Background
Methods
Patient population
Data collection
Statistical analysis
Results
Characteristics of the study population
Variables | Training dataset (China data) n = 96 | Test dataset (Iran data) n = 43 | P value |
---|---|---|---|
Age (years), mean (SE) | 63.47 (1.36) | 63.37 (2.70) | 0.972 |
Male, n (%) | 49 (51.0) | 30 (69.8) | 0.039 |
Smoking history, n (%) | 1 (1.0) | 2 (4.7) | 0.064 |
Symptoms on admission, n (%) | |||
Fever | 86 (89.6) | 20 (46.5) | < 0.001 |
Cough | 78 (81.3) | 23 (53.5) | 0.222 |
Fatigue | 86 (89.6) | 27 (42.2) | < 0.001 |
Shortness of breath | 70 (72.9) | 24 (55.8) | 0.983 |
Headache | 17 (17.7) | 4 (9.3) | 0.453 |
Diarrhea | 20 (20.8) | 1 (2.3) | 0.017 |
Coexisting disorder, n (%) | |||
Hypertension | 33 (34.4) | 12 (27.9) | 0.748 |
Diabetes | 16 (16.7) | 8 (18.6) | 0.296 |
Chronic obstructive pulmonary disease | 3 (3.1) | 2 (4.7) | 0.444 |
Cerebral infarction | 1 (1.0) | 1 (2.3) | 0.42 |
Coronary heart disease | 10 (10.4) | 0 | 0.057 |
Chronic kidney disease | 3 (3.1) | 1 (2.3) | 0.768 |
Chronic liver disease | 0 | 0 | – |
Respiratory rate, mean (SE) | 27.24 (0.57) | 22.76 (1.09) | < 0.001 |
Heart rate, mean (SE) | 92.89 (1.749) | 90.34 (1.88) | 0.323 |
In-hospital deaths, n (%) | 31 (32.3) | 13 (30.2) | 0.809 |
Feature selection
Derivation and validation of NSL model and NL model
Variable | AUC | 95%CI | SE | Sensitivity (%) | Specificity (%) |
---|---|---|---|---|---|
Training data from China cohort | |||||
LDH (U/L) | 0.880 | 0.813–0.948 | 0.034 | 97 | 71 |
NE (%) | 0.879 | 0.812–0.946 | 0.034 | 84 | 82 |
SaO2 (%) | 0.849 | 0.758–0.940 | 0.046 | 87 | 78 |
LY (%) | 0.852 | 0.776–0.929 | 0.039 | 77 | 80 |
NLR | 0.858 | 0.783–0.933 | 0.038 | 81 | 82 |
CKMB (U/L) | 0.829 | 0.746–0.912 | 0.042 | 87 | 69 |
D-dimer (μg/mL) | 0.763 | 0.641–0.885 | 0.062 | 73 | 88 |
CRP (μg/mL) | 0.807 | 0.723–0.892 | 0.043 | 100 | 51 |
Integrated models | |||||
All variables with information gain > 0.2 | |||||
LDH + NE + SaO2 + LY + NLR + CKMB + D-dimer + CRP | 0.945 | 0.897–0.992 | 0.024 | 97 | 83 |
NE was selected for modeling | |||||
LDH + NE + SaO2 + CKMB + D-dimer + CRP | 0.945 | 0.900–0.989 | 0.023 | 93 | 84 |
LDH + NE + SaO2 + CKMB + D-dimer | 0.942 | 0.898–0.987 | 0.023 | 97 | 78 |
LDH + NE + SaO2 + CKMB | 0.937 | 0.887–0.988 | 0.026 | 83 | 94 |
LDH + NE + SaO2 (NSL risk score) | 0.932 | 0.884–0.981 | 0.025 | 97 | 78 |
LDH + NE (NL risk score) | 0.903 | 0.843–0.963 | 0.031 | 94 | 82 |
LY was selected for modeling | |||||
LDH + SaO2 + LY + CKMB + D-dimer + CRP | 0.948 | 0.904–0.992 | 0.022 | 97 | 84 |
LDH + SaO2 + LY + CKMB + D-dimer | 0.944 | 0.901–0.987 | 0.022 | 86 | 88 |
LDH + SaO2 + LY + CKMB | 0.932 | 0.880–0.984 | 0.026 | 97 | 77 |
LDH + SaO2 + LY | 0.934 | 0.886–0.982 | 0.025 | 90 | 88 |
LDH + LY | 0.903 | 0.843–0.964 | 0.031 | 90 | 82 |
NLR was selected for modeling | |||||
LDH + SaO2 + NLR + CKMB + D-dimer + CRP | 0.930 | 0.866–0.995 | 0.033 | 83 | 95 |
LDH + SaO2 + NLR + CKMB + D-dimer | 0.945 | 0.901–0.989 | 0.022 | 79 | 95 |
LDH + SaO2 + NLR + CKMB | 0.933 | 0.882–0.983 | 0.026 | 77 | 97 |
