3.1General condition of critically ill patients
3.1.1Sample characteristic
A total of 634 cases took part in the study, including 600 effective cases and 75 cases of LEDVT, with an incidence of 12.50%. The study subjects included 315 males (52.50%) and 285 females (47.50%) The minimum age of the Non-LEDVT group was 24 years old, the oldest was 93 years old, and the average age was 65.14±14.25 years old. During admission to the ICU, patients with LEDVT mainly had respiratory, neurological, circulatory and digestive diseases. Among patients with respiratory system dysfunction as the main disease, the incidence of LEDVT was 36.00%, the nervous system was 29.30%, and the circulatory system was 16.00%. Digestive system was 5.30%, other system damage was 13.30%. The main diseases of the patient's respiratory system in this study included type II respiratory failure, chronic obstructive pulmonary disease, severe pneumonia and falling pneumonia. The nervous system was mainly dominated by cerebral hemorrhage and ischemic diseases. The circulatory system included acute and chronic heart failure, coronary atherosclerotic heart disease, cardiac arrest, and arrhythmia. The digestive system was dominated by gastrointestinal bleeding, abnormal liver function and severe pancreatitis. There were 34 patients with deep vein thrombosis of the left lower extremity, 26 patients with deep vein thrombosis of the right lower extremity, and 15 patients with bilateral deep vein thrombosis, as shown in Table 1.
Table 1 Types of underlying diseases in critically ill patients and location of LEDVT
General characteristics
|
N=600
|
Type of underlying disease [n(%)]
|
|
Respiratory diseases
|
250(41.67)
|
Nervous system disease
|
180(30.83)
|
Circulatory system disease
|
104(17.33)
|
Digestive system diseases
|
23(3.83)
|
Urinary system diseases
|
4(0.67)
|
Other system diseases
|
34(5.67)
|
Location of LEDVT [n(%)]
|
|
Left lower extremity
|
34(45.33)
|
Right lower extremity
|
26(34.67)
|
Both lower extremity
|
15(20.00)
|
3.1.2 Factors of LEDVT in critically ill patients
The factors determined by univariate analysis. The results showed that APACHE-Ⅱ score, WBC, PLT, FIB, D-Dimer, DM, MV, CRRT, swollen legs, congestion heart failure (CHF), malignant tumor, central venous catheterization(CVC), major surgery> 45 minutes, family history of VTE, stroke and multiple injuries were interest between the two groups (P <0.05), as shown Table 2.
Table 2 Comparison of general information of patients in Non-LEDVT and LEDVT groups
Variable
|
Non-LEDVT (n=525)
|
LEDVT (n=75)
|
t/χ2 /Z
|
P
|
Gender(n)
|
|
|
0.0242)
|
0.877
|
male
|
275
|
40
|
|
|
female
|
250
|
35
|
|
|
Age(n)
|
|
|
5.0101)
|
0.171
|
18-40
|
110
|
14
|
|
|
41-60
|
112
|
10
|
|
|
61-74
|
134
|
18
|
|
|
≥75
|
169
|
33
|
|
|
BMI (n)
|
|
|
1.6262)
|
0.202
|
<25kg/m2
|
300
|
37
|
|
|
≥25kg/m2
|
225
|
38
|
|
|
DM (n)
|
|
|
8.1582)
|
0.004
|
yes
|
190
|
40
|
|
|
no
|
335
|
35
|
|
|
IABPM (n)
|
|
|
2.6672)
|
0.102
|
yes
|
455
|
70
|
|
|
no
|
70
|
5
|
|
|
MV (n)
|
|
|
11. 0912)
|
0.001
|
yes
|
215
|
46
|
|
|
no
|
310
|
29
|
|
|
EN (n)
|
|
|
0.6352)
|
0.426
|
yes
|
325
|
50
|
|
|
no
|
200
|
25
|
|
|
Dehydration (n)
|
|
|
0.1012)
|
0.751
|
yes
|
325
|
45
|
|
|
no
|
200
|
30
|
|
|
Sedative (n)
|
|
|
0.3882)
|
0.533
|
yes
|
300
|
40
|
|
|
no
|
255
|
35
|
|
|
Analgesic (n)
|
|
|
0.8312)
|
0.362
|
yes
|
315
|
40
|
|
|
no
|
220
|
35
|
|
|
CRRT [n (n)]
|
|
|
