Characteristics of the study population
The baseline characteristics and main lifestyle factors for cases and controls are shown in Table 1. Compared with controls, cases were more likely to have a higher proportion of tobacco, alcohol consumption, tumor history and worse oral hygiene. In addition, the distribution of gender, education level, BMI, residence was significantly different between the case and control group (p<0.05). General differences of fatty acid intake were observed between case and control, intake of SFA such as C14:0, C16:0, C18:0 are higher in case than control group (p<0.001), intake of monounsaturated fatty acids(MUFA) such as C18:1 is higher in case than control group (p<0.001). The distribution of dietary fatty acids between the case and control are shown in Supplement Figure 1.
Table1 Characteristics of the case(n=446) and control(n=448)
Variable
|
Case
|
Control
|
P
|
Age
|
|
|
<0.001
|
<49
|
94(21.1%)
|
210(46.9%)
|
|
≥49
|
352(78.9%)
|
238(53.1%)
|
|
Gender
|
|
|
0.002
|
Male
|
258(57.8%)
|
213(47.5%)
|
|
Female
|
188(42.2%)
|
235(52.2%)
|
|
Education
|
|
|
<0.001
|
low
|
77(17.3%)
|
204(45.5%)
|
|
high
|
369(82.7%)
|
244(54.5%)
|
|
Marital status
|
|
|
0.699
|
Married
|
408(91.5%)
|
413(92.2%)
|
|
Unmarried and Others
|
38(8.5%)
|
35(7.8%)
|
|
BMI
|
|
|
0.024
|
<18.5
|
39(8.7%)
|
19(4.2%)
|
|
18.5~
|
284(63.7%)
|
297(66.3%)
|
|
≥24
|
123(27.6%)
|
132(29.5%)
|
|
Residence
|
|
|
0.008
|
Country
|
258(57.8%)
|
298(66.5%)
|
|
City
|
188(42.2%)
|
150(33.5%)
|
|
Occupation
|
|
|
0.231
|
Farmers and Workers
|
148(33.2%)
|
132(295%)
|
|
Others
|
298(66.8%)
|
316(66.8%)
|
|
Tobacco smoking
|
|
|
0.001
|
No
|
259(58.1%)
|
307(68.5%)
|
|
Yes
|
187(41.9%)
|
141(31.5%)
|
|
Alcohol drinking
|
|
|
<0.001
|
No
|
294(65.9%)
|
349(77.9%)
|
|
Yes
|
152(34.1%)
|
99(22.1%)
|
|
Tumor history
|
|
|
<0.001
|
No
|
374(83.9%)
|
413(92.2%)
|
|
Yes
|
72(16.1%)
|
35(7.8%)
|
|
Oral hygiene score
|
|
|
<0.001
|
0-2
|
79(17.7%)
|
167(37.3%)
|
|
3-5
|
257(57.6%)
|
248(55.4%)
|
|
6-8
|
110(24.7%)
|
33(7.4%)
|
|
Identification of fatty acid patterns
By applying PCA, we found 4 principal components which could explain 75.7% of the variance of the dietary fatty acid consumption, as the scree plot was shown in Figure 1. And we chose the first principal component because it explained 33.2% of the variation of 32 fatty acids. The components is characterized by saturated fatty acid (the “SFA” pattern), which mainly included octanoic acid (C8:0), undecanoic acid (C11:0), lauric acid (C12:0), myristic acid (C14:0), pentacarbonate (C15:0), and (C16:1) in MUFA. Factor loadings of the individual fatty acids in the “SFA” patterns are shown in Table 2. Additionally, correlation analysis among individual fatty acids were performed, and heatmap were derived using correlation coefficients among individual fatty acids. A similar pattern was identified in cluster analysis, as fatty acids adjacent in the tree had similar loading values (Figure 1).
Table 2 Factor loading of the individual fatty acids in the FA pattern
Type of fatty acid
|
Name
|
Loading of the FA pattern *
|
Saturated fatty acids
|
|
|
6:0
|
Caproic
|
0.334
|
8:0
|
Caprylic
|
0.731
|
10:0
|
Capric
|
0.596
|
11:0
|
Undecanoic
|
0.702
|
12:0
|
Lauric
|
0.783
|
13:0
|
Tridecanoic
|
0.618
|
14:0
|
Myristic
|
0.848
|
15:0
|
Pentadecanoic
|
0.720
|
16:0
|
Palmitic
|
0.787
|
17:0
|
Heptadecanoic
|
0.525
|
18:0
|
Stearic
|
0.792
|
19:0
|
Nonadecanoic
|
0.586
|
20:0
|
Arachidic
|
0.559
|
22:0
|
Behenic
|
0.037
|
Monounsaturated fatty acids
|
|
|
14:1
|
Myristoleic
|
0.548
|
15:1
|
Pentadecenoic
|
0.575
|
16:1
|
Palmitoleic
|
0.871
|
17:1
|
Heptadecenoic
|
0.449
|
18:1
|
Oleic
|
0.588
|
20:1
|
Eicosenoic
|
0.615
|
22:1
|
Erucic
|
-0.083
|
Polyunsaturated fatty acids
|
|
|
16:2
|
Hexadecadienoic
|
0.577
|
18:2
|
Linoleic
|
0.210
|
18:3
|
Octadecadienoic
|
0.443
|
20:2
|
Eicosadicnoic
|
0.695
|
20:3
|
Eicosatrienoic
|
0.326
|
20:4
|
Arachidonic
|
0.769
|
20:5
|
Eicosapentaenoic
|
0.387
|
22:3
|
Docosatrienoic
|
0.319
|
22:4
|
Docosatetraenoic
|
0.36
|
22:5
|
Docosapentaenoic
|
0.365
|
22:6
|
Docosahexaenoic
|
0.369
|
* We analyzed only the principal component 1 with the greatest explanatory degree and listed its factor load. This principal component explained 33.2% of the variation in all 32 fatty acids.
