Patient characteristics
Of the 359 participants with lung nodules who were considered for the study, 48 were excluded from the final analysis (22 had no CT imaging or tumor biomarkers or were lost to follow-up, 21 had no definite diagnosis, 5 had lung metastasis). Finally, 311 patients with a definite diagnosis were included in our analysis, and the median follow-up time was 18 months. Of these patients, 129 were categorized as having malignant nodules, 173 were male, and the mean age was 57 years old. The pathology of the lung nodules is shown in Table 1; nine participants in the benign group had a clinical diagnosis.
Table 1
The pathology of lung nodules
Pathology of lung nodule
|
Frequency
|
Percent (%)
|
Benign
|
Tuberculosis
|
29
|
22.5
|
pneumomycosis
|
9
|
7.1
|
Chronic inflammation
|
51
|
39.5
|
Organized pneumonia
|
2
|
1.6
|
hamartoma
|
7
|
5.4
|
pulmonary sclerosing pneumocytoma
|
1
|
0.8
|
granuloma
|
6
|
4.7
|
|
Other inflammation
|
15
|
11.6
|
|
Total
|
129
|
100
|
Malignant
|
squamous carcinoma
|
22
|
12.1
|
|
adenocarcinoma
|
130
|
71.3
|
|
Small cell lung cancer
|
12
|
6.6
|
|
Large cell carcinoma
|
1
|
.5
|
|
Adenoid cystic carcinoma
|
5
|
2.7
|
|
Undifferentiated carcinoma
|
12
|
6.6
|
|
Total
|
182
|
100.0
|
Univariate and multivariate logistic regression analysis
As shown in Table 2, age, smoking history, and family history of cancer, nodule number, size, shape, margin, burr sign, vascular notch, lobulation sign, crest, pleural traction sign, mediastinal/hilar lymphadenopathy, calcification, CEA, CA125, and CYFRA211 were statistically different between the benign and malignant groups, based on univariate analysis. Except for lung nodule diameter, there were no significant differences in clinical features between the training and validation sets. The mean size was 18.49 ± 7.37 in the training set and 16.42 ± 8.70 in the validation set, which is a minimal difference in clinical practice.
Table 2
Comparison of clinical features between benign and malignant groups in the derivation set
Variable
|
Total
|
benign (%)
|
malignant (%)
|
Wald X2
|
p
|
Gender
|
212
|
92
|
120
|
1.349
|
0.291
|
male
|
125
|
58 (63.0)
|
67 (55.8)
|
|
|
female
|
87
|
34 (37.0)
|
53 (44.2)
|
|
|
Age
|
212
|
51.83±13.10
|
61.42±10.56
|
25.160
|
0.000
|
Smoking
|
67
|
20 (21.7)
|
47 (39.2)
|
7.148
|
0.008
|
Drinking
|
36
|
12 (14.3)
|
24 (20.7)
|
1.354
|
0.245
|
Family history of cancer
|
14
|
3 (3.3)
|
11 (9.2)
|
2.704
|
0.100
|
Number
|
212
|
92
|
120
|
13.282
|
0.000
|
1
|
144
|
50 (54.3)
|
94 (78.3)
|
|
|
≥2
|
68
|
42 (45.7)
|
26 (21.7)
|
|
|
Type
|
212
|
92
|
120
|
0.734
|
0.392
|
solid
|
152
|
69 (75.0)
|
83 (69.2)
|
|
|
GGO
|
34
|
13 (14.1)
|
21 (17.5)
|
|
|
subsolid
|
26
|
10 (10.9)
|
16 (13.3)
|
|
|
Size
|
212
|
16.46±7.77
|
20.04±6.66
|
11.832
|
0.001
|
Shape
|
212
|
92
|
120
|
9.790
|
0.002
|
circle
|
58
|
33 (35.9)
|
25 (20.8)
|
|
|
oval
|
49
|
25 (27.2)
|
24 (23.1)
|
|
|
unregular
|
105
|
34 (37.0)
|
71 (59.2)
|
|
|
Margin
|
212
|
92
|
120
|
1.220
|
0.269
|
clear
|
134
|
62 (67.4)
|
72 (60.0)
|
|
|
ambiguous
|
78
|
30 (32.6)
|
48 (36.8)
|
|
|
Cavity
|
28
|
11 (12.0)
|
17 (14.2)
|
0.221
|
0.638
|
Vascular notch
|
147
|
42 (45.7)
|
105 (87.5)
|
37.463
|
0.000
|
Lobulation
|
79
|
9 (9.8)
|
70 (58.3)
|
41.563
|
0.000
|
Burr sign
|
73
|
13 (14.1)
|
60 (50.0)
|
26.492
|
0.000
|
Pleural traction
|
46
|
12 (13.0)
|
34 (28.3)
|
6.862
