Characteristics of Patients
We identified a total of 7,215 patients from the database eventually. The selecting flow diagram was presented in Fig. 1 and the characteristics of the patients was showed in Table 1.
Table 1
Characteristics of Patients with Esophageal Carcinoma after Surgery
Whole cohort (n = 7284) Training cohort (n = 5100) Validation cohort (n = 2184)
|
Variables
|
N
|
%
|
N
|
%
|
N
|
%
|
Age
|
< 50
|
683
|
9.5
|
479
|
9.5
|
204
|
9.4
|
|
50–59
|
1948
|
27.0
|
1349
|
26.7
|
599
|
27.7
|
|
60–69
|
2841
|
39.3
|
2003
|
39.6
|
838
|
38.7
|
|
70–79
|
1491
|
20.7
|
1055
|
20.9
|
436
|
20.2
|
|
≥ 80
|
252
|
3.5
|
166
|
3.3
|
86
|
4.0
|
Gender
|
Female
|
1129
|
15.6
|
809
|
16.0
|
320
|
14.8
|
|
Male
|
6086
|
84.4
|
4243
|
84.0
|
1843
|
85.2
|
Race
|
White
|
6496
|
90.0
|
4560
|
90.3
|
1936
|
89.5
|
|
Black
|
397
|
5.5
|
278
|
5.5
|
119
|
5.5
|
|
Others
|
322
|
4.5
|
214
|
4.2
|
108
|
5.0
|
Site
|
Upper
|
131
|
1.8
|
97
|
1.9
|
34
|
1.6
|
|
Middle
|
977
|
13.6
|
679
|
13.4
|
298
|
13.7
|
|
Lower
|
5656
|
78.4
|
3958
|
73.3
|
1698
|
78.5
|
|
Overlapping
|
226
|
3.1
|
164
|
3.3
|
62
|
2.9
|
|
Unknown
|
225
|
3.1
|
154
|
3.1
|
71
|
3.3
|
Histology
|
ESCC
|
1359
|
18.8
|
958
|
18.9
|
401
|
18.5
|
|
EAC
|
5357
|
74.3
|
3743
|
74.1
|
1614
|
74.6
|
|
Others
|
499
|
6.9
|
351
|
7.0
|
148
|
6.9
|
Grade
|
Well
|
478
|
6.6
|
336
|
6.6
|
142
|
6.6
|
|
Moderately
|
2798
|
38.8
|
1974
|
39.1
|
824
|
38.1
|
|
Poorly
|
3104
|
43.0
|
2153
|
42.6
|
951
|
44.0
|
|
Undifferentiated
|
94
|
1.3
|
66
|
1.3
|
28
|
1.3
|
|
Unknown
|
741
|
10.3
|
523
|
10.4
|
218
|
10.0
|
TNM stage
|
I
|
1509
|
20.9
|
1086
|
21.5
|
423
|
19.6
|
|
II
|
2367
|
32.8
|
1633
|
32.3
|
734
|
33.9
|
|
III
|
2638
|
36.6
|
1854
|
36.7
|
784
|
36.2
|
|
IV
|
701
|
9.7
|
479
|
9.5
|
222
|
10.3
|
Radiation
|
No
|
2524
|
35.0
|
1787
|
35.4
|
737
|
34.1
|
|
Yes
|
4691
|
65.0
|
3265
|
64.6
|
1426
|
65.9
|
Chemo
|
No
|
2121
|
29.4
|
1484
|
29.4
|
637
|
29.5
|
|
Yes
|
5094
|
70.6
|
3568
|
70.6
|
1526
|
70.5
|
LNR
|
0
|
4492
|
62.2
|
3187
|
63.1
|
1305
|
60.3
|
|
< 0.05
|
264
|
3.7
|
169
|
3.3
|
95
|
4.4
|
|
0.05–0.3
|
1602
|
22.2
|
1090
|
21.6
|
512
|
26.7
|
|
> 0.3
|
857
|
11.9
|
606
|
12.0
|
251
|
11.6
|
Size
|
0–20 mm
|
1373
|
19.0
|
995
|
19.7
|
378
|
17.5
|
|
21–50 mm
|
2769
|
38.4
|
1945
|
38.5
|
824
|
38.1
|
|
> 50 mm
|
1628
|
22.6
|
1100
|
21.8
|
528
|
24.4
|
|
UNK
|
1445
|
20.0
|
1012
|
20.0
|
433
|
20.2
|
In general, 6,086 patients (84.4%) were male, and 4,584 patients (63.5%) were more than 60 years old. compared with other sites, the lower esophagus was the most common site. The white race made up 90% of the entire cohort. More than half of the study population received perioperative radiation (65.0%) and chemotherapy (70.6%). Of all the histology types, EAC accounted for the largest proportion (74.3%). Nearly half of the patients were in advanced stages (III, IV) at the time of diagnosis. There were 2,723 patients (37.8%) with regional LN positive, with 2,459 patients (34.1%) LNR higher than 0.05.
