Patients’ Characteristics
3861 US patients extracted from the SEER database were randomly divided into training cohort (n = 2702) and a validation cohort (n = 1159) with a ratio of 7:3. The median age at diagnosis in the training and validation cohorts was 62 years (interquartile range, 53-69 years) and 60 years (interquartile range, 51-69 years), respectively. The majority of the patients were white (69.2 and 71.2%), married (47.6 and 48.3%), and insured (95.4 and 96.3%). Most of the tumors were at pathological Grade III or IV and bigger than 50 mm in both cohorts. Nearly half of the patients were histologically diagnosed with carcinosarcoma. The distribution of different SEER stages was close to agreement with a little higher rate in localized group (37.4 and 42.5%). Nearly half of the patients were in AJCC stage I (41.0 and 45.1%) and only less than one tenth were in AJCC stage II (9.7 and 8.1%). Most of the patients received surgery (90.8 and 92.4%), with a few receiving radiotherapy and over half receiving chemotherapy in both cohorts.
The summary of these clinical characteristics was listed in table 1.
Table 1
Patient characteristics in the study
Variable
|
Training Cohort
(n = 2702)
|
Validation Cohort
(n = 1159)
|
Medium age at diagnosis,
(25th-75th percentile)
|
62 (53-69)
|
60 (51-69)
|
Race n (%)
|
|
|
White
|
1869 (69.2)
|
808 (71.2)
|
Black
|
591 (22.1)
|
243 (20.2)
|
Other
|
242 (8.6)
|
108 (8.6)
|
Marital status n (%)
|
|
|
Married
|
1257 (47.6)
|
561 (48.3)
|
Single
|
586 (22.3)
|
261 (22.7)
|
SDW
|
736 (25.6)
|
293 (25.5)
|
Unknown
|
123 (4.6)
|
44 (3.5)
|
Insurance record n (%)
|
|
|
Yes
|
2584 (95.4)
|
1117 (96.3)
|
No
|
118 (4.6)
|
42 (3.7)
|
Tumor size n (%)
|
|
|
≤ 50 mm
|
710 (23.1)
|
313 (23.8)
|
> 50 mm
|
1596 (61.4)
|
677 (61.6)
|
Unknown
|
396 (15.6)
|
169 (14.6)
|
Pathological grade n (%)
|
|
|
I
|
169 (6.4)
|
76 (6.0)
|
II
|
345 (12.0)
|
135 (11.4)
|
III
|
1193 (43.0)
|
527 (44.5)
|
IV
|
995 (38.7)
|
421 (38.2)
|
Histological type n (%)
|
|
|
Sarcoma
|
76 (4.3)
|
20 (2.8)
|
Leiomyosarcoma
|
532 (28.3)
|
215 (26.6)
|
Adenosarcoma
|
124 (4.4)
|
45 (3.8)
|
Stromal sarcoma
|
455 (14.1)
|
207 (15.5)
|
Carcinosarcoma
|
1515 (48.8)
|
672 (51.3)
|
SEER stage n (%)
|
|
|
Localized
|
1116 (37.4)
|
537 (42.5)
|
Regional
|
825 (29.1)
|
321 (28.2)
|
Distant
|
761 (33.5)
|
301 (29.3)
|
AJCC stage n (%)
|
|
|
I
|
1225 (41.0)
|
573 (45.1)
|
II
|
236 (9.7)
|
84 (8.1)
|
III
|
572(20.0)
|
222 (20.0)
|
IV
|
669 (29.3)
|
280 (26.9)
|
Surgery status n (%)
|
|
|
Yes
|
2516 (90.8)
|
1090 (92.4)
|
No/Unknown
|
186 (9.2)
|
69 (7.6)
|
Radiotherapy status n (%)
|
|
|
Yes
|
782 (26.3)
|
313 (24.6)
|
No/Unknown
|
1920 (73.7)
|
846 (75.4)
|
Chemotherapy status n (%)
|
|
|
Yes
|
1420 (52.4)
|
598 (53.2)
|
No/Unknown
|
1282 (47.6)
|
561 (46.8)
|
Variable Screening and Multivariate Cox Regression Analysis Results
