Patient Characteristics
The characteristics of the 103 patients with AE-IPF are shown in Table 1. Among the 88 men and 15 women (mean age, 73.7 years; range, 55–89 years) there were 3 current smokers, 81 former smokers, and 19 never-smokers. Mean (SD) number of pack-years was 57.9 (31.2), and 94 cases were associated with a UIP pattern and 9 cases were associated with a probable UIP pattern. Table 1 also shows data for laboratory variables, interstitial lung disease markers (such as KL-6 and SP-D), HRCT pattern of AE-IPF, and PaO2/FiO2 ratio at AE-IPF onset. PaO2/FiO2 ratio at AE-IPF onset was significantly but weakly negatively correlated with GAP stage (r=−0.2937, p<0.05) and JRS classification of IPF disease severity (r=−0.3986, p<0.01). All patients were treated with high-dose corticosteroid pulse therapy (methylprednisolone 1000 mg/day for 3 days). Corticosteroid dose was tapered after pulse therapy (0.5–1.0 mg/kg/day). Some patients were treated with combination treatment: cyclosporin A (CsA) 2.5 mg/kg/day was administered orally, sivelestat 4.8 mg/kg/day was administered intravenously for the first 14 days, and recombinant human thrombomodulin (rhTM) 0.06 mg/kg/day was administered intravenously for the first 6 days.
Table 1
Characteristics of 103 patients with AE-IPF
|
Overall
(n=103)
|
Survivors
(n=55)
|
Non-survivors (n=48)
|
P-value
|
Male sex n (%)
|
88 (85.4%)
|
45(81.8%)
|
43(89.6%)
|
0.4
|
Age median (range)
|
73.7±6.9
|
73.4±7.2
|
74.1±6.6
|
0.76
|
Smoking history:
|
|
|
|
|
Never/Former/Current
|
19/81/3
|
10/44/1
|
9/37/2
|
0.98
|
Pack-years (mean ± SD)
|
57.9 ± 31.2
|
58.3±29.3
|
56.4±34.3
|
0.56
|
%FVC
|
71.0±71.8
|
72.9±21.0
|
68.6±19.1
|
0.44
|
%DLco
|
35.5±30.4
|
37.2±29.6
|
33.6±31.4
|
0.57
|
GAP score
|
2.7±2.2
|
2.8±2.3
|
2.6±2.2
|
0.68
|
Laboratory and radiological findings at the onset of AE-IPF
|
|
|
PaO2/FiO2 ratio
|
252.2±85.6
|
283.3±79.3
|
217.4±79.3
|
<0.001*
|
KL-6 (U/ml)
|
1535.3±1856.8
|
1291.4±875.2
|
1748.2±2397.2
|
0.58
|
SP-D (ng/ml)
|
438.6±356.6
|
395.9±321.9
|
487.6±390.3
|
0.22
|
SP-A (ng/ml)
|
103.0±58.5
|
101.9±60.6
|
104.4±56.5
|
0.98
|
LDH (U/l)
|
373.9±186.1
|
329.1±68.7
|
424.3±253.3
|
0.007*
|
CRP (mg/dl)
|
8.0±6.4
|
6.2±5.5
|
10.0±6.9
|
0.002*
|
Diffuse HRCT pattern, n (%)
|
73 (70.9)
|
30(54.5)
|
43(89.6)
|
<0.001*
|
*P<0.05
%FVC: Forced vital capacity, % of predicted value, %DLco Carbon monoxide diffusing capacity -% of predicted value, GAP score: Gender-Age-Physiology score, KL-6: Krebs von der Lungen-6, SP-D: surfactant protein D, SP-A: surfactant protein A, LDH: lactate dehydrogenase, CRP: C-reactive protein, HRCT: high-resolution computed tomography, CS: corticosteroid, PMX-DHP: Polymyxin B-immobilized fiber column-direct hemoperfusion, rhTM: recombinant human thrombomodulin
|
Logistic regression analysis for 3-month mortality
In univariate logistic regression analysis including all variables, 3-month mortality was significantly associated with PaO2/FiO2 ratio (odds ratio [OR] 0.9898, 95% CI 2.1317–15.9630, P = 0.0002), diffuse HRCT pattern (OR 5.8333, 95% CI 0.98436–0.9953, P = 0.0006), serum LDH concentration (OR 1.0068, 95% CI 1.0022–1.0115, P = 0.0041), and serum CRP concentration (OR 1.1147, 95% CI 1.0366–1.1988, P = 0.0034). Neither the severity of IPF nor pulmonary function test results before AE-IPF were significantly associated with 3-month mortality. Furthermore, no AE-IPF treatment was significantly associated with 3-month mortality. Thus, these factors were not used as candidate variables in multivariate analysis.
