9)	2 (0.6)		4 (1.3)	3 (1.0)
Never	742 (99.7)	742 (99.7)		316 (99.1)	317 (99.4)		309 (98.7)	310 (99.0)
Histologyb			0.407			0.880			0.884
I+II	54 (7.3)	46 (6.2)		24 (7.5)	23 (7.2)		25 (8.0)	26 (8.3)
III	690 (92.7)	698 (93.8)		295 (92.5)	296 (92.8)		26 (92.0)	287 (91.7)
VCA-IgAc			0.461			0.682			0.243
<80	222 (29.8)	231 (31.0)		79 (24.8)	84 (26.3)		69 (22.0)	78 24.9)
80–320	377 (50.7)	354 (47.6)		167 (52.4)	156 (48.9)		166 (53.0)	145 (46.3)
⩾320	145 (19.5)	159 (21.4)		73 (22.9)	79 (24.8)		78 (24.9)	90 (28.8)
EA-IgAc			0.764			0.832			0.681
<10	364 (48.9)	374 (50.3)		143 (44.8)	142 (44.5)		133 (42.5)	125 (39.9)
10–40	233 (31.3)	220 (29.6)		104 (32.6)	99 (31.0)		96 (30.7)	106 (33.9)
⩾40	147 (19.8)	150 (20.2)		72 (22.6)	78 (24.5)		84 (26.8)	82 (26.2)
BMI			0.115			0.887			0.876
Mean	21.998	21.754		23.350	23.315		23.328	23.283
SD	2.784	3.184		2.706	3.361		3.026	4.118
Median	21.887	21.454		23.508	23.147		23.438	23.147
T-stage			0.574			0.844			0.567
T1	125 (16.8)	127 (17.1)		62 (19.4)	57 (17.9)		51 (16.3)	50 (16.0)
T2	177 (23.8)	161 (21.6)		65 (20.4)	67 (21.0)		61 (19.5)	73 (23.3)
T3	253 (34.0)	276 (37.1)		113 (35.4)	122 (38.2)		117 (37.4)	103 (32.9)
T4	189 (25.4)	180 (24.2)		79 (24.8)	73 (22.9)		84 (26.8)	87 (27.8)
N-stage			0.113			0.181			0.434
N0	151 (20.3)	124 (16.7)		80 (25.1)	63 (19.7)		76 (24.3)	66 (21.1)
N1	432 (58.1)	456 (61.3)		174 (54.5)	200 (62.7)		163 (52.1)	179 (57.2)
N2	124 (16.7)	138 (18.5)		53 (16.6)	48 (15.0)		57 (18.2)	57 (18.2)
N3	37 (5.0)	26 (3.5)		12 (3.8)	8 (2.5)		17 (5.4)	11 (3.5)
Clinical stage			0.425			0.581			0.402
I	49 (6.6)	46 (6.2)		28 (8.8)	21 (6.6)		22 (7.0)	18 (5.8)
II	195 (26.2)	186 (25.0)		78 (24.5)	89 (27.9)		63 (20.1)	80 (25.6)
III	279 (37.5)	310 (41.7)		128 (40.1)	130 (40.8)		129 (41.2)	118 (37.7)
IV	221 (29.7)	202 (27.2)		85 (26.6)	79 (24.8)		99 (31.6)	97 (31.0)
RT			0.240			0.666			0.793
CRT	522 (70.2)	501 (67.3)		226 (70.8)	221 (69.3)		218 (69.6)	221 (70.6)
IMRT	222 (29.8)	243 (32.7)		93 (29.2)	98 (30.7)		95 (30.4)	92 (29.4)
CT			0.530			0.464a			0.369a
No	141 (19.0)	121 (16.3)		57 (17.9)	66 (20.7)		91 (29.1)	90 (28.8)
IC	137 (18.4)	154 (20.7)		77 (24.1)	61 (19.1)		82 (26.2)	64 (20.4)
CC	228 (30.6)	244 (32.8)		97 (30.4)	111 (34.8)		80 (25.6)	101 (32.3)
IC+CC	202 (27.2)	194 (26.1)		76 (23.8)	73 (22.9)		51 (16.3)	52 (16.6)
CC+AC	25 (3.4)	24 (3.2)		9 (2.8)	5 (1.6)		7 (2.2)	4 (1.3)
IC+CC+AC	11 (1.5)	7 (0.9)		3 (0.9)	3 (0.9)		2 (0.6)	2 (0.6)

Abbreviations: AC=adjuvant chemotherapy; BMI=body mass index; CC=concurrent chemotherapy; CRT=conventional radiotherapy; CT=chemotherapy; EA=early antigen; IC=induction chemotherapy; IgA=immunoglobulin A; IMRT=intensity-modulated radiotherapy; RT=radiotherapy; SD=standard deviation; VCA=viral capsid antigen.