A total of 3,983 deaths occurred, including 3,209 cancer-specific deaths and 774 deaths from other causes during a follow-up period of up to 155 months. The mean follow-up period was 39.8 months. The cancer-specific mortalities of 1-year, 3-year, and 5-year were 14.6%, 35.7%, 41.6%, and the competing mortalities were 3.2%, 6.2%, and 7.8% respectively
The corresponding CIF curves of competing risk model were presented in Fig. 2. For males, the cumulative mortality of EC was significantly higher than females while it did not show statistic difference as for competing causes. Compared with EAC, ESCC and other histology types had a higher cancer-specific mortality but it was inapplicable for death of competing causes. Greater cumulative mortality of EC was observed when patients presented with older age, worse differentiation, greater tumor size, later the TNM stage, later histology stage, and more positive LN. In addition, higher probability of death from competing causes was shown in other all risk factors except gender, grade, and T stage.
The OS survival curves of risk factors above were presented in Supplement Fig. 1. It was demonstrated that greater survival probability was associated with younger age, female, well differentiation (grade I), smaller tumor size, earlier TNM stage, EAC histology type, earlier histology stage, absent or less positive LN, and lower LNR.
Multivariate Cox Regression Analysis
Multivariate cox regression analysis of training cohort was conducted to identify significant risk factors, presenting in Table 2. Age, race, gender, tumor location, histology, grade, TNM stage, tumor size, radiation, chemotherapy, LN positive, LNR were taken into consideration. Perioperative radiation (HR = 1.07, 95%CI: 0.96–1.19) was not included in the nomograms because of not presenting statistical significance. Compared with the white, although the black patients suffered a higher risk of death (HR = 1.28, 95% CI: 1.12–1.46), race was not included in the nomograms because racial information was not widely applicable. Moreover, the number of positive LNs showed significant difference in predicting survival of EC after surgery, whereas it was also excluded to avoid multicollinearity.
Table 2
Multivariate Cox Analysis of the Training Cohort
Variables Overall Survival
|
|
|
HR
|
95% CI
|
P value
|
Age
|
< 50
|
Ref
|
|
|
|
|
50–59
|
1.19
|
1.05
|
1.35
|
0.005*
|
|
60–69
|
1.33
|
1.18
|
1.50
|
< 0.001*
|
|
70–79
|
1.79
|
1.57
|
2.03
|
< 0.001*
|
|
≥ 80
|
2.31
|
1.92
|
2.79
|
< 0.001*
|
Gender
|
Female
|
Ref
|
|
|
|
|
Male
|
1.21
|
1.09
|
1.32
|
< 0.001*
|
Site
|
Upper
|
Ref
|
|
|
|
|
Middle
|
0.87
|
0.69
|
1.10
|
0.241
|
|
Lower
|
0.75
|
0.60
|
0.94
|
0.013*
|
|
Overlapping
|
0.83
|
0.63
|
1.09
|
0.187
|
|
Unknown
|
0.78
|
0.59
|
1.05
|
0.105
|
Histology
|
ESCC
|
Ref
|
|
|
|
|
EAC
|
0.88
|
0.80
|
0.97
|
0.012*
|
|
Others
|
0.99
|
0.86
|
1.14
|
0.947
|
Grade
|
Well
|
Ref
|
|
|
|
|
Moderately
|
1.19
|
1.00
|
1.41
|
0.025*
|
|
Poorly
|
1.36
|
1.15
|
1.62
|
< 0.001