|
0.009
|
Crest
|
38
|
4 (4.3)
|
34 (28.3)
|
15.472
|
0.000
|
Mediastinal/hilar lymphadenopathy
|
42
|
12 (13.0)
|
40 (33.3)
|
10.872
|
0.001
|
Emphysema
|
49
|
17 (18.5)
|
32 (26.7)
|
1.947
|
0.163
|
Calcification
|
21
|
16 (17.4)
|
5 (4.2)
|
8.750
|
0.003
|
CEA
|
212
|
1.99±1.71
|
69.55±6.38
|
11.185
|
0.001
|
CA125
|
206
|
14.16±11.72
|
38.87±43.70
|
8.780
|
0.003
|
CYFRA211
|
206
|
2.01±1.15
|
4.67±12.33
|
13.426
|
0.000
|
NSE
|
205
|
12.74±5.77
|
12.21±8.61
|
0.258
|
0.611
|
In the univariate regression analysis, risk factors associated with lung nodule malignancy included older age, smoking, fewer nodules, longer diameter, nonregular shape, vascular notch, lobulation, burr sign, pleural traction, crest, mediastinal/hilar lymphadenopathy, no calcification, higher CEA value, CA125, and CYFRA211. In the multivariate logistic regression analysis, number, vascular notch, mediastinal/hilar lymphadenopathy, lobulation, age, gender, CEA, CA125, and CYFRA211 were associated with lung nodule malignancy (Table 3).
Table 3
The univariate and multivariate analysis of risk factors for lung nodule malignancy
Variable
|
Univariate analysis
|
Multivariate analysis
|
|
OR
|
95% CI
|
p
|
OR
|
95% CI
|
p
|
Number
|
0.329
|
0.181-0.598
|
0.000
|
.163
|
0.058-0.453
|
0.001
|
Vascular notch
|
8.333
|
4.226-16.432
|
0.000
|
4.641
|
1.814-11.874
|
0.001
|
Lobulation
|
12.911
|
5.932-28.101
|
0.000
|
10.921
|
4.024-29.636
|
0.000
|
ML
|
3.333
|
1.630-6.819
|
0.001
|
2.974
|
1.050- 8.425
|
0.040
|
Age
|
1.073
|
1.044-1.103
|
0.000
|
1.073
|
1.025-1.124
|
0.003
|
Gender
|
1.349
|
0.774-2.353
|
0.291
|
3.277
|
1.305-8.228
|
0.011
|
CEA
|
1.309
|
1.118-1.533
|
0.001
|
1.151
|
0.939-1.410
|
0.176
|
CA125
|
1.028
|
1.009-1.046
|
0.003
|
1.039
|
1.011-1.068
|
0.006
|
CYFRA211
|
1.432
|
1.182-1.736
|
0.000
|
1.344
|
0.994-1.817
|
0.055
|
Abbreviations: OR: odds ratio, CI: confidence interval, ML: Mediastinal/hilar lymphadenopathy. |
The ZZU model and ROC curves
Regression equation of the ZZU prediction model: X= −7.619−1.187*number+1.535*vascular+2.391*lobulation+1.090*ML+0.07*age+0.140*CEA+0.038*CA125+0.295*CYFRA211+1.187*gender
P = exp (X)/(1+exp (X))
The performance of the regression equation was assessed by ROC curve analysis, as shown in Figure 2. The AUC of the ZZU model in the derivation cohort was 0.93 (95% CI 0.90–0.97) and the AUC of the ZZU model in the validation set was 0.91 (95% CI 0.85–0.98). Validation of lung nodules with mediastinal/hilar lymphadenopathy: the AUC of the model was 0.95 (95% CI: 0.90–0.99). Validation of lung nodules without mediastinal/hilar lymphadenopathy: the AUC of the model was 0.91 (95% CI: 0.87–0.95).
The cut-off value was calculated by Youden index, and the maximum of the Youden index was 0.734. The predicted cut-off value was 0.6279, the sensitivity was 83.2%, the specificity was 90.2%, the positive predictive value (PPV) was 88.2%, the negative predictive value was 83.7%, and the accuracy was 86.3%. The Hosmer-Lemeshow goodness-of-fit test result was 0.221 (p > 0.05). The expected value and observed values fit well, as shown in Figure 3.
Nomogram of clinical prediction model for lung nodules
The nomogram of the prediction model for malignancy of lung nodules is shown in Figure 4, and eight independent variables were selected and visualized. We calculated the malignant risk of lung nodules based on the variables.