11.4272)
|
0.001
|
yes
|
50
|
17
|
|
|
no
|
445
|
58
|
|
|
Transfusion (n)
|
|
|
4.8181)
|
0.090
|
PTL
|
140
|
60
|
|
|
RBC
|
160
|
60
|
|
|
plasma
|
|
|
|
|
Swollen legs (n)
|
|
|
6.6972)
|
0.010
|
yes
|
78
|
20
|
|
|
no
|
447
|
55
|
|
|
Lung disease (n)
|
|
|
0.2122)
|
0.645
|
yes
|
253
|
99
|
|
|
no
|
47
|
21
|
|
|
Pulmonary dysfunction (n)
|
|
|
0.1012)
|
0.751
|
yes
|
200
|
30
|
|
|
no
|
325
|
45
|
|
|
Sepsis (n)
|
|
|
2.7803)
|
0.095
|
yes
|
20
|
6
|
|
|
no
|
505
|
69
|
|
|
CHF(n)
|
|
|
5.8122)
|
0.016
|
yes
|
56
|
15
|
|
|
no
|
479
|
60
|
|
|
Malignant tumor (n)
|
|
|
6.1712)
|
0.013
|
yes
|
80
|
20
|
|
|
no
|
445
|
55
|
|
|
CVC(n)
|
|
|
21.1452)
|
0.000
|
yes
|
160
|
43
|
|
|
no
|
365
|
32
|
|
|
Surgery>45min (n)
|
|
|
5. 4182)
|
0.020
|
yes
|
100
|
22
|
|
|
no
|
455
|
53
|
|
|
Family history of VTE (n)
|
|
|
12.0782)
|
0.001
|
yes
|
8
|
6
|
|
|
no
|
517
|
69
|
|
|
Stroke[n (n)]
|
|
|
4.0822)
|
0.043
|
yes
|
150
|
30
|
|
|
no
|
375
|
45
|
|
|
Multiple trauma
(n)
|
|
|
5.1392)
|
0.023
|
yes
|
15
|
6
|
|
|
no
|
510
|
69
|
|
|
APACHE-Ⅱ(score,±s)
|
16.00(13.00,22.00)
|
21.00(17.00,22.00)
|
-4.7033)
|
<0.001
|
WBC(10^9/L,±s)
|
9.62(6.40,13.71)
|
11.07(7.57,15.35)
|
-2.3153)
|
0.021
|
PLT(10^9/L,±s)
|
156.00(107.00,210.00)
|
172.00(121.00,287.00)
|
-2.6803)
|
0.007
|
Hb(g/L,±s)
|
101.00(82.00,125.00)
|
103.00(84.00,123.00)
|
-0.2703)
|
0.788
|
HCT(%,±s)
|
29.90(25.40, 38.75)
|
29.00(25.30,35.00)
|
-0.9273)
|
0.354
|
ATPP(s)
|
23.00(34.00,63.00)
|
22.00(33.00,56.00)
|
-1.891
|
0.059
|
C-protein(g/L)
|
69.70(17.92,163.26)
|
83.14(17.76,190.56)
|
-0.7383)
|
0.461
|
FIB(g/L,±s)
|
4.18(2.97,5.5)
|
4.81(3.15,6.16)
|
-1.9933)
|
0.046
|
D-Dimer(ng/mL,±s)
|
3.85(2,7.49)
|
5(2.99,15.52)
|
-3.4643)
|
0.001
|
ps:1)Hierarchical chi-square test;2)chi-square test;3)Mann-Whitney. Red blood cell volume( HCT), Activated partial thromboplastin time(APTT),Fibrinogen(FIB).
3.1.3 Prognostic Evaluation
Among the critically ill patients, the length of ICU stay and the 28-day mortality rate between the Non-LEDVT group and the LEDVT group were statistically significant (both P<0.05), as shown in Table 3.
Table 3 Comparison of prognostic indicators between Non-LEDVT group and LEDVT group
Variable
|
Non-LEDVT
|
LEDVT
|
c²/F/Z
|
P
|
LOS (d)
|
15.00(10.00,18.00)
|
16.00(12.00,21.00)
|
-2.856
|
<0.0012)
|
28-day mortality rate(n,%)
|
40(5.52)
|
9(12.00)
|
3.500
|
0.0481)
|
ps:1)chi-square test;2)Mann-Whitney. Length of ICU stay(LOS)
3.2 BP-ANN RAM of LEDVT
3.2.1 Determine the structure of BP-ANN RAM
Following the Virchow three elements of thrombosis and the results of statistical analysis, the influencing factors of interest general data were included in the input layer. In addition, the age, gender, and BMI that were closely related to the occurrence of VTE were also included in the input layer[18, 19]. According to the empirical formula , the number of hidden layer neurons was calculated to be 5~14. Through repeated training, it was finally determined that the number of hidden layer neurons was 9 and the number of output layer neurons was 2. The activation function was determined according to the AUC values corresponding to different activation functions. The activation function ReLU corresponds to the highest AUC, so the ReLU was used as the activation function of the BP-ANN RAM. The learning rate was 0.18, the number of iterations was 30000, and the R value was 0.82, were detailed in Table 4.
