Of the 52,963 ICU admissions from MIMIC-III database, 5784 patients fulfilled the definition of sepsis. 3310 participants were identified into our analysis according to the inclusion criteria, afterwards patients with preexisting liver disease (n=526) was excluded from our study. At last 456 (16.4%) patients were consisted of jaundice group, and the remaining 2328 (83.6%) patients did not develop Jaundice (Fig. 1).
Patient characteristics before matching.
Table 1 showed the notable differences in baseline characteristics between jaundice and non-jaundice groups of sepsis before propensity score matching. The jaundice group had a higher prevalence of other race (20.6% vs.16.3%; P=0.029), certain preexisting medical comorbidities, including AFIB (31.8% vs. 26.7%; P=0.029) and malignancy (28.5% vs.21.4%; P=0.001), and lower prevalence of COPD (8.9% vs. 14.9%; P=0.029), stroke (3.5% vs. 8.4%; P <0.001) and renal disease (14.9% vs.21.7%; P=0.001) than the non-jaundice group.
There was also significantly higher severity of illness in Jaundice group than non-jaundice group, with higher SOFA score (8.4±4.0 vs. 5.7±3.0; P<0.001), higher heart rate (112.9±24.0 vs. 107.7±21.3; P<0.001), respiratory rate (29.4±7.2 vs. 28.6±6.5; P=0.027), and also lower systolic pressure (144.5±27.0 vs. 148.8±24.9; P=0.001), diastolic pressure (84.5±19.7 vs. 87.3±20.3; P=0.009). On 24h after admission to ICU, jaundice group had higher rate of vasopressor treatment (46.7 vs. 39.3%; P=0.004), and changed liver function biomarker, as higher AST (639.0±1434.6 vs. 168.7±634.9; P<0.001), ALT (532.3±1371.9 vs. 122.8±514.8; P<0.001),TBIL 24h (5.3±5.3 vs. 0.7±0.5; P<0.001), TBIL 48h (3.7±5.0 vs. 0.3±0.4; P<0.001), TBIL 72h (4.4±4.8 vs. 0.7±0.5; P<0.001), TBIL 7d (5.3±6.4 vs. 0.6±0.4; P<0.001), TBIL at the day of discharge (5.5±6.5 vs. 0.7±0.5; P<0.001) and lower serum albumin (3.0±0.7 vs. 3.2±0.7; P<0.001). The Elixhauser comorbidity score (5.3±7.3 vs. 3.8±7.1; P<0.001), serum lactate (4.1±3.6 vs. 3.0±2.5; P<0.001), and INR (2.2±2.2 vs. 1.7±1.6; P<0.001) were also higher in jaundice group. And the jaundice group had lower serum platelet (215.6±123.8 vs. 261.4±134.6; P<0.001), bicarbonate (23.1±4.5 vs. 24.5±4.7; P<0.001), sodium (140.1±5.0 vs. 140.9±5.9; P=0.013) and arterial pCO2 (46.0±14.9 vs. 48.1±16.1; P=0.030) than the non-jaundice group.
The overall in-hospital mortality of patients with sepsis was 12.1%. Patients in jaundice group had significantly higher rates of in-hospital mortality (17.9% vs.11.0%, P<0.001), 30d mortality (24.0% vs.16.6%, P<0.001), 90d mortality (28.3%vs.19.9%, P<0.001), 180d mortality (30.3% vs.22.5%, P<0.001), and one-year mortality after discharge (32.8% vs.25.0%, P<0.001) than in non-jaundice group, respectively. Kaplan-Meier’s analysis also showed that one-year survival rate was significantly lower in jaundice group than in non-jaundice group (P=0.0149) before matching (Fig.2A).
Univariate and stepwise multivariate Cox hazard analysis of risk factors for mortality in sepsis before matching.
Further, univariate and stepwise multivariate Cox hazard analysis were used to analyze independent risk factors for mortality in sepsis (Table 2). Age [Hazard ratio, HR 1.029; 95% CI (1.009-1.049); P=0.005], preexisting malignancy [HR 3.244; 95% CI (1.729-6.085); P<0.001], SOFA score [HR 1.179; 95% CI (1.054-1.318); P=0.004] and serum total bilirubin at hospital discharge [HR 1.050; 95% CI (1.022-1.079); P<0.001] were independent risk factors for one-year mortality before matching by stepwise multivariate Cox hazard analysis.
Propensity score analysis
Then, we conducted propensity-score matching to controll possible confounders. One-to-one propensity-score matching yielded 432 pairs (Table 3), baseline characteristics for patient were well balanced between the two groups after matching. Standardized biases for all variables were 0.05 or less. In jaundice group after matching, jaundice were still prone to bring more severity of sepsis, with higher SOFA sore (8.5±4.0 vs. 5.6±3.0; P<0.001), heart rate (113.0±24.1 vs. 108.3±21.1; P=0.002) and respiratory rate (29.4±7.3 vs. 28.5±6.3; P=0.043), lactate (4.1±3.6 vs. 3.1±2.6; P<0.001), serum BUN (37.7±29.8 vs. 33.3±23.6; P=0.017); lower maximum serum platelet (214.3±123.5 vs. 261.9±134.7; P<0.001), and bicarbonate (23.0±4.5 vs. 24.6±4.5; P<0.001) on 24h admissionthan the matched non-jaundice group. Jaundice group also had significantly higher levels of liver injury related indicators, as AST (649.5±1434.7 vs. 221.3±847.0; P<0.001), ALT (541.2±1391.9 vs. 146.9±567.8; P<0.001), TBIL 24h (5.4±5.4 vs. 0.8±0.5; P<0.001), TBIL 48h (5.8±5.5 vs. 0.8±0.5; P<0.001).
The matched results also showed that jaundice was associated with significantly increased hospital mortality (17.6% vs. 11.3%P=0.012), but not statistically increased in 30d, 90d, 180d, and one-year mortality in overall population (Table 3, Fig. 2B).
Univariate and multivariate regression analysis to explore independent predictors affecting jaundice in sepsis after matching
Factors associated with the development of jaundice were entered into multivariate regression analysis, which revealed that mechanical ventilation [OR 0.310; 95% CI (0.222-0.432); P<0.001], serum platelet [OR 0.998; 95% CI (0.997-1.000); P=0.015] and serum bicarbonate at 24h admission [OR 0.962; 95% CI (0.929-0.996); P=0.030] were the independent protective factors of new presented jaundice, SOFA score [OR 1.314; 95% CI (1.248-1.385); P<0.001] was the only independent risk factor of new onset of jaundice during course of sepsis (Table 4).