Fisher's exact test.

Based on the criteria of WHO histological type (1991): I—Squamous-cell carcinomas, II—Differentiated non-keratinising carcinoma, III—Undifferentiated non-keratinising carcinoma.

In accordance with the criteria adopted in previous studies.

The baseline characteristics of all patients before matching (Supplementary Tables 1 and 2) and the excluded patients by matching (Supplementary Tables 3 and 4) were statistically different across sex. However, the included males and females after matching in each group had similar mean of age and BMI, smoking status, drinking status, histology, titres of VCA-IgA and EA-IgA, T-stage, N-stage, clinical stage, radiation techniques and chemotherapy regimens (Tables 1 and 2).

Survival outcomes

Compared with male patients, female counterparts showed significant advantage across all end points in both early stage (DSS rates at 5 years 97.1% vs 91.7%, P=0.002; OS rates at 5 years 97.1% vs 91.7%, P=0.003; DMFS rates at 5 years 97.9% vs 93.4%, P=0.006; and LRFS rates at 5 years 94.8% vs 92.0%, P=0.017; Figure 1A–D) and advanced stage groups (DSS rates at 5 years 86.2% vs 80.7%, P=0.014; OS rates at 5 years 85.8% vs 80.6%, P=0.021; DMFS rates at 5 years 85.5% vs 80.4%, P=0.006; and LRFS rates at 5 years 91.5% vs 87.5%, P=0.042; Figure 1E–H).

Figure 1

Survival outcomes of male and female patients in early stage group (A–D) and advanced stage group (E–H).

However, this female survival advantage across all end points was limited at premenopausal age (⩽45 years) (DSS rates at 5 years 91.5% vs 87.1%, P=0.015; OS rates at 5 years 91.3% vs 87.3%, P=0.023; DMFS rates at 5 years 89.8% vs 85.4%, P=0.028; and LRFS rates at 5 years 91.6% vs 88.4%, P=0.042; Figure 2A–D). There were no significant differences in DSS (rates at 5 years 91.0% vs 87.0%, P=0.100), OS (rates at 5 years 90.7% vs 86.8%, P=0.693) or LRFS (rates at 5 years 93.4% vs 91.7%, P=0.330) between female and male patients at menopausal age (46–54 years), except DMFS (rates at 5 years 91.9% vs 86.0%, P=0.021). (Figure 2E–H) Furthermore, the female survival advantage totally vanished when patients reached the postmenopausal age (⩾55 years) (DSS rates at 5 years 78.9% vs 78.4%, P=0.525; OS rates at 5 years 78.9% vs 78.0%, P=0.518; DMFS rates at 5 years 86.8% vs 85.6%, P=0.344; and LRFS rates at 5 years 88.4% vs 89.3%, P=0.509; Figure 2I–L).

Figure 2

Survival outcomes of male and female patients in premenopausal age group (A–D), menopausal age group (E–H) and postmenopausal age group (I–L).

Multivariate analysis

Accounting for age (continuous), smoking, drinking, histology, titres of VCA-IgA (<80/80–320/⩾320) and EA-IgA (<10/10–40/⩾40), BMI (continuous), T-stage, N-stage, clinical stage, radiation techniques and chemotherapy regimens in multivariate analysis, the significant female advantage persisted for DSS, OS, DMFS and LRFS, regardless of clinical stage (Table 3).

Table 3

Summary of important prognostic factors in multivariate analysis for early and advanced stage groupsa

	Early stage (I+II)		Advanced stage (III+IVa-b)
	Hazard ratio (95% CI)	P	Hazard ratio (95% CI)	P
Disease-specific survival
Sex (male as reference)	0.34 (0.18–0.64)	0.001	0.75 (0.59–0.94)	0.014
Age (continuous)	1.06 (1.03–1.08)	<0.001	1.04 (1.03–1.05)	<0.001
T-stage	NS		1.41 (1.20–1.65)	<0.001
N-stage	8.01 (2.48–25.83)	<0.001	1.84 (1.58–2.14)	<0.001
Body mass index (continuous)	NS		0.94 (0.91–0.98)	0.003
Overall survival
Sex (male as reference)	0.36 (0.19–0.67)	0.001	0.77 (0.61–0.96)	0.022
Age (continuous)	1.06 (1.03–1.08)	<0.001	1.04 (1.03–1.05)	<0.001
T-stage	NS		1.42 (1.21–1.66)	<0.001
N-stage	8.24 (2.56–26.56)	<0.001	1.82 (1.57–2.11)	<0.001
Body mass index (continuous)	NS		0.94 (0.91–0.98)	0.003
Distant metastasis-free survival
Sex (male as reference)	0.33 (0.15–0.74)	0.007	0.72 (0.57–0.91)	0.006
T-stage	NS		1.39 (1.18–1.63)	<0.001
N-stage	4.55 (1.38–14.97)	0.013	2.03 (1.74–2.36)	<0.001
Locoregional relapse-free survival
Sex (male as reference)	0.49 (0.28–0.87)	0.015	0.73 (0.54–0.99)	0.039
Age (continuous)	NS		1.02 (1.01–1.03)	0.008
T-stage	NS	0.421	1.26 (1.01–1.57)	0.044
N-stage	3.82 (1.63–8.95)	0.002	1.31 (1.07–1.62)	0.011