Table 4. Comparison of different activation functions of BP-ANN RAM
Activation functions
|
AUC
|
Sensitivity(%)
|
Specificity(%)
|
Sigmoid
|
0.776
|
72.73
|
69.98
|
tanh
|
0.763
|
63.64
|
79.92
|
ReLU
|
0.828
|
72.73
|
81.29
|
3.2.2The prediction ability of BP-ANN RAM and compare to different RAM
The area under the curve of the BP-ANN RAM was 0.828. The sensitivity was 72.73% and the specificity was 81.28%. BP-ANN, Caprini, Padua, and Wells RAM have certain predictive capabilities for the occurrence of LEDVT in critically ill patients. Comparing BP-ANN RAM with the previous three RAM, the AUC, sensitivity, and sensitivity of the BP-ANN model were the highest value, the predictive ability was better than the other three models significantly, the difference was statistically significant (Z=7.991, P<0.001; Z=3.435, P<0.006; Z=3.041, P<0.024). In this study, the best diagnostic threshold of the Caprini RAM was 8 points, and the sensitivity and specificity were well balanced, about 68.00%. In addition, the best diagnostic threshold of the Padua model was 6 points, which shows a good specificity of 77.90%. The best diagnostic threshold of the Wells model was 2 points, and its specificity was slightly higher than the sensitivity of this model at 66.29%, were detailed in Table 5 and Figure1.
Table 5. Comparison of prediction capabilities of Caprini, Padua, Wells and BP-ANN RAM
Models
|
AUC
|
Sensitivity(%)
|
Specificity(%)
|
+LR
|
-LR
|
Caprini
|
0.727
|
68.000
|
67.810
|
2.110
|
0.470
|
Padua
|
0.626
|
45.330
|
77.900
|
2.050
|
0.700
|
Wells
|
0.605
|
49.330
|
66.290
|
1.460
|
0.760
|
BP-ANN
|
0.828
|
72.730
|
81.290
|
3.890
|
0.340
|
ps: likelihood ratio, LR
3.2.2 Importance of neurons in the input layer of BP-ANN RAM
The results of normalized importance analysis the input layer neurons shown that the factors have an important influence on the formation of LEDVT were D-Dimer, APACHE-Ⅱ, CHF, WBC, CRRT FIB and BMI, a total of 7 influencing factors, and their importance were greater than 50. 00%, the specific values are 100%, 76.9%, 74.3%, 71.3% and 67.5%, 55.1% and 52.8% respectively, as shown in table6.
Table 6. Rank of independent variable importance on the BP-ANN RAM
Rank
|
variable
|
Importance
|
Normalized importance(%)
|
1
|
D-Dimer
|
0.134
|
100.0%
|
2
|
APACHE-Ⅱ
|
0.103
|
76.9%
|
3
|
CHF
|
0.099
|
74.3%
|
4
|
WBC (10^9)
|
0.095
|
71.3%
|
5
|
CRRT
|
0.090
|
67.5%
|
6
|
FIB
|
0.074
|
55.1 %
|
7
|
BMI
|
0.071
|
52.8%
|
8
|
Age
|
0.057
|
42.8%
|
9
|
PLT
|
0.055
|
41.0%
|
10
|
DM
|
0.042
|
31.4%
|
11
|
MV
|
0.035
|
26.5%
|
12
|
Multiple trauma
|
0.033
|
24.5%
|
13
|
Lower extremity edema
|
0.030
|
22.5%
|
14
|
CVC
|
0.018
|
13.3%
|
15
|
Stroke
|
0.018
|
10.9%
|
16
|
Gender
|
0.015
|
13.3%
|
17
|
Family history of VTE
|
0.011
|
8.2%
|
18
|
Malignant tumor
|
0.010
|
7.8%
|
19
|
Surgery>45min
|
0.010
|
7.2%
|