Introduction

Pseudomonas aeruginosa (P. aeruginosa) is a monotrichous, non-spore-forming, Gram-negative, facultative aerobic bacterium.¹ According to the 2019 Global Burden of Disease Study, P. aeruginosa ranks as the fifth leading cause of infection-related mortality among all bacterial pathogens investigated and is the second most common pathogen responsible for both community-acquired and hospital-acquired pneumonia.² A prevalence survey identified P. aeruginosa as the fifth most common cause of healthcare-associated infections, with a prevalence of 7.1% in tertiary hospitals and an overall prevalence of 24.7% in Asia.^3,4 In China, the prevalence of P. aeruginosa-related infections in tertiary healthcare institutions is 28.7%, with its isolation frequency ranking second only to Acinetobacter baumannii.⁵

The antimicrobial resistance of P. aeruginosa has garnered significant concern, as it is one of the six leading pathogens responsible for resistance-related mortality and is included in the ESKAPE pathogen group.⁶ The World Health Organization (WHO) has classified carbapenem-resistant P. aeruginosa as a “critical priority” pathogen.⁷ Current studies demonstrate that P. aeruginosa exhibits resistance to multiple commonly used antimicrobial agents, including carbapenems (imipenem resistance rate: 31.3%; meropenem: 24.4%), cephalosporins (ceftazidime: 18.5%; cefepime: 12.0%), and aminoglycosides (gentamicin: 25.1%).^8–10 Notably, the resistance rates of P. aeruginosa exhibit geographical variations globally. The European Centre for Disease Prevention and Control (ECDC) 2021 report indicated that 18.7% of P. aeruginosa isolates were resistant to carbapenems, while 13.4% were resistant to three or more antimicrobial classes.⁹ Indian studies have shown an increasing trend in P. aeruginosa resistance.¹¹ In Asian regions, the resistance situation of P. aeruginosa remains severe, with an overall prevalence of multidrug-resistant P. aeruginosa (MDR-PA) reaching 46.0%, extensively drug-resistant P. aeruginosa (XDR-PA) at 19.6%, extended-spectrum β-lactamase-positive P. aeruginosa (ESBL-positive PA) at 33.4%, and metallo-β-lactamase-positive P. aeruginosa (MBL-positive PA) at 16.0%.³ Among Chinese patients with hospital-acquired pneumonia, the infection rates of MDR-PA and XDR-PA were as high as 44.4% and 18.2%, respectively.¹² And the PA resistance rates also vary among different populations, with ICU patients showing the MDR-PA infection rate of 48.7%.¹³ Among various pneumonia patients, ventilator-associated pneumonia patients exhibit the highest MDR-PA infection rate, followed by hospital-acquired pneumonia patients.¹²

Mortality associated with P. aeruginosa infections remains a critical concern,¹⁴ particularly in immunocompromised and hospitalized patients, with rates ranging from 25–39% for pneumonia and 18–61% for bacteremia.^15,16 Notably, infections caused by multidrug-resistant isolates correlate with even higher mortality rates (40–70%).^14–17

Therefore, infection risk prediction tools are critical for reducing mortality and alleviating the economic burden associated with MDR-PA infections. Previous predictive models for P. aeruginosa infections have each focused on different aspects. Wang et al developed a predictive model using Cox regression analysis and nomograms to estimate the risk of recurrent hemoptysis following bronchial artery embolization and P. aeruginosa infection, providing valuable clinical insights; however, this model did not specifically address antibiotic resistance.¹⁸ Tartof et al’s model specifically evaluates the resistance of P. aeruginosa to carbapenems and broad-spectrum β-lactams in hospitalized patients, with its applicability primarily centered on specific antibiotics.¹⁹ Wijit et al’s research focused on the risk of P. aeruginosa infection in community-acquired pneumonia,²⁰ which, while valuable for specific populations, does not cover hospital-acquired infections. Therefore, this study base on multiple dimensions such as demographic characteristics, antibiotic use, and laboratory examinations of hospitalized patients, and focuses on developing and validating a predictive model for the risk of MDR-PA pneumonia in hospitalized patients.

