A total of 114 patients (78 male and 36 female patients) with IEEs were included in this study (median age:22.0 years, IQR: 10-44 years). There were 80 and 34 patients, respectively, drawn at random for the training and validation groups. There were no significant differences in the clinicopathologic characteristics between the training and the validation groups (all p values >0.05) (Table 1). The range for DFS was 1.00–178.70 months and the median time was 30.10 months. The range for OS was 3.40–183.20 months and the median time was 35.65 months. Figure 1 presents the diagram for study flow.
Table 1. Demographic, clinical and MR-VASARI (Visually Accessible Rembrandt Images) features of patients with intracranial extraventricular ependymoma (IEE).
Characteristic
|
Training group (n=80)
|
Validation group(n=34)
|
P-value
|
Demographics characteristics
|
|
|
|
Gender, n (%)
|
|
|
0.38
|
Male
|
57(71.25%)
|
21(61.76%)
|
|
Female
|
23(28.75%)
|
13(38.24%)
|
|
Age [median (IQR), years]
|
23.50(12.00~47.75)
|
13.00(5.00~37.75)
|
0.06
|
Clinical characteristics [mean,(SD)]
|
|
|
|
Eos
|
0.16(0.19)
|
0.23(0.36)
|
0.25
|
BUN
|
3.68(1.87)
|
3.63(1.58)
|
0.93
|
Scr
|
65.00(23.12)
|
60.81(35.65)
|
0.07
|
D-dimer
|
0.31(0.65)
|
0.34(0.55)
|
0.83
|
DFS, n (%)
|
|
|
0.55
|
0
|
42(52.50%)
|
20(58.82%)
|
|
1
|
38(47.50%)
|
14(41.18%)
|
|
OS, n (%)
|
|
|
0.13
|
0
|
67(83.75%)
|
24(70.59%)
|
|
1
|
13(16.25%)
|
10(29.41%)
|
|
VASARI features
[mean,(SD)]
|
|
|
|
F1
|
3.05(2.52)
|
3.00(2.46)
|
0.63
|
F3
|
1.61(1.14)
|
1.79(1.27)
|
0.98
|
F10
|
1.58(0.55)
|
1.68(0.59)
|
0.62
|
F13
|
2.71(0.48)
|
2.65(0.65)
|
0.10
|
Significant at P<0.05.
Note: F1, tumor locations; F3, eloquent brain; F10, T1/FLAIR ratio; F13, definition of the non-enhancing margin. In DFS and OS, 0 represents that no endpoint event has occurred, while 1 represents that an endpoint event has occurred.
Abbreviations: Eos, Eosinophil; BUN, blood urea nitrogen; Scr, Serum creatinine; OS,overall survival; DFS, disease-free survival.
Figure 1. Flowchart for the study cohort of patients with intracranial extraventricular ependymoma (IEE).
In the training group, multivariate Cox proportional hazards regression analysis identified the clinical variables such as eosinophil, blood urea nitrogen, serum creatine, and MR-VASARI feature such as F13 as significant prognostic factors for DFS (Table 2). The factors including blood urea nitrogen, D-dimer, F1, F3, and F10 were significant predictors for OS(p<0.05) (Table 3).
Table 2. Multivariate Cox proportional hazards regression analysis of clinical and MR-VASARI features for disease-free survival rate in patients with intracranial extraventricular ependymoma.
Variable
|
Hazard Ratio [95% CI]
|
P-value
|
Clinic
|
|
|
Eos
|
11.62 [1.94,69.66]
|
0.007
|
BUN
|
0.76 [0.57,1.02]
|
0.068
|
Scr
|
0.99 [0.97,1.01]
|
0.182
|
VASARI
|
|
|
F13
|
0.56 [0.33,0.95]
|
0.030
|
Significant at P<0.05.
Note: Data are hazard ratio estimates; 95% CI in parentheses for variables included in Cox regression models.
Abbreviations: Eos, Eosinophil; BUN, blood urea nitrogen; Scr, Serum creatinine; VASARI, Visually Accessible Rembrandt Images; F13,definition of the non-enhancing margin.
Table 3. Multivariate Cox proportional hazards regression analysis of clinical and MR-VASARI features for overall survival rate in patients with intracranial extraventricular ependymoma.
Clinic
|
Hazard Ratio [95% CI]
|
P-value
|
BUN
|
0.60 [0.36,0.99]
|
0.046
|
D-dimer
|
1.60 [1.12,2.30]
|
0.010
|
VASARI
|
|
|
F1
|
1.45 [1.13,1.86]
|
0.004
|
F3
|
1.41 [1.00,1.99]
|
0.049
|
F10
|
5.61 [1.79,17.61]
|
0.003
|
Significant at P<0.05.
