Clinical data and calculus composition
The workflow is shown in Figure 1. 480 UUTC patients were included in our research, including 328 from Fujian Provincial Hospital South Branch and 152 from Fujian Provincial Hospital. In South Branch (construction cohort), the medium age was 51 (42,59) years. 63.82% were men. The most common stone type was calcium oxalate monohydrate (COM) (66.45%), followed by carbonate apatite (CaPa) (17.11%) and calcium oxalate dihydrate (COD) (13.16%). The incidence rates of postoperative infection and sepsis were 9.87% and 3.29%, respectively. In Fujian Provincial Hospital (validation cohort), The medium age was 54 (44,62) years. 60.37% were men. The most common type was the same as the construction cohort: COM stone (58.54%), CaPa (23.17%), and COD (16.16%). The incidence rates of postoperative infection and sepsis were 9.45% and 3.05%. Details are described in Table S1 and Figure S1.
Infection-related infrared spectrum features extraction and IR-infection score construction
In the construction cohort, 3736 infrared wavelengths were calculated by FTIR. By univariate analysis, 1306 wavelengths were identified to be related to postoperative infection in UCTC patients (P < 0.05) (Figure 2A, B). After the greedy recursive deletion strategy, 16 relatively independent infection-related wavelengths were anchored (intraclass correlation coefficient < 0.9) (Figure 3A). In the end, 4 features, including WL_1261.22, WL_1320.03, WL_1479.13, and WL_1966.07, were finally made up of the IR-infection score system by LASSO analysis (Figure 3B). We calculated each patient's IR-infection score based on the formula as follows: IR-infection score = -2.23639 + ∑i (-1.05822) ×(WL_1261.22 transmittance) + (2.46380) ×(WL_1320.03 transmittance) + (-1.17410) ×(WL_1479.13 transmittance) + (0.98789) × (WL_1966.07 transmittance). As for the validation cohort, 694 wavelengths were identified to be related to postoperative infection (Figure 2A, B). When two cohorts intersect, we got 653 shared infrared wavelengths, giving us information about influential substances that can trigger postoperative infections (Figure 2C and Files S2).
Assessment of IR-infection score for postoperative infection
In the construction cohort, IR-infection score had a favorable performance in predicting postoperative infection of UUTC patients (AUC = 0.708) (Figure 4A). Then, calibration curves showed the reliability and stability of our IR-infection score (P = 0.756) (Figure 4A). Moreover, the decision curves showed the superiority of the IR-infection score based on the net benefit and threshold probabilities (6.3%-45.4%) (Figure 4A). Our IR-infection score also functioned well in the validation cohort (AUC = 0.707) (Figure 4B). And its prediction power was also validated by calibration curves (P = 0.148) (Figure 4B) and the decision curves (2.0%-16.7%) (Figure 4B).
Comparison between IR-infection score and traditional indicators
To demonstrate the better clinical value of the IR-infection score, we compared its AUC with 12 preoperative indicators in the two-insert cohort. The AUC value of IR-infection score was statistically significantly higher than infection stone, WBC, PCT, URBC, UPRO, UNIT, UPH, and UGLU (all P<0.05) (Table 1). In addition, the IR-infection score showed an improvement in predictive ability compared to UWBC (Difference value (D-value) = 0.015), UBACT (D-value = 0.065), and urine culture (D-value = 0.042) (Table 1). The same trends were found in construction and validation cohorts (Table S2, 3).
Table 1 Comparison between IR-infection score and traditional indicators in the two-combined cohort.
Variables
|
AUC
|
D-value of AUC
|
Delong
|
Variables
|
AUC
|
D-value of AUC
|
Delong
|
Infection stone 1
|
0.633
|
0.077
|
0.049*
|
UPRO
|
0.547
|
0.163
|
0.001**
|
Infection stone 2
|
0.611
|
0.099
|
0.013*
|
UBACT
|
0.645
|
0.065
|
0.226
|
BWBC
|
0.514
|
0.196
|
0.004**
|
UNIT
|
0.589
|
0.121
|
0.015*
|
PCT
|
0.570
|
0.140
|
0.004**
|
UPH
|
0.581
|
0.129
|
0.038*
|
URBC
|
0.496
|
0.214
|
<0.001***
|
UGLU
|
0.509
|
0.201
|
<0.001***
|
UWBC
|
0.695
|
0.015
|
0.771
|
Urine culture
|
0.668
|
0.042
|
0.453
|
*: P < 0.05, **: P < 0.01, ***: P < 0.001.
Risk factors for postoperative infection in UUTC patients
In the construction cohort, univariate analysis showed that postoperative infection was related to IR-infection score, gender, calculus volume, surgery procedure, operation time, infection stone 1, infection stone 2, PCT, UWBC, UPRO, UBACT, UNIT, and urine culture (all P < 0.05) (Table 2). Multivariate logistic regression analysis anchored the IR-infection score as independent risk factors (P < 0.05) (Table 2). Details are shown in Table S4.