In multivariate logistic regression analysis with selected explanatory variables, 3-month mortality was significantly associated with PaO2/FiO2 ratio (OR 0.9920, 95% CI 0.9856–0.9984, P = 0.0139), diffuse HRCT pattern (OR 3.2399, 95% CI 1.0268–10.2228, P = 0.0450), serum LDH concentration (OR 1.0048, 95% CI 0.9983–1.0113, P = 0.1500), and serum CRP concentration (OR 1.0959, 95% CI 1.0145–1.1838, P = 0.0201) (Table 2).
Table 2
Results of multivariate logistic regression analysis with selected variables as explanatory variables (upper) and after stepwise variable selection (lower)
Variables
|
Partial regression coefficient
|
95% CI of partial regression coefficient
|
Odds ratio
|
95% CI of Odds ratio
|
P value
|
|
Diffuse HRCT pattern
|
1.1755
|
-0.0265-2.3246
|
3.2399
|
1.0268-10.2228
|
0.0450
|
*
|
KL-6
|
-0.0006
|
-0.0011-0.0000
|
0.9994
|
0.9989-1.0000
|
0.0628
|
|
SP-D
|
0.0011
|
-0.0006-0.0027
|
1.0011
|
0.9994-1.0027
|
0.1994
|
|
LDH
|
0.0048
|
-0.0017-0.0113
|
1.0048
|
0.9983-1.0113
|
0.1500
|
|
CRP
|
0.0916
|
0.0144-0.1687
|
1.0959
|
1.0145-1.1838
|
0.0201
|
*
|
PaO2/FiO2 ratio
|
-0.0081
|
-0.0145-0.0016
|
0.9920
|
0.9856-0.9984
|
0.0139
|
*
|
|
|
|
|
|
*P<0.05
|
|
Stepwise variable selection
Variables
|
Partial regression coefficient
|
95% CI of partial regression coefficient
|
Odds ratio
|
95% CI of Odds ratio
|
P value
|
|
Diffuse HRCT pattern
|
1.3386
|
0.2430-2.4342
|
3.8137
|
1.2751-11.4065
|
0.0166
|
*
|
CRP
|
0.0943
|
0.0217-0.1669
|
1.0989
|
1.0219-1.1817
|
0.0109
|
*
|
PaO2/FiO2 ratio
|
-0.0082
|
-0.0142- -0.0022
|
0.9919
|
0.9859-0.9978
|
0.0074
|
**
|
|
|
|
|
|
*P<0.05, **P<0.01
|
CI: confidence interval, HRCT: high-resolution computed tomography, KL-6: Krebs von der Lungen-6, SP-D: surfactant protein D, SP-A: surfactant protein A, LDH: lactate dehydrogenase, CRP: C-reactive protein |
In multivariate logistic regression analysis with selected explanatory variables, followed by stepwise variable selection, 3-month mortality was significantly associated with PaO2/FiO2 ratio (OR 0.9919, 95% CI 0.9859–0.9978, P = 0.0074), diffuse HRCT pattern (OR 3.8137, 95% CI 1.2751–11.4065, P = 0.0166), and serum CRP concentration (OR 1.0989, 95% CI 1.0219–1.1817, P = 0.0109) (Table 2). The c-index was 0.8027 (95% CI 0.7167–0.8886, P<0.0001). In addition, 1-month mortality was significantly associated with PaO2/FiO2 ratio (OR 0.9910, 95% CI 1.0179–0.9973, P = 0.0053) and serum CRP concentration (OR 1.0965, 95% CI 1.0179–1.1812, P = 0.0152) in multivariate analysis.