Abbreviations: CI=confidence interval; EA=early antigen; IgA=immunoglobulin A; NS=not significant; VCA=viral capsid antigen.

Adjustment for age (continuous), smoking, drinking, VCA-IgA (<80/80–320/⩾320), EA-IgA (<10/10–40/⩾40), pathology, body mass index (continuous), T-stage, N-stage, radiation techniques and chemotherapy regimens with forward LR method.

With adjustment for the same covariates, female sex was an independent, significant protective predictor of DSS, OS, DMFS and LRFS for patients at premenopausal age, along with DMFS for patients at menopausal age. However, sex was not significantly associated with the DSS, OS or LRFS of patients at menopausal age, or any of the end points of patients at postmenopausal age (Table 4).

Table 4

Summary of important prognostic factors in multivariate analysis for three age groupsa

	Premenopausal age (⩽45 years)		Menopausal age (46–54 years)		Postmenopausal age (⩾55 years)
	Hazard ratio (95% CI)	P	Hazard ratio (95% CI)	P	Hazard ratio (95% CI)	P
Disease-specific survival
Sex (male as reference)	0.64 (0.46–0.89)	0.008	0.66 (0.40–1.09)	0.102	1.00 (0.99–1.01)	0.329b
Age (continuous)	NS		NS		1.04 (1.01–1.07)	0.005
T-stage	1.72 (1.43–2.07)	<0.001	1.55 (1.18–2.04)	0.002	1.43 (1.20–1.71)	<0.001
N-stage	2.14 (1.74–2.63)	<0.001	2.02 (1.47–2.78)	<0.001	1.72 (1.40–2.11)	<0.001
Body mass index (continuous)	NS		0.89 (0.81–0.97)	0.008	NS
Overall survival
Sex (male as reference)	0.66 (0.47–0.91)	0.012	1.00 (0.99–1.01)	0.659b	1.00 (0.99–1.00)	0.307b
Age (continuous)	NS		NS		1.05 (1.02–1.08)	0.001
T-stage	1.74 (1.45–2.01)	<0.001	1.53 (1.17–2.00)	0.002	1.46 (1.23–1.74)	<0.001
N-stage	2.10 (1.71–2.58)	<0.001	2.07 (1.51–2.84)	<0.001	1.70 (1.39–2.08)	<0.001
Body mass index (continuous)	NS		0.90 (0.82–0.98)	0.015	NS
Distant metastasis-free survival
Sex (male as reference)	0.69 (0.51–0.94)	0.017	0.55 (0.33–0.92)	0.021	0.80 (0.52–1.23)	0.303
T-stage	1.63 (1.38–1.93)	<0.001	NS		1.47 (1.16–1.85)	0.001
N-stage	2.30 (1.90–2.77)	<0.001	3.00 (2.21–4.07)	<0.001	2.19 (1.68–2.85)	<0.001
Body mass index (continuous)	NS		0.87 (0.80–0.96)	0.003	NS
Locoregional relapse-free survival
Sex (male as reference)	0.69 (0.49–0.97)	0.034	0.74 (0.39–1.38)	0.340	1.00 (0.98–1.01)	0.519b
T-stage	1.30 (1.09–1.55)	0.004	1.53 (1.10–2.15)	0.013	1.31 (1.02–1.69)	0.038
N-stage	1.36 (1.09–1.71)	0.008	NS		NS

Abbreviations: CI=confidence interval; EA=early antigen; IgA=immunoglobulin A; NS=not significant; VCA=viral capsid antigen.

Adjustment for age (continuous), smoking, drinking, VCA-IgA (<80/80–320/⩾320), EA-IgA (<10/10–40/⩾40), pathology, body mass index (continuous), T-stage, N-stage, radiation techniques and chemotherapy regimens with forward LR method.

Cox regression model with time-dependent covariates.

Discussion

The most appealing result of this study is the convincing prognostic advantage in DSS, OS, DMFS and LRFS from female sex for patients with nasopharyngeal carcinoma using the propensity score matching analysis. Currently, this propensity score matching analysis, along with multivariate analysis, provides the fairest comparison of matched male and female patients to evaluate the sex effect. This protective effect of female sex is fairly consistent with that reported in the literature for NPC (Lee ; Xiao ) and other cancers (Hidaka ; McGovern ; Bohanes ; Cheung ; Joosse ).