Materials and Methods

Study Design and Data Source

This retrospective study was conducted at Chengdu University Affiliated Hospital (a large tertiary comprehensive teaching hospital with over 1400 open beds in Southwest China) from January 1, 2023 to April 30, 2024. The analysis exclusively included bronchial lavage fluid specimens from patients with P. aeruginosa-positive cultures. Exclusion criteria comprised: 1) negative P. aeruginosa cultures, 2) specimens without bacterial culture, and 3) cases of recurrent infection (only the initial episode was retained) (Figure 1).

Figure 1 The flowchart of patients’ enrollment.

Patient demographics and clinical parameters were extracted from electronic medical records and analyzed using WHONET 5.6. Bacterial identification and antimicrobial susceptibility testing were performed with the MicroScan WalkAway 40 automated system, following standardized protocols for clinical microbiological specimen processing. All bronchoscopic procedures adhered to current guidelines for specimen collection and laboratory analysis. Quality control was ensured using P. aeruginosa ATCC27853 reference strains. Antimicrobial susceptibility interpretations complied with the most recent CLSI guidelines, with annual updates applied. The study protocol received ethical approval from the Institutional Review Board of Chengdu University Affiliated Hospital (PJ2024-048-01), with waiver of informed consent granted for retrospective data analysis. The research will follow the principles of the Declaration of Helsinki and the Measures for Ethical Review of Biomedical Research Involving Human Subjects issued by China’s National Health Commission.

Measures of Variables

The following variables were systematically extracted from electronic medical records:

Demographic parameters: Age and sex; Clinical characteristics: Smoking history; total hospital length of stay; ICU admission status; pre-sampling hospitalization duration; pre-sampling ICU stay duration; pre-existing respiratory disorders; major comorbidities (cardiovascular diseases, neurological disorders, diabetes mellitus, malignancies, renal insufficiency); ventilation modalities; mechanical ventilation duration; nasogastric intubation; indwelling urinary catheterization; arteriovenous catheterization; thoracic drainage; grade III/IV surgical procedures during hospitalization; antibiotic exposure profiles; and duration of systemic corticosteroid administration (intravenous or oral); Laboratory parameters: Leukocyte count; hemoglobin concentration; platelet count; high-sensitivity C-reactive protein (hs-CRP); interleukin-6 (IL-6); procalcitonin (PCT) levels; serum albumin; D-dimer; and antimicrobial susceptibility profiles for P. aeruginosa.

Definitions

Pneumonia is an acute lower respiratory tract infection caused by pathogens such as bacteria, viruses, or fungi, characterized by inflammatory cell infiltration and exudative changes in alveolar spaces, accompanied by fever, cough, and radiographic pulmonary infiltrates.²¹ Multidrug resistance is defined as acquired resistance to at least one agent from three or more antimicrobial classes: carbapenems, β-lactamase inhibitor combinations, cephalosporins, aminoglycosides, and fluoroquinolones.²² Carbapenem-resistant P. aeruginosa is defined as P. aeruginosa that is resistant to one or more carbapenem agents.²³ Definitions of Antimicrobial Susceptibility Categories for P. aeruginosa: Susceptible (S): A P. aeruginosa strain is classified as susceptible to a specific antimicrobial agent when the minimum inhibitory concentration (MIC) is ≤ the established susceptibility breakpoint, indicating that standard dosing regimens are highly likely to achieve clinically effective plasma and tissue concentrations for therapeutic success. Intermediate (I): A strain is designated intermediate when the MIC exceeds the susceptibility breakpoint but remains ≤ the resistance breakpoint. This category suggests uncertain therapeutic efficacy, as clinical response may be achievable only under specific conditions (eg, higher-than-standard doses or localized infections where the drug accumulates at elevated concentrations).Resistant (R): Resistance is confirmed when the MIC exceeds the resistance breakpoint, signifying a high probability of therapeutic failure even with maximum tolerated doses, due to established resistance mechanisms or inadequate drug penetration at the infection site.²⁴