New risk scoring system for predicting 3-month mortality after AE-IPF
To develop the grading system, cut-off values for variables were determined by ROC curve analysis. The cut-off value was 250 for PaO2/FiO2 ratio and 5.5 for CRP concentration. The variables identified as significant risk factors for 3-month mortality were a PaO2/FiO2 ratio of <250 (OR 3.1632, 95% CI 1.2497–8.0065, P = 0.0151), diffuse HRCT pattern (OR 4.1872, 95% CI 1.4445–12.1378, P = 0.0084), and serum CRP concentration ≥5.5 (OR 2.7347, 95% CI 1.1045–6.7710, P = 0.0296), all with integral weights of 1 (Table 3). The c-index was 0.7739 (95% CI 0.6855–0.8623, P<0.0001). Points were assigned to variable categories to create a point-score model (PCR index), as shown in Table 4. The total score (PCR score) was calculated as [1 × PaO2/FiO2 ratio (<250): (P)] + [1 × CRP (≥5.5 mg/dl): (C)] + [1 × HRCT (radiological) findings (diffuse HRCT pattern): (R)]. The score ranged from 0 to 3. The diagnostic performance of the PCR score for predicting 3-month mortality was examined by ROC analysis, and the area under the ROC curve (AUC) was 0.7686 (95% CI 0.6812–0.8544, P<0.0001). The optimal cut-off value for PCR score was 2, as indicated by the Youden index, and sensitivity was 0.771 and specificity was 0.655 for predicting cause-specific survival (Fig. 2). To calibrate the PCR model, we evaluated agreement between the predicted probability by PCR score and observed probability. Figure 3 shows the calibration plot. The observed and predicted probabilities were in excellent agreement in the Hosmer–Lemeshow test (P=1.000). Ultimately, a grading system was created by grouping point scores into 4 groups. Three-month mortality was 7.7% for a score of 0, 29.4% for a score of 1, 54.8% for a score of 3, and 80.0% for a score of 4 (Fig. 4). The log-rank test for trend revealed that survival curves were consistent with PCR scores (p<0.0001). In addition, we confirmed that values for Kaplan–Meier analysis of the 3-month mortality rate for each PCR score and the estimated 3-month mortality rate calculated by the bootstrap method exhibited similar trends (Table 4).
Table 3
Results of multivariate logistic regression analysis with categorized selected variables
Variables
|
Partial regression coefficient
|
95% CI of partial regression coefficient
|
Odds ratio
|
95% CI of Odds ratio
|
P value
|
Integral Weight
|
Diffuse HRCT pattern
|
1.4320
|
0.3677-2.4963
|
4.1872
|
1.4445-12.1378
|
0.0084*
|
1
|
High CRP (≥5.5)
|
1.0060
|
0.0994-1.9126
|
2.7347
|
1.1045-6.7710
|
0.0296*
|
1
|
Low PaO2/FiO2 ratio (<250)
|
1.1516
|
0.2229-2.0802
|
3.1632
|
1.2497-8.0065
|
0.0151*
|
1
|
*P<0.05 CI: confidence interval, HRCT: high-resolution computed tomography, CRP: C-reactive protein |
Table 4
PCR index and grading system.
Predictor
|
Value
|
Score
|
PaO2/FiO2ratio
(P)
|
<250
|
1
|
≧250
|
0
|
CRP
(C)
|
≧5.5
|
1
|
<5.5
|
0
|
HRCT pattern
(R)
|
Diffuse
|
1
|
Non-diffuse
|
0
|
Total Possible Points 3
Measured value (n=103)
|
3-month mortality estimated by bootstrap method
|
95%CI
|
Total score of PCR index
|
No. of patients
|
3-month mortality
|
Lower
|
Upper
|
0
|
n=13
|
0.077 (7.7%)
|
0.083 (8.3%)
|
0.072
|
0.081
|
1
|
n=34
|
0.294 (29.4%)
|
0.291 (29.1%)
|
0.291
|
0.300
|
2
|
n=31
|
0.548 (54.8%)
|
0.554 (55.4%)
|
0.564
|
0.576
|
3
|
n=25
|
0.800 (80.0%)
|
0.803 (80.3%)
|
0.798
|
0.808
|
CI: confidence interval, HRCT: high-resolution computed tomography, CRP: C-reactive protein
|