The remarkable sex differences in survival were used to be presumably explained by sex differences in lifestyle behaviour and diagnostic delays in other cancers (Bohanes ; Joosse ). However, in this propensity-matched study, several behavioural factors (e.g., smoking and drinking status and BMI) and multiple indicators related to diagnostic delays (e.g., T-stage, N-stage, clinical stage, titre of VCA-IgA and EA-IgA) were well balanced, and even taking these confounders into account caused little shift from crude to adjusted hazard ratios and failed to overturn the significant sex effect. More importantly, sex remained the independent prognostic value across all end points (DSS, OS, DMFS and LRFS) in locoregionally advanced nasopharyngeal carcinoma, as demonstrated in patients with early stage, although the relative female advantage declined from the 51–67% advantage in early stage to a nearly 30% advantage in advanced stage (Table 3). Therefore, our findings indicate that similar to oesophageal cancer (Bohanes ) and melanoma (Joosse ), common lifestyle behaviours and diagnostic delays cannot fully explained the survival differences of male and female NPC patients, and the underlying biologic traits of female sex may have a pivotal role in a much more profound way.

The hormonal differences, especially oestrogen and oestrogen receptor (ER), are very representative of the biologic traits. The levels of oestrogen and ER in the female are known to differ before and after menopause; so there is an oestrogen and ER hypothesis that in postmenopausal women, the survival advantage from oestrogen against men should decline or even completely vanish. Since age is commonly considered as a surrogate for menopause, and only 5% of women enter menopause after age 55 years (McKinlay ), the survival differences of male and female patients were examined in three age groups. We found that the sex differences existed across all end points in premenopausal age (⩽45 years), restricted to DMFS in menopausal age (46–54 years), and totally disappeared after menopause (⩾55 years). Therefore, this finding highly supports the oestrogen and ER hypothesis. Actually, this oestrogen-related sex disparity had already been displayed in the incidence of developing nasopharyngeal carcinoma (Xie ). Unfortunately, little is known about the association of female hormone and survival or the underlying mechanism. This is likely to be the functional result of genetic variants, for example, the VEGF-2578 CC genotype, which was associated with tumour progression and frequently involved with the male patients as indicated by Nasr . Additionally, it was reported that inhibition of ER-α with a repressor (NAG7) could promote nasopharyngeal carcinoma invasion via upregulation of JNK2/AP-1/MMP1 pathways (Huang ).

Apart from the hormonal differences, another way in which the intrinsic biologic traits of sex directly exert is the response rate and probability of side effects from treatment, especially the chemotherapy. Sex-biased expression levels of metabolic enzymes and transporters in liver and kidney lead to different pharmacokinetics for most common anti-cancer drugs. In women, half-life is often longer, which exactly results in a better response rate of cisplatin in female NPC without increasing toxicity (Schmetzer and Florcken, 2012). Finally, other literature-mentioned plausible explanations for the female advantage include the differences in immune homeostasis (Bouman ) and body iron stores (Mascitelli and Goldstein, 2013). Further researches are warranted to confirm or exclude any of these hypothetical biologic explanations.

The major strength of this study lies in the investigation of sex effect in nasopharyngeal carcinoma using propensity score matching and multivariate analysis. This directly addressed the limitations of divergent confounders, treatment heterogeneity and selection bias associated with the retrospective assessment of observational data (Austin, 2009). Additional strength is that the common hypotheses to explain the sex differences were tested, for the first time, in separate groups of matched male and female nasopharyngeal carcinoma patients.

Anyway, it was a limitation that the presented data were derived from a single institution in endemic area with expertise in diagnosing and treating this disease. Moreover, since data on DNA copy number of the Epstein-Barr virus were missing in most of cases, VCA-IgA and EA-IgA were taken as the surrogate. Finally, true anamnesis on menopausal status, data on hormonal analysis and information on hormone replacement therapy were missing in this retrospective study. However, stratified analysis by three age groups was a valuable alternative to indirectly disclose the correlation of survival differences across sex with hormone, because age is commonly considered as a surrogate for menopause. These issues would be addressed in the coming prospective study.

In conclusion, sex significantly affected the survival of nasopharyngeal carcinoma, with a definite female advantage across all end points, independent of other prognostic factors. This female survival advantage persisted in all stages of this cancer, but disappeared among postmenopausal women. It was strongly associated with the underlying biologic traits of sex, rather than the behavioural sex disparities. Sex is of great necessity to be stratified for analysis in the upcoming randomised controlled trials.