According to the Cox stepwise regression analysis, tumors’ primary site, years of diagnosis and surgery site were excluded for further study due to no difference in the prognosis for US patients. Multivariate Cox regression analyses demonstrated that age at diagnosis [hazard ratio (HR) = 1.0116, p < 0.001], being black (HR = 1.1698, p < 0.05), single (HR = 1.2181 vs. married, p < 0.05), SDW (HR = 1.2965 vs. married, p<0.001), tumor size (> 50 mm, HR = 1.4861 vs. ≤ 50 mm, p < 0.001; tumor size unknown, HR = 1.3345, p < 0.01), higher pathology grade(Grade III,HR = 7.3773 vs. pathology grade I, p < 0.001; IV, HR = 7.0185, p < 0.001), SEER stage (regional, HR = 1.8809 vs. localized, p < 0.001; distant, HR = 2.5199, p < 0.001), and higher AJCC stage (III,HR = 1.7459 vs. AJCC stage I, p < 0.001; IV,HR = 2.2275, p < 0.001) were all independent risk factors. Meanwhile, we found that leiomyosarcoma (HR = 0.6550 vs. sarcoma, p < 0.01), carcinosarcoma (HR = 0.6099 vs. sarcoma, p < 0.01), insurance (no insurance HR = 1.4851, p < 0.01), receiving surgery therapy (no/unknown HR = 2.7559, p < 0.001), adjuvant radiotherapy (no/unknown HR = 1.3267, p < 0.001), and adjuvant chemotherapy (no/unknown HR = 1.5355, p < 0.001) were protective factors for surviving US. The results also indicated that other race, other marital status, pathology grade II, adenosarcoma, stromal sarcoma and AJCC stage II were not significant risk factors (P > 0.05). The results of multivariate Cox regression analysis were presented table 2.
Table 2
Selected variables in the SEER database by multivariate Cox regression analysis (training cohort)
Variable
|
Hazard ratio
|
95% CI
|
p value
|
Age at diagnosis
|
1.0116
|
1.0061-1.0172
|
0.000***
|
Race
|
|
|
|
White
|
Reference
|
|
|
Black
|
1.1698
|
1.0225-1.3383
|
0.022*
|
Other
|
0.8481
|
0.6698-1.0739
|
0.171
|
Marital status
|
|
|
|
Married
|
Reference
|
|
|
Single
|
1.2181
|
1.0436-1.4217
|
0.012*
|
SDW
|
1.2965
|
1.1230-1.4969
|
0.000***
|
Other
|
1.2816
|
0.9740-1.6863
|
0.076
|
Insurance record
|
|
|
|
Yes
|
Reference
|
|
|
No
|
1.4851
|
1.1367-1.9404
|
0.004**
|
Tumor size
|
|
|
|
≤ 50 mm
|
Reference
|
|
|
> 50 mm
|
1.4861
|
1.2674-1.7425
|
0.000***
|
Unknown
|
1.3345
|
1.0802-1.6487
|
0.007**
|
Pathological grade
|
|
|
|
I
|
Reference
|
|
|
II
|
1.3135
|
0.7662-2.2518
|
0.321
|
III
|
7.3773
|
4.5441-11.9771
|
0.000***
|
IV
|
7.0185
|
4.3528-11.3165
|
0.000***
|
Histological type
|
|
|
|
Sarcoma
|
Reference
|
|
|
Leiomyosarcoma
|
0.6550
|
0.4839-0.8866
|
0.006**
|
Adenosarcoma
|
0.6288
|
0.3919-1.0089
|
0.054
|
Stromal sarcoma
|
0.9274
|
0.6623-1.2984
|
0.661
|
Carcinosarcoma
|
0.6099
|
0.4535-0.8202
|
0.001**
|
SEER stage
|
|
|
|
Localized
|
Reference
|
|
|
Regional
|
1.8809
|
1.3831-2.5580
|
0.000***
|
Distant
|
2.5199
|
1.6708-3.8005
|
0.000***
|
AJCC stage
|
|
|
|
I
|
Reference
|
|
|
II
|
1.0136
|
0.7096-1.4478
|
0.941
|
III
|
1.7459
|
1.2843-2.3735
|
0.000***
|
IV
|
2.2275
|
1.4818-3.3484
|
0.000***
|
Surgery status
|
|
|
|
Yes
|
Reference
|
|
|
No/Unknown
|
2.7559
|
2.2503-3.3751
