3.1. Basic information for patients with ESCC
A significant difference was observed in the number of patients with a history of alcohol consumption (40% of the N+ group and 22% of the N0 group) (P=0.011). A significant difference was also observed in terms of the BMI between the two groups [N+: 20.74 (18.96 - 23.27), N0: 22.31 (19.36 - 25.14), P=0.031)]. No statistical differences were observed in terms of the other clinical characteristics, such as sex, history of diabetes, and age (P > 0.05). (Table. 1).
Table. 1. Basic information for patients with oesophageal squamous carcinoma
Variable
|
N+ (n=60)
|
N0 (n=123)
|
P
|
Age(years)
|
67 (64-72.75)
|
69 (64-73)
|
0.320
|
Sex[cases(%)]
|
|
|
0.800
|
Male
|
44 (73.3)
|
88 (71.5)
|
|
Female
|
16 (26.7)
|
35 (28.5)
|
|
Diabetes[cases(%)]
|
4 (6.7)
|
13 (93.3)
|
0.393
|
Smoking[cases(%)]
|
31 (51.7)
|
66 (53.7)
|
0.800
|
Alcohol[cases(%)]
|
24 (40)
|
27 (22)
|
0.011
|
Body mass index (kg/m2)
|
20.74 (18.96-23.27)
|
22.31 (19.36-25.14)
|
0.031
|
3.2. Tumour information
A comparative analysis between the two groups of patients yielded a statistically significant difference between superficial infiltration (T1-T2) in 43.3% of the N+ group and 67.5% of the N0 group, and deep infiltration (T3-T4) in 56.7% of the N+ group and 32.5% of the N0 group (P=0.002). However, no significant difference was observed in terms of the tumour location and the degree of tumour differentiation between the two groups of patients (P > 0.05). (Table. 2).
Table. 2. Tumour information
Variable
|
N+ (n=60)
|
N0 (n=123)
|
P
|
Histological differentiation
[cases(%)]
|
|
|
0.214
|
Well
|
5(8.3)
|
17(13.8)
|
|
Moderate
|
40(66.7)
|
87(70.7)
|
|
Poor
|
15(25.0)
|
19(15.4)
|
|
T stage[cases(%)]
|
|
|
0.002
|
T1-T2
|
26(43.3)
|
83(67.5)
|
|
T3-T4
|
34(56.7)
|
40(32.5)
|
|
3.3. Laboratory test indicators
The laboratory indices of the two groups of patients were analysed using the independent sample t-test. Statistically significant differences were observed in terms of RDW [N+:0.81 (0.75-0.92), N0: 0.78 (0.73-0.84), P=0.033) ] and prealbumin (N+: 225.65±69.70, N0: 251.38±72.12, P=0.023). No significant differences were observed in terms of the remaining laboratory indices (P>0.05). (Table. 3).
Table. 3. Laboratory test indicators
Variable
|
N+ (n=60)
|
N0 (n=123)
|
P
|
|
WBC
|
5.74 (4.74-6.64)
|
5.52 (4.71-6.80)
|
0.934
|
|
MCV
|
94.55 (90.48-97.80)
|
93.10 (90.50-97.70)
|
0.736
|
|
RDW
|
0.81 (0.75-0.92)
|
0.78 (0.73-0.84)
|
0.033
|
|
LYM
|
1.31 (0.93-1.77)
|
1.34 (1.08-1.67)
|
0.726
|
|
NEU
|
3.67 (2.65-4.61)
|
3.60 (2.88-4.87)
|
0.781
|
|
MON
|
0.38 (0.30-0.43)
|
0.35 (0.28-0.43)
|
0.231
|
|
MCH
|
31.45 (30.20-32.48)
|
31.50 (30.40-32.50)
|
0.853
|
|
PLT
|
181.25±53.10
|
182.52±59.37
|
0.936
|
|
MPV
|
9.75 (8.92-12.20)
|
10.10 (9.20-11.50)
|
0.162
|
|
PDW
|
16.15 (15.90-16.58)
|
16.20 (15.90-16.50)
|
0.999
|
|
NLR
|
2.89 (1.80-4.26)
|
2.78 (1.94-3.75)
|
0.738
|
|
PLR
|
125.74 (101.26-196.93)
|
128.70 (104.85-171.57)
|
0.745
|
|
Albumin
|
40.10 (36.03-42.88)
|
41.40 (38.40-43.40)
|
0.048
|
|
Lactic acid
|
2.00 (1.51-2.70)
|
1.80 (1.40-2.30)
|
0.312
|
|
β-microgiobulin
|
1.90 (1.40-2.50)
|
1.90 (1.40-2.30)
|
0.613
|
|
Prealbumin
|
225.65±69.70
|
251.38±72.12
|
0.023
|
|
α1-globulin
|
4.00 (3.70-4.90)
|
4.00 (3.60-5.00)
|
0.833
|
|