Note: Data are hazard ratio estimates; 95% CI in parentheses for variables included in Cox regression models.
Abbreviations: BUN, blood urea nitrogen; VASARI, Visually Accessible Rembrandt Images; F1, tumor locations; F3, eloquent brain; F10, T1/FLAIR ratio.
We constructed three prediction models based on clinical data, MR-VASARI feature and the two features, respectively (Table 4). The higher the c-index, the better the predictive ability of the model. For OS, the clinical prediction model had a c-index of 0.732, the MR-VASARI feature-based model had a c-index of 0.729, and model combining both clinical and MR-VASARI features had a c-index test of 0.768. For DFS, the c-indices for clinical, MR-VASARI and combined models were 0.694, 0.576, and 0.714, respectively. The results showed that the c-index of the combined model incorporating both clinical and MRI-VASARI features was the highest among the three models.
Table 4. Concordance indices (C-index) for clinical, MR-VASARI, and combined prognostic models.
|
DFS
|
OS
|
|
Training (95%CI)
|
Validation (95%CI)
|
Training (95%CI)
|
Validation (95%CI)
|
|
|
|
|
|
Clinic
|
0.694 [0.596,0.791]
|
0.679 [0.548,0.810]
|
0.734 [0.559,0.908]
|
0.732 [0.592,0.872]
|
MR-VASARI
|
0.576 [0.494,0.658]
|
0.596 [0.452,0.740]
|
0.821 [0.719,0.924]
|
0.729 [0.602,0.857]
|
Combined
|
0.714 [0.621,0.806]
|
0.683 [0.546,0.819]
|
0.848 [0.770,0.926]
|
0.768 [0.663,0.873]
|
Note: The 95% confidence interval is indicated in parentheses. OS, overall survival, DFS, disease-free survival; VASARI, Visually Accessible Rembrandt Images.
Figure 2 presents the predicted probabilities and visualized nomograms for 3-, 5-, and 7-year survival in patients with IEE. The model incorporated all factors related to survival. The nomogram plot revealed that blood urea nitrogen and eosinophil had the strongest correlations with OS and DFS, respectively. D-dimer, serum creatinine, F1, F3, F10, and F13 were the next most significant predictors of survival. Each variable was assigned a score according to its prognostic value. The total score for all variables was converted into predictions of 3-, 5-, and 7-year survival probabilities for OS and DFS. The scores for each chosen variable could be used to calculate the likelihood that the patient would survive. Calibration curves were used to graphically evaluate the nomograms and found good agreement in predicting 3-, 5-, and 7-year survival in the training and validation groups(Figure 3).
Figure 2. Nomogram predicting 3-, 5- and 7- year disease-free survival and overall survival of patients with intracranial extraventricular ependymoma (IEE).
Note:The nomogram summed the points identified on the scale for each variable. The total points projected on the bottom scales indicate the probabilities of disease-free survival (DFS) and overall survival (OS) at 3-, 5-, and 7 years. A: nomogram for disease-free survival (DFS); B: nomogram for overall survival (OS).
Abbreviations: Eos, Eosinophil; BUN, blood urea nitrogen; Scr, Serum creatinine; F1, tumor locations; F3, eloquent brain; F10, T1/FLAIR ratio; F13, definition of the non-enhancing margin.
Figure 3. Calibration curves for disease-free survival (DFS) and overall survival (OS) based on predictions of the nomogram.
Note: Calibration curves for predicting (A) DFS and (B) OS in the training and validation cohorts. Nomogram predicted survival is plotted on the x-axis and actual survival is plotted on the y-axis.
Kaplan-Meier survival curves calculated the endpoints of each endpoint event on different models separately and produced a relevant separation of OS and DFS survival curves (Figure 4). Log-rank tests were performed in DFS and OS. Only the combined model had a p-value closest to 0.05 for DFS. For OS, both the MR-VASARI features model and the combined model had a p-value <0.05. The smaller the p-value of the log-rank test, the more significant difference existed in the survival distribution of the model.
Figure 4. Kaplan-Meier survival models for disease-free survival (DFS) and overall survival (OS).
Note: A: Clinical model, B: MR-VASARI model, and C: Combined model (1: disease-free survival (DFS), 2: overall survival (OS)). Each graph contained the training group on the left and the validation group on the right. The red line indicated the patients without an endpoint event (0), and the blue line indicated the patients with an endpoint event (1). The log‐rank test was applied to test for significant separation of survival curves and calculation of p‐values. Significant separation of DFS curves was observed in the combined model (C1) in the validation group (p = 0.068). Significant separation of the survival curves for OS was observed in the MR-VASARI (B2) and combined (C2) models in the validation group (p = 0.035, p = 0.036).