|
0.000***
|
Radiotherapy status
|
|
|
|
Yes
|
Reference
|
|
|
No/Unknown
|
1.3267
|
1.1567-1.5216
|
0.000***
|
Chemotherapy status
|
|
|
|
Yes
|
Reference
|
|
|
No/Unknown
|
1.5355
|
1.3509-1.7453
|
0.000***
|
SDW Separated, divorced, and widowed, SEER Surveillance, Epidemiology, and End Results, HR hazard ratio, AJCC American Joint Committee on cancer
*p < 0.05, **p < 0.01, ***p < 0.001
|
Nomogram Construction
The nomogram was constructed for predicting the 1-, 3-, and 5-year CSS based on the multivariate Cox regression analysis (Figure 2). The pathological grade was set as reference scale ranging from 0 to 100 because it had the largest coefficient absolute value. The probabilities of 1-, 3- and 5-year CSS could be easily calculated by adding the points value of each variable. .As shown in Fig.2, pathological grade contributed the greatest influence to CSS, followed by age at diagnosis, surgery status, SEER stage, AJCC stage, histological grade, chemotherapy, insurance record, tumor size, race, radiotherapy, and finally marital status.
Nomogram Comparison and Evaluation
The C-indexes in both training and validation cohorts were higher than the AJCC 7th edition staging system (0.796 vs. 0.706, 0.767 vs. 0.713), indicating that our new model showed better discriminative ability. Furthermore, AUCs of the training cohort for the 1-,3-,5-year CSS demonstrated superior were significantly larger (0.842, 0.845,0.860, respectively) than the traditional system (0.755, 0.772, and 0.774, respectively). Likewise, the ROC values were also significantly larger (0.833, 0.798, 0.797 at 1-,3-,5-year vs0.763, 0.741, and 0.747, respectively) (Fig. 3).
Validation and Calibration of the Nomogram
The NRI values for the 1-, 3- and 5-year CSS rates in the training cohort were 64.6% (95% confidence interval [CI] = 55.3-73.5%), 59.0% (95% CI = 50.7%-67.9%) and 62.2% (95% CI = 52.8-71.4%), respectively, and in the validation cohort, 47.2% (95% CI = 25.0-63.1%), 37.6% (95% CI = 14.1-51.4%) and 29.9% (95% CI = 7.4-55.0%), respectively. Moreover, the IDI values for the 1-, 3- and 5-year CSS rates in the training cohort were 8.64, 9.63 and 9.50%, respectively, and 3.98, 5.79 and 5.88% in the validation cohort (all p < 0.001). These results all indicated that the nomogram predicted prognosis with greater accuracy than that of the AJCC model.
Calibration plots which predicting the 1-, 3- and 5-year CSS for the training and validation cohorts were nearly identical to the practical observations, which demonstrated that the new model had great calibration ability (Fig. 4).
Clinical Usefulness
Finally, DCA curve was used to evaluate the clinical usability and benefits of the nomogram. Compared to the AJCC staging system, all the DCA curves showed larger net benefits of the new model in both the training and validation cohorts (Fig. 5). These results demonstrated that the new model had better clinical benefits than the AJCC staging system.