Statistical Analysis

Continuous variables with normal distribution were described as mean ± standard deviation, while non-normally distributed variables were presented as median (interquartile range). Categorical variables were expressed as frequency (percentage). Comparative analyses between MDR-PA and DS-PA groups were performed using unpaired t-test or Wilcoxon rank-sum test for continuous variables, and Pearson’s chi-square test or Fisher’s exact test for categorical variables. The Least Absolute Shrinkage and Selection Operator (LASSO) regression technique was applied for data dimensionality reduction and predictor selection. A multivariate logistic regression analysis was employed to develop the MDR-PA prediction model and corresponding nomogram. Internal validation was conducted through bootstrap resampling (resamples=1000). The model’s discriminative performance was assessed using receiver operating characteristic (ROC) curves and area under the curve (AUC). All statistical analyses were performed using R software (version 4.4.1), with statistical significance set at P<0.05.

Results

Baseline Characteristics

During the study period (January 1, 2023–April 30, 2024), 3132 bronchoalveolar lavage fluid (BALF) specimens were initially collected. After applying inclusion/exclusion criteria, 209 patients with P. aeruginosa-positive cultures were enrolled, comprising 44.98% (94/209) multidrug-resistant (MDR) strains and 55.02% (115/209) non-MDR (drug-sensitive, DS) strains. The demographic and clinical characteristics of the study cohort are summarized in Table 1. Antibiotic usage are presented in Table 2, with laboratory findings documented in Table 3. Univariate analysis revealed statistically differences between the two groups (P<0.05):age, total hospitalization duration, hospitalization duration before specimen collection, cor pulmonale, cardiac functional insufficiency, remote cerebral infarction, chronic renal insufficiency, ventilator support, ventilator use duration, mechanical ventilation, tracheostomy, nasogastric tube insertion, urinary catheter insertion, arteriovenous catheterization, thoracic drainage tube, cefoperazone-sulbactam exposure, aminoglycosides administration, imipenem-cilastatin therapy, linezolid use, combination therapy (≥2 antibiotics),concomitant antifungal treatment, leukocyte count, neutrophil percentage, hemoglobin level, serum albumin concentration, procalcitonin (PCT) level, d-dimer value.

Table 1 Demographic and Clinical Characteristics of Study Participants

Table 2 Antibiotic Usage of Study Participants

Table 3 Laboratory Findings of Study Participants

Construction of Predictive Models

1) Univariate Analysis: After conducting univariate analysis on 66 variables, we identified 26 variables that were statistically significant. These variables covered demographic and clinical characteristics, antibiotic usage, and laboratory findings. 2) LASSO Regression Analysis: Although the 26 variables identified in the univariate analysis were statistically significant, not all of them had a substantial impact on the outcome. Therefore, we chose to use LASSO regression analysis to select the most important predictors from the pool of potential variables. LASSO regression effectively reduces model complexity by penalizing less important variables, thereby enhancing the model’s generalizability and preventing overfitting. In this study, we employed cross-validation for predictor selection (Figure 2) and determined the tuning parameter (lambda) based on both the minimum criterion (left dashed line) and the 1-SE criterion (right dashed line). Ultimately, we followed the 1-SE criterion (right dashed line) for predictor selection (Figure 3). Through LASSO regression analysis, we selected 13 predictors, including age, total hospitalization duration, comorbidities (cor pulmonale, heart failure, old cerebral infarction, chronic renal insufficiency), clinical interventions (mechanical ventilation requirement, tracheostomy status, indwelling gastric tube), antimicrobial therapies (cefoperazone-sulbactam, aminoglycoside antibiotics, antifungal agents), and hemoglobin levels. 3) Model Development and Validation: Patients included in the study were randomly divided into training and validation cohorts (7:3 ratio) using a random number table In the training cohort, we developed a multivariable logistic regression model to predict the risk of MDR-PA infection using the variables selected by LASSO regression. The model was subsequently validated in the validation cohort. The model demonstrated good discriminative ability, with an area under the ROC curve (AUC) of 0.853 (95% confidence interval [CI], 0.793–0.914) in the training cohort and 0.972 (95% CI, 0.931–1.0) in the validation cohort (Figures 4 and 5). To make the model more intuitive and clinically applicable, we constructed a nomogram (Figure 6). The nomogram converts the regression coefficients of each predictor into scores, sums them to obtain a total score, and then translates the total score into a predicted probability, providing clinicians with a practical tool for estimating infection risk. 4) Model Calibration: The calibration performance of the model was assessed using the Hosmer-Lemeshow test, which showed non-significance (p=0.989) (Figure 7), indicating a close agreement between predicted and observed outcomes in the training cohort. These results suggest that the model has good discrimination and calibration consistency, effectively identifying high-risk patients for MDR-PA infection and providing strong support for clinical decision-making.

