Patient characteristics, hematological indicators, and Univariate analyses

As shown in Fig. 1 and Table 1, a total of 118 patients with PA BSIs were included in this study. Among these, 46 patients were in the death group and 72 were in the survival group. Additionally, based on the occurrence of SIC, 31 patients were classified into the SIC group and 58 into the non-SIC group. The median age of the overall cohort was 59.6 years, and males accounted for a higher proportion of patients in both groups.

Fig. 1

A flow diagram of patient selection

Table 1 Demographic characteristics and Univariate analyses

There were notable differences between the death and survival groups in several variables. ICU admission was significantly more common in the mortality group (76.1% vs. 50.0%, P = 0.005), indicating more severe illness. Similarly, patients in the mortality group were more likely to have received invasive ventilation (76.1% vs. 43.1%, P < 0.001). Empirical sensitive antimicrobial therapy was significantly lower in the death group (21.7% vs. 76.4%, P < 0.001), underscoring the potential importance of appropriate antimicrobial treatment in survival outcomes. of all patients, 65 underwent ESAT. Among them, 36 patients were treated with cephalosporins, 19 with carbapenems, 1 with polypeptides, 5 with quinolones, 2 with aminoglycosides, and 2 with penicillins.

In contrast, no significant differences were observed between the death and survival groups in terms of age (60.52 vs. 58.11 years, P = 0.402), gender (58.7% vs. 73.6%, P = 0.091), or smoking history (P = 0.467). Likewise, no significant differences were found between the groups in terms of comorbidities such as hypertension and diabetes.

Further analysis between the SIC and non-SIC groups revealed that patients in the SIC group had significantly elevated levels of inflammatory markers such as CRP and PCT compared to the non-SIC group (P = 0.002 and P < 0.001, respectively). However, no significant differences were observed between these groups regarding ICU admission, invasive procedures like surgery, or the use of CRRT.

Multivariable analysis and construction of prediction model for death in PA BSIs

Based on the outcomes of univariate analysis, 12 significant factors were identified for further analysis (Table 1 and 2). The results of the multivariate analysis are summarized in Table 3 and 2, where we developed three logistic regression models to predict mortality. Model 1: ESAT, steroid therapy, invasive ventilation, CAD, and PTA; Model 2: ESAT, invasive ventilation, CAD, and PTA; Model 3: ESAT, invasive ventilation, and PTA.

Table 2 Univariate analyses of clinical and laboratory results around the first positive blood cultureTable 3 Multivariate logistic regression analysis of risk factors for mortality in PA BSIsTable 4 The construction of prediction model for mortality

Collinearity analysis revealed that the variance inflation factor (VIF) values were all less than 10, indicating no significant multicollinearity (Table 3). We use the Z test to discover that there were no statistically significant differences between the three models. However, Model 2 exhibited higher specificity (Table 4). Based on Fig. 2 and the comparison of AUC curves across the three models, it can be concluded that Model 2 has a specificity of 0.903 with higher specificity indicating a lower false-positive rate. Considering clinical applicability and simplicity, Model 2 was selected as the effective predictive model. In this model, ESAT, CAD, PTA, and invasive ventilation were included to construct the multivariate logistic regression equation. The results showed that: CAD (OR = 10.315, 95% CI: 1.746–60.950, P = 0.010) and invasive ventilation (OR = 3.926, 95% CI: 1.246–12.371, P = 0.020) increased the risk of mortality in adults with PA BSIs. PTA (OR = 0.965, 95% CI: 0.943–0.987, P = 0.002) and ESAT (OR = 0.039, 95% CI: 0.011–0.136, P < 0.001) were identified as independent protective factors.

Fig. 2

The construction of prediction models for PA BSI

Using Model 2, we constructed a nomogram to predict mortality in patients with PA BSIs (Fig. 3). The model’s discrimination was assessed using the Bootstrap resampling method, with 1,000 replications for internal validation. The calibration curve was plotted, and the ROC curve was generated. The AUC of the ROC was 0.908 (greater than 0.7), indicating that the prediction results of the model are consistent with the actual results and have good calibration (Fig. 5A and B). The decision curve shows that this prediction model could have high clinical application value (Fig. 5C).

Fig. 3

A nomogram of model 2 for predicting mortality in adult patients with PA BSIs

Multivariable analysis and construction of prediction model for SIC

Stepwise multivariate analysis (Table 5) was conducted on five significant factors from the univariate analysis (Table 1 and 2). The results revealed the following: CRP (OR = 1.011, 95% CI: 1.004–1.019, P = 0.003) and PCT (OR = 1.030, 95% CI: 1.009–1.052, P = 0.005) were identified as factors that increased the risk of SIC in adult patients with PA BSIs. HB (OR = 0.963, 95% CI: 0.938–0.988, P = 0.004) was identified as an independent protective factor against the risk of SIC in these patients.

Table 5 Multivariate logistic regression analysis of risk factors for SIC in PA BSIs

Based on the binary logistic regression results, a nomogram model was constructed to predict the risk of SIC in patients with PA BSIs (Fig. 4). The Bootstrap resampling method was used to evaluate the model's discrimination, with 1,000 resampling iterations for internal validation. A calibration curve was plotted, and an ROC curve was generated. The ROC curve yielded an area under the curve (AUC) of 0.817 (greater than 0.7), indicating good predictive accuracy of the model (Fig. 5D–F).

Fig. 4

A nomogram of a model for predicting the occurrence of SIC in adult patients with PA BSIs

Fig. 5

A and D, ROC curve and area under the curve of models for predicting mortality and the occurrence of SIC in adults of PA BSIs; B and E, Calibration curve of models; C and F, Decision curve of models