Additionally, we evaluated the correlations between the “SFA” pattern with intakes of nutrients and food groups, results of which was shown in Supplement Table 2. The “SFA” pattern was positively associated with intake of protein, total fat, (r=0.207, 0.368 respectively, all p<0.001), but negatively related to fiber (r=-0.185, P<0.001). As for food groups, the “SFA” pattern was positively correlated to the intakes of fish, eggs, dairy and red meat (r=0.372, 0.320, 0.283, 0.282, respectively, all p<0.05), but negatively correlated with grain and vegetables (r=-0.403, -0.100, respectively, all p<0.05).
Association of the “SFA” pattern with risk of oral cancer
Table 3 presents the OR and 95% CI for oral cancer across the tertile categories for the “SFA” pattern. A positive association between the “SFA” pattern and risk of oral cancer was observed. In the crude model, those in the highest tertile of the “SFA” pattern had an increased risk of oral cancer compared with the lowest tertile, with a significant linear trend (OR=3.054; 95% CI: 2.184–4.265; Ptrend <0.004). In model 1, after adjusting for gender, age, marital status, residence, BMI, occupation and tumor history, the individuals in the highest tertile of the “SFA” pattern tended to have higher odds for oral cancer (OR=2.874; 95%CI: 1.964-4.205; Ptrend <0.001) compared with those in the lowest tertile. In model 2, this result remained significant after further adjustment for lifestyle factors, including tobacco smoking, alcohol drinking and oral hygiene score (OR=3.325; 95% CI: 2.222-4.975; Ptrend <0.001). Moreover, this association was similarly observed when we treated the “SFA” pattern score as a continuous variable in the rude model, model 1 and model 2 (OR=1.697, 95%CI: 1.465-1.965; OR=1.652, 95%CI: 1.402-1.947; OR= 1.772, 95%CI: 1.490-2.106).
Table 3 The relation between the “SFA” pattern and oral cancer risk
Model
|
Tertiles of the “SFA” pattern score*
|
Ptrend
|
“SFA” pattern score†
|
I
|
II
|
III
|
case/control(n)
|
138/196
|
152/183
|
144/190
|
|
|
crude
|
1.0(reference)
|
1.431(1.033-1.984)
|
3.054(2.187-4.265)
|
<0.001
|
1.697(1.465-1.965)
|
model1#
|
1.0(reference)
|
1.617(1.115-2.344)
|
2.874(1.964-4.205)
|
<0.001
|
1.652(1.402-1.947)
|
model2#
|
1.0(reference)
|
1.830(1.241-2.698)
|
3.325(2.222-4.975)
|
<0.001
|
1.772(1.490-2.106)
|
*Three categories were obtained by tertiles of the fatty acid pattern score .Each participant was assigned a fatty acid pattern score.
#Multivariable-adjusted Logistic regression models. Mode l adjusted for demographic characteristics: gender, age, marital status, residence, BMI, tumor, occupation, education.
Model 2 adjusted for demographic characteristics and tobacco smoking, drinking, oral hygiene score.
† Fatty acid pattern score treated as a continuous variable in the crude and two adjusted models
Furthermore, we visualized the association between the “SFA” pattern score and the risk of oral cancer using restricted cubic splines. The risk of oral cancer increased with the increase of “SFA” pattern score. However, the risk of oral cancer was relatively flat until around -0.68 of “SFA” pattern scores and then started to increase rapidly afterwards (Pnon-linearity =0.097) (Figure 2).
Stratification analysis between the “SFA” pattern and oral cancer risk
Associations between the “SFA” pattern and oral cancer risk were stratified by demographic characteristics and life style exposure factors, results of which was shown in Figure 3. Positive association between oral cancer risk and the “SFA” pattern was observed in all subgroups except for the lower oral hygiene score group. There was no evidence of effect modification by sex, tobacco smoking, alcohol drinking, or oral hygiene score (Pheterogeneity>0.1). The association varied across age (Figure 3; I2=87.8%, Pheterogeneity=0.004). The interaction was further tested by multiplying the “SFA” pattern score with age, and a multiplicative interaction was observed (Pinteraction < 0.001).