LDH + SaO2 + NLR | 0.933 | 0.883–0.971 | 0.025 | 87 | 88 |
LDH + NLR | 0.919 | 0.866–0.971 | 0.027 | 90 | 82 |
LDH wasn’t selected for modeling | |||||
NE + SaO2 | 0.919 | 0.865–0.972 | 0.027 | 97 | 78 |
Test data from Iran cohort | |||||
LDH (U/L) | 0.746 | (0.574–0.919) | 0.0881 | 77 | 70 |
NE (%) | 0.851 | 0.719–0.984 | 0.068 | 85 | 82 |
SaO2 (%) | 0.869 | 0.702–1.000 | 0.085 | 85 | 97 |
Combined models | |||||
LDH + NE + SaO2 (NSL risk score) | 0.910 | 0.758–1.000 | 0.077 | 92 | 96 |
LDH + NE (NL risk score) | 0.871 | 0.734–1.000 | 0.071 | 92 | 82 |
Nomogram prediction for in-hospital death of severe patients
Development of risk scoring system for predicting in-hospital death
Variables | Categories | Reference value (Wij) | Bi | Regression units Βi (Wij—WiREF) | Points assigned Βi (Wij—WiREF)/B |
---|---|---|---|---|---|
NSL risk score (NE + SaO2 + LDH) | |||||
NE (%) | 0.127 | ||||
≤ 60 | 55 (WiREF) | 0.000 | 0 | ||
60.1–70 | 65 | 1.270 | 4 | ||
70.1–80 | 75 | 2.540 | 8 | ||
80.1–90 | 85 | 3.810 | 12 | ||
≥ 90.1 | 95 | 5.080 | 16 | ||
SaO2 (%) | -0.175 | ||||
100–96 | 98 (WiREF) | 0.000 | 0 | ||
95–91 | 93 | 0.875 | 3 | ||
90–86 | 88 | 1.750 | 6 | ||
≤ 85 | 83 | 2.625 | 9 | ||
LDH (U/L) | 0.003 | ||||
≤ 221 | 171 (WiREF) | 0.000 | 0 | ||
222–321 | 271 | 0.300 | 1 | ||
322–421 | 371 | 0.600 | 2 | ||
422–521 | 471 | 0.900 | 3 | ||
522–621 | 571 | 1.200 | 4 | ||
622–721 | 671 | 1.500 | 5 | ||
722–821 | 771 | 1.800 | 6 | ||
822–921 | 871 | 2.100 | 7 | ||
922–1021 | 971 | 2.400 | 8 | ||
≥ 1022 | 1071 | 2.700 | 9 | ||
NL risk score (NE + LDH) | |||||
NE (%) | 0.158 | ||||
≤ 60 | 55 (WiREF) | 0.000 | 0 | ||
60.1–70 | 65 | 1.580 | 4 | ||
70.1–80 | 75 | 3.160 | 8 | ||
80.1–90 | 85 | 4.740 | 12 | ||
≥ 90.1 | 95 | 6.320 | 16 | ||
LDH (U/L) | 0.004 | ||||
≤ 221 | 171 (WiREF) | 0 | 0 | ||
222–321 | 271 | 0.400 | 1 | ||
322–421 | 371 | 0.800 | 2 | ||
422–521 | 471 | 1.200 | 3 | ||
522–621 | 571 | 1.600 | 4 | ||
622–721 | 671 | 2.000 | 5 | ||
722–821 | 771 | 2.400 | 6 | ||
822–921 | 871 | 2.800 | 7 | ||
922–1021 | 971 | 3.200 | 8 | ||
≥ 1022 | 1071 | 3.600 | 9 |
NSL risk score (NE + SaO2 + LDH) | NL risk score (NE + LDH) | ||
---|---|---|---|
Point of total | Estimate of risk of hospital mortality (%) | Point of total | Estimate of risk of hospital mortality (%) |
0 | 0.18 | 0 | 0.22 |
1 | 0.25 | 1 | 0.33 |
2 | 0.33 | 2 | 0.49 |
3 | 0.45 | 3 | 0.74 |
4 | 0.61 | 4 | 1.09 |
5 | 0.82 | 5 | 1.62 |
6 | 1.10 | 6 | 2.40 |
7 | 1.48 | 7 | 3.54 |
8 | 1.99 | 8 | 5.20 |
9 | 2.67 | 9 | 7.56 |
10 | 3.57 | 10 | 10.87 |
11 | 4.76 | 11 | 15.39 |
12 | 6.32 | 12 | 21.35 |
13 | 8.34 | 13 | 28.82 |
14 | 10.94 | 14 | 37.66 |
15 | 14.22 | 15 | 47.40 |
16 | 18.29 | 16 | 57.35 |
17 | 23.20 | 17 | 66.73 |
18 | 28.97 | 18 | 74.95 |
19 | 35.50 | 19 | 81.70 |
20 | 42.63 | 20 | 86.94 |
21 | 50.07 | 21 | 90.85 |
22 | 57.52 | 22 | 93.68 |
23 | 64.63 | 23 | 95.67 |
24 | 71.16 | 24 | 97.06 |
25 | 76.91 | 25 | 98.01 |
26 | 81.80 | ||
27 | 85.85 | ||
28 | 89.12 | ||
29 | 91.71 | ||
30 | 93.72 | ||
31 | 95.27 | ||
32 | 96.45 | ||
33 | 97.35 | ||
34 | 98.02 |