Patient selection from SEER
This study included surgical patients with DTC derived from the SEER from 2010 to 2015. A total of 17,659 were involved in our study, of which 12,363 were randomly divided to the modeling cohort while 5,296 cases were placed into the validation cohort in a ratio of 7: 3 (Fig. 1). Of the enrolled patients, 13,317 were female (75.4%), and 14,223 (80.5%) were White. Based on the optimal cutoff value in age (age ≤ 54, 55–68, and age ≥ 69), 4767 (27.0%) were between 55–68 years old. Besides, the majority of the grade is I (81.4%) while 57.5% were in the T1 stage, 75.4% were in the N0 stage and 98.5% were in the M0 stage. For the modeling and validation cohort, the median follow-up period was 43.3 (range, 0–83 months) months and 43.0 (range, 0–83 months) months, respectively. The clinicopathological features of patients enrolled in the study are summarized in Table 1.
Table 1
Clinic-pathologic characteristics of patients in the modeling and validation cohorts
Variables
|
All patients, n (%)
|
Modeling Cohort, n (%)
|
Validation Cohort, n (%)
|
P
|
Total
|
17,659 (100.0)
|
12,363 (70.0)
|
5,296 (30.0)
|
|
Age
|
|
|
|
0.25
|
≤ 54
|
10,935 (61.9)
|
7,608 (61.5)
|
3,327 (62.8)
|
|
55–68
|
4,767 (27.0)
|
3,378 (27.3)
|
1,389 (26.2)
|
|
≥ 69
|
1,957 (11.1)
|
1,377 (11.1)
|
580 (11.0)
|
|
Race
|
|
|
|
0.71
|
White
|
14,223 (80.5)
|
9,941 (80.4)
|
4,282 (80.9)
|
|
Black
|
1,352 (7.7)
|
947 (7.7)
|
405 (7.6)
|
|
Other*
|
2,084 (11.8)
|
1,475 (11.9)
|
609 (11.5)
|
|
Sex
|
|
|
|
0.22
|
Male
|
4,342 (24.6)
|
3,008 (24.3)
|
1,334 (25.2)
|
|
Female
|
13,317 (75.4)
|
9,355 (75.7)
|
3,962 (74.8)
|
|
Grade
|
|
|
|
0.56
|
I
|
14,375 (81.4)
|
10,069 (81.4)
|
4,306 (81.3)
|
|
II
|
2,604 (14.7)
|
1,812 (14.7)
|
792 (14.9)
|
|
III
|
539 (3.1)
|
376 (3.0)
|
163 (3.1)
|
|
IV
|
141 (0.8)
|
106 (0.9)
|
35 (0.7)
|
|
AJCC T stage(7th )
|
|
|
|
0.87
|
T1
|
10,147 (57.5)
|
7,111 (57.5)
|
3,036 (57.3)
|
|
T2
|
3,028 (17.1)
|
2,128 (17.2)
|
900 (17.0)
|
|
T3
|
3,810 (21.6)
|
2,648 (21.4)
|
1,162 (21.9)
|
|
T4
|
674 (3.8)
|
476 (3.9)
|
198 (3.7)
|
|
AJCC N stage(7th )
|
|
|
|
0.20
|
N0
|
13,319 (75.4)
|
9,358 (75.7)
|
3,961 (74.8)
|
|
N1
|
4,340 (24.6)
|
3,005 (24.3)
|
1,335 (25.2)
|
|
AJCC M stage(7th )
|
|
|
|
0.28
|
M0
|
17,399 (98.5)
|
12,173 (98.5)
|
5,226 (98.7)
|
|
M1
|
260 (1.5)
|
190 (1.5)
|
70 (1.3)
|
|
*, American Indian & AK Native & Asian &Pacific Islander; Grade I: Well differentiated; Grade II: Moderately differentiated; Grade III: Poorly differentiated; Grade IV: Undifferentiated or anaplastic. |
Independent prognostic factors in the modeling cohort
Clinic-pathologic characteristics, including age, race, sex, grade, AJCC TNM stage, survival time, and vital status, were obtained from the training set. Age, race, sex, grade, AJCC T stage, AJCC N stage, and AJCC M stage were significantly identified in the modeling cohort (P < 0.05, Table 2), Multivariate analysis showed the following important risk factors for survival: age 55–68 (HR = 3.197, P < 0.001 vs age ≤ 54), age ≥ 69 (HR = 8.522, P < 0.001 vs age ≤ 54), male (HR = 1.269, P = 0.012 vs female), grade III (HR = 3.120, P < 0.001 vs grade), grade IV (HR = 6.593, P < 0.001 vs grade I), AJCC stage T4 (HR = 2.387, P < 0.001 vs AJCC stage T1), AJCC stage N1 (HR = 1.337, P = 0.006 vs AJCC stage N0), and AJCC stage M1 (HR = 3.475, P < 0.001 vs AJCC stage M0). Multivariate analysis revealed that these 7 variables are significantly related to OS, showing as independent predictive factors of survival (P < 0.05; Table 2).