Table 2 Comparison of clinical characteristics and IR-infection score between postoperative infection and non-postoperative infection groups.
|
Construction cohort
n = 328
|
P value
|
Validation cohort
n = 152
|
P value
|
|
Infection,
n = 31
|
Non-infection, n = 297
|
Univariate analysis
|
Multivariate analysis
|
Infection,
n = 15
|
Non-infection, n = 137
|
Univariate analysis
|
Multivariate analysis
|
Gender, n(male) (%)
|
9
(29.03)
|
189
(63.64)
|
<0.001***
|
0.069
|
5
(33.33)
|
92
(67.15)
|
0.010**
|
0.719
|
Calculus volume, cm3 (IQR)
|
2.08
(0.62,7.59)
|
0.70
(0.27,1.67)
|
0.001**
|
0.621
|
3.46
(1.04,13.14)
|
1.24
(0.51,2.89)
|
0.046*
|
0.201
|
RIRS / PCNL
or both, n (%)
|
14
(45.16)
|
69
(23.23)
|
0.008**
|
0.195
|
11
(73.33)
|
85
(62.04)
|
0.389
|
-
|
Operation time, min (IQR)
|
125
(90,144)
|
92
(60,129)
|
0.008**
|
0.259
|
95
(62.5,101.5)
|
75
(57,95)
|
0.383
|
-
|
Infection stone 1, n (%)
|
18
(58.06)
|
98
(33.00)
|
0.005**
|
0.904
|
8
(53.33)
|
32
(23.36)
|
0.028*
|
0.385
|
Infection stone 2, n (%)
|
13
(41.94)
|
63
(21.21)
|
0.009**
|
0.554
|
6
(40.00)
|
20
(14.60)
|
0.034*
|
0.292
|
PCT, n (%)
|
6
(19.35)
|
9
(3.03)
|
<0.001***
|
0.051
|
2
(13.33)
|
6
(4.38)
|
0.180
|
-
|
UWBC, n (%)
|
27
(87.10)
|
163
(54.88)
|
<0.001***
|
0.426
|
13
(86.67)
|
45
(32.85)
|
<0.001***
|
0.095
|
UPRO, n (%)
|
6
(19.35)
|
19
(6.40)
|
0.026*
|
0.208
|
2
(13.33)
|
16
(11.68)
|
1.000
|
0.382
|
UBACT, n (%)
|
14
(45.16)
|
77
(25.93)
|
0.023*
|
0.413
|
11
(73.33)
|
33
(24.09)
|
<0.001***
|
0.290
|
UNIT, n (%)
|
7
(22.58)
|
19
(6.40)
|
0.005**
|
0.270
|
4
(26.67)
|
8
(5.84)
|
0.020*
|
0.128
|
UPH, (IQR)
|
6.5
(5.8,6.5)
|
6.0
(5.5,6.5)
|
0.532
|
-
|
6.5
(6.0,7.0)
|
6.0
(5.5,6.5)
|
0.022*
|
0.119
|
Urine culture, n (%)
|
14
(45.16)
|
38
(12.79)
|
<0.001***
|
0.155
|
7
(46.67)
|
14
(10.22)
|
<0.001***
|
0.263
|
IR-infection score, (IQR)
|
-2.19
(-2.61,-1.14)
|
-2.59
(-2.80,-2.37)
|
<0.001***
|
0.031*
|
-2.21
(-2.56,-1.20)
|
-2.63
(-2.81,-2.39)
|
0.009**
|
0.716
|
*: P < 0.05, **: P < 0.01, ***: P < 0.001. IQR: interquartile range. Positive PCT is defined as serum concentration > 0.25ng/ml. The positive result of URBC, UWBC, UPRO are defined as concentrations ≥ 2+. A positive urine bacterial count is a bacterial concentration > 114/ul.
Predictive model construction
In the construction cohort, we first targeted 13 risk factors by univariable analysis (all P<0.05) to build the nomogram, including IR-infection score, gender, calculus volume, surgical procedure, operation time, infection stone 1, infection stone 2, PCT, UWBC, UPRO, UBACT, UNIT, and urine culture. Each factor's coefficient of nomogram is shown in Fig. 5A. We calculated each patient's risk score based on the formula below: risk score = -1.48536 + ∑i (-0.85843) × (gender) + (0.01028) × (operation time) + (1.21489) × (PCT) + (1.0574) × (urine culture) + (0.83396) × (IR-infection score). The nomogram we drew well exhibited its prediction power. (Fig. 5B). The AUC of the nomogram was 0.819 in the construction cohort and 0.828 in the validation cohort. Then, calibration curves showed our nomogram was reliable (Pconstruction cohort = 0.423, Pvalidation cohort = 0.526). Also, the decision curves showed the superiority of the nomogram based on the net benefit and threshold probabilities (2.2%-86.3% in the construction cohort and 2.9%-30.9% in the validation cohort) (Fig. 5C, D).
Assessment of IR-infection score and nomogram for postoperative sepsis
In the construction cohort, univariate analysis showed that sepsis was related to IR-infection score, gender, BMI, calculus volume, surgical procedure, operation time, infection stone 1, WBC, UWBC, UPRO, UBACT, UNIT, and urine culture (all P < 0.05). Subsequently, multivariate analysis anchored gender and IR-infection score as independent risk factors in the construction cohort (all P < 0.05). Details are shown in Table S5. Compared with 12 traditional indicators, the IR-infection score had more outstanding predictive performance in the two-combined cohort. The same trends were also found in construction and validation cohorts (Table S6-8). In the construction cohort, the IR-infection score and nomogram were all favorable in predicting sepsis (AUC IR−infection score = 0.690, AUC nomogram = 0.931) (Figure S2A, B). They also functioned well in the validation cohort (AUC IR−infection score = 0.824, AUC nomogram = 0.878) (Figure S2C, D).