|
|
Undifferentiated
|
1.43
|
1.01
|
2.01
|
0.022*
|
|
Unknown
|
1.07
|
0.87
|
1.33
|
0.896
|
TNM stage
|
I
|
Ref
|
|
|
|
|
II
|
1.73
|
1.49
|
2.03
|
< 0.001*
|
|
III
|
2.23
|
1.87
|
2.66
|
< 0.001*
|
|
IV
|
2.06
|
1.65
|
2.58
|
< 0.001*
|
Size
|
0–20 mm
|
Ref
|
|
|
|
|
21–50 mm
|
1.15
|
1.04
|
1.27
|
0.006*
|
|
> 50 mm
|
1.27
|
1.14
|
1.43
|
< 0.001*
|
|
Unknown
|
1.12
|
0.98
|
1.25
|
0.102
|
Radiation
|
No
|
Ref
|
|
|
|
|
Yes
|
1.07
|
0.96
|
1.19
|
0.234
|
Chemo
|
No
|
Ref
|
|
|
|
|
Yes
|
0.67
|
0.60
|
0.75
|
< 0.001*
|
LNR
|
0
|
Ref
|
|
|
|
|
< 0.05
|
1.01
|
0.84
|
1.21
|
0.906
|
|
0.05–0.25
|
1.51
|
1.39
|
1.65
|
< 0.001*
|
|
> 0.25
|
2.47
|
2.24
|
2.71
|
< 0.001*
|
HR: Hazard ratio; CI: Confidence interval; TNM stage: tumor node metastasis stage; LNR: lymph node ratio. |
Construction and Validation of Nomograms for OS and CSS
The training cohort consisted of 70% of patients. For the training cohort, the nomograms of OS and CSS were presented in Fig. 3 and Fig. 4. Of all risk factors, age, gender, tumor site, histology, grade, TNM stage, tumor size, chemotherapy, and LNR were included. The 1-, 3-, and 5 years of predicted OS and CSS were lined at the bottom. The C-indexes of prognostic nomograms were 0.748 (95% CI: 0.738–0.758) for OS, and 0.795 (95% CI: 0.785–0.804) for CSS, which suggested a good discrimination ability for predicting OS and CSS of EC after surgery. In addition, a total of 2,184 patients (30%) were assigned into the validation cohort. The C-indexes of the nomograms were 0.752 (95% CI: 0.738–0.765) for OS and 0.804 (95% CI: 0.790–0.817) for CSS.
The efficiency of nomograms was also evaluated by calibration plots in both training cohort and validation cohort. The calibration plots of 1-, 3-, and 5-years survival probabilities were shown in Supplement Fig. 2(A-F) for the OS, and in Supplement Fig. 3(A-F) for the CSS. Calibration plots of training cohort demonstrated the optimal agreement between the prediction by nomogram and actual observation in the probability of 1-, 3-, and 5-year OS and CSS. Meanwhile, it suggested a high agreement between the predicted and actual survival in the validation cohort.
Comparison of Nomogram and TNM system
The ROC curves were carried out to compare the difference of discriminatory accuracy between the established nomograms and TNM staging system, presenting in Fig. 5. The AUC values of ROC curves for predicting 1-, 3-, and 5-year OS and CSS were calculated (Table 3). The AUC values of 1-, 3-, and 5-year OS were 0.703, 0.731, 0.736 respectively, with 0.628, 0.677, 0.688 for TNM stage. Similarly, it demonstrated that the AUC values of established nomograms were higher than the TNM staging system in terms of CSS.
Table 3
Comparison of the AUC values between nomograms and TNM stages.
|
Overall survival
|
|
Cancer-specific survival
|
|
1 year
|
3 year
|
5 year
|
|
1 year
|
3 year
|
5 year
|
Nomogram
|
0.703
|
0.731
|
0.736
|
|
0.718
|
0.745
|
0.753
|
TNM Stage
|
0.628
|
0.677
|
0.688
|
|
0.645
|
0.687
|
0.700
|