Table 2
Univariate and multivariate analysis of OS in the modeling (n = 12,363)
Variables
|
Univariate analysis
|
Multivariate analysis
|
P value
|
HR (95% Cl)
|
P value
|
Age
|
<0.001
|
|
|
≤54
|
|
Reference
|
|
55–68
|
|
3.197 (2.507–4.076)
|
<0.001
|
≥69
|
|
8.522 (6.689–10.858)
|
<0.001
|
Race
|
0.008
|
|
|
Black
|
|
Reference
|
|
White
|
|
0.491 (0.370–0.651)
|
<0.001
|
Other
|
|
0.558 (0.384–0.811)
|
0.002
|
Sex
|
<0.001
|
|
|
Female
|
|
Reference
|
|
Male
|
|
1.269 (1.054–1.528)
|
0.012
|
Grade
|
<0.001
|
|
|
I
|
|
Reference
|
|
II
|
|
1.142 (0.886–1.472)
|
0.305
|
III
|
|
3.120 (2.338–4.163)
|
<0.001
|
IV
|
|
6.593 (4.510–9.637)
|
<0.001
|
AJCC T stage(7th )
|
<0.001
|
|
|
T1
|
|
Reference
|
|
T2
|
|
1.038 (0.773–1.394)
|
0.804
|
T3
|
|
1.162 (0.910–1.484)
|
0.229
|
T4
|
|
2.387 (1.717–3.318)
|
<0.001
|
AJCC N stage(7th )
|
<0.001
|
|
|
N0
|
|
Reference
|
|
N1
|
|
1.337 (1.085–1.649)
|
0.006
|
AJCC M stage(7th )
|
<0.001
|
|
|
M0
|
|
Reference
|
|
M1
|
|
3.475 (2.629–4.595)
|
<0.001
|
HR: hazard ratio; Cl: confidence interval. |
Construction of a prognostic nomogram
Based on the 7 Clinicopathological features, we developed a prognostic nomogram to predict the 3- and 5-year OS (Fig. 2). The nomogram shows that age is most closely related to prognosis, followed by the grade, AJCC M stage, AJCC T stage, race, AJCC N stage, and sex. We can simply and intuitively calculate the survival probability of a single patient through the cumulative score of each selected variable (Table 3).
Table 3
Detailed scores for all variables in nomogram
Variables
|
Nomogram Score
|
Age
|
|
≤54
|
0
|
55–68
|
54
|
≥69
|
100
|
Race
|
|
Black
|
33
|
White
|
0
|
Other
|
6
|
Sex
|
|
Female
|
0
|
Male
|
11
|
Grade
|
|
I
|
0
|
II
|
6
|
III
|
53
|
IV
|
88
|
AJCC T stage(7th )
|
|
T1
|
0
|
T2
|
2
|
T3
|
7
|
T4
|
41
|
AJCC N stage(7th )
|
|
N0
|
0
|
N1
|
14
|
AJCC M stage(7th )
|
|
M0
|
0
|
M1
|
58
|
Validation of a prognostic nomogram
The OS nomogram was validated both internally (model cohort) and externally (validation cohort). For the model validation, the C-index of the OS nomogram was 0.829 (95% CI: 0.807–0.851), and the C-index of the external validation was 0.833 (95% CI: 0.803–0.862). Calibration plots of the nomogram close to the dotted line showed the good consistency between the predicted 3-, 5-year survival rates and the actual observations in the modeling and validation groups (Fig. 3).