α2-globulin
|
9.05 (8.10-10.18)
|
9.30 (8.30-10.30)
|
0.563
|
|
β1-globulin
|
5.80 (5.40-6.50)
|
5.70 (5.40-6.10)
|
0.208
|
|
β2-globulin
|
4.90 (4.30-5.48)
|
4.80 (4.20-5.40)
|
0.834
|
|
γ-globulin
|
17.10 (15.55-19.40)
|
17.10 (15.50-19.00)
|
0.923
|
|
Abbreviations: WBC, white blood cell; MCV, mean corpuscular volume; RDW, red blood cell distribution width; LYM, lymphocyte; NEU, neutrophile granulocyte; MON, mononuclear cell; MCH, mean corpuscular hemoglobin; PLT, platelet; MPV, mean platelet volume; PDW, platelet distribution width; NLR, neutrophil to lymphocyte ratio; PLR, platelet to lymphocyte ratio.
3.4.Univariate and multivariate logistic regression results
The t-test and chi-square test: revealed a statistical difference in terms of the following factors: history of alcohol consumption, BMI, the depth of tumour infiltration, RDW, and prealbumin. These factors were included in the univariate and multivariate logistic regression analyses, which revealed that a history of alcohol abuse (odds ratio [OR]=2.143, 95%confidence interval [CI]=1.059-4.335, P=0.034), the depth of tumour infiltration (OR= 2.816, 95%CI=1.444-5.439, P=0.002), and RDW (OR=24.749, 95%CI=1.600-382.792, P= 0.022) were independent risk factors for lymph node metastasis in ESCC. (Table. 4).
Table. 4. Univariate and multivariate analysis of lymph node metastasis in patients with oesophageal squamous carcinoma
|
|
Univariate
|
|
Multivariate
|
|
OR
|
95%CI
|
P
|
|
OR
|
95%CI
|
P
|
Alcohol
|
2.370
|
1.213-4.633
|
0.012
|
|
2.143
|
1.059-4.335
|
0.034
|
BMI
|
0.898
|
0.816-0.988
|
0.027
|
|
|
|
|
Prealbumin
|
0.995
|
0.990-0.999
|
0.025
|
|
|
|
|
T-stage
|
2.713
|
1.438-5.120
|
0.002
|
|
2.816
|
1.444-5.439
|
0.002
|
RDW
|
17.901
|
1.269-252.527
|
0.033
|
|
24.749
|
1.600-382.792
|
0.022
|
Albumin
|
-
|
-
|
0.077
|
|
|
|
|
Abbreviation: OR, odds ratio; CI, confidence interval; BMI, body mass index; RDW, red cell distribution width
3.5. Risk prediction model construction
The results of the multifactor logistic regression analysis revealed that; the regression equation was logit (P) = -4.00 + 0.783 × (history of alcohol consumption) + 0.983 × (depth of tumour infiltration) + 3.209 × (RDW), and the prediction model was expressed in the form of a column line graph. The ROC curve of this model was plotted according to the predicted probability, and the AUC was 0.700 (95%CI=0.619-0.782, P<0.001), with sensitivities and specificities of 58.3% and 78.0%, respectively. (Fig. 1).
3.6. Assessing the clinical usefulness of the model
When the model predicted the probability of lymph node metastasis in ESCC (hereafter referred to as the threshold probability) as 15-48%, the decision curves were higher than the two extreme lines (as shown in the figure), indicating that when the threshold probability was 15-48%, the patients under the model prediction benefitted from the corresponding intervention, thus having clinical value. (Fig. 2 and 3).
3.7. Internal validation of the model
The calibration curve was plotted after the Bootstrap method was used to draw calibration curves for 1000 internal validations. The model still possessed some discriminatory ability, and its prediction curve remained consistent with the actual incidence curve, which had a calibrated C-index of 0.683. (Fig. 4).