Figure 2 Selection of predictive factors using LASSO regression analysis with cross-validation.

Figure 3 Tuning parameter (lambda) selection for LASSO regression deviance based on the minimum criterion (left dashed line) and the 1-SE criterion (right dashed line).

Figure 4 AUC of the predictive and validation models-train.

Figure 5 AUC of the predictive and validation models-test.

Figure 6 Nomogram for assessing multidrug-resistant Pseudomonas aeruginosa infection risk.

Figure 7 Calibration curve of the predictive model.

Discussion

The threat of multidrug-resistant MDR-PA infection is on the rise, and there is an urgent need for predictive tools to guide early clinical intervention. Utilizing data from a comprehensive tertiary hospital in Southwest China, we developed and validated a nomogram that integrates multidimensional information, including demographics, clinical invasive procedures, antibiotic exposure, and laboratory tests, to predict MDR-PA pulmonary infection.

In recent years, the prevalence of P. aeruginosa has surged dramatically, raising concerns. Our study found that MDR-PA accounted for 44.98% of all P. aeruginosa-infected patients, which is consistent with the 2025 study by Zulkifli et al, reporting a prevalence of 46.0% for MDR-PA strains in Asia and Africa.³ This study developed a predictive model for MDR-PA infections using exclusively BALF specimens. Compared to prior studies incorporating heterogeneous specimen types, the sterile collection protocol of BALF effectively mitigates oropharyngeal flora contamination,²⁵ thereby significantly enhancing the specificity of lower respiratory tract pathogen detection. This specimen standardization strategy improves the model’s applicability to severe infection scenarios, such as ventilator-associated pneumonia.²⁶ However, the requirement for bronchoscopy in BALF acquisition necessitates the exclusion of hemodynamically unstable patients and non-severe cases, which may result in potential underestimation of MDR-PA infection risks in specific populations.

The risk factors for MDR-PA infection identified in this study are in line with global epidemiological trends while revealing regional specificity. Age is one of the key host factors influencing the risk of MDR-PA pneumonia. The results show that the risk of MDR-PA pneumonia increases with age. From a pathophysiological perspective, the age-related decline in immune function may be the reason why elderly patients are susceptible to MDR - PA.^20,27 Moreover, elderly patients often have multiple underlying diseases, which increase the burden of comorbidities. These diseases not only disrupt the respiratory barrier function but also exacerbate the drug resistance of P. aeruginosa through long-term use of corticosteroids/antibiotics. Our study found that cor pulmonale, heart failure, chronic renal insufficiency, and old cerebral infarction increase the risk of MDR - PA pneumonia. Among them, old cerebral infarction as a risk factor is particularly noteworthy, as this finding has been underestimated in previous literature.^27,28 This association may stem from chronic neuroinflammation or post-stroke immune suppression, which creates a favorable environment for the colonization or infection of MDR - PA.

Prolonged hospitalization is one of the significant risk factors for MDR-PA pneumonia. Our study revealed that the risk of MDR-PA pneumonia in long-term hospitalized patients is significantly higher than that in short-term hospitalized patients, with a higher isolation rate of MDR-PA. As the length of hospital stay increases, patients are exposed to a greater variety of pathogens in the hospital environment, including drug resistant bacteria.²⁹ In long-term care hospitals, the multidrug resistance rate of P. aeruginosa is higher than that in general hospitals.³⁰ Reducing the length of hospital stay is an important strategy to decrease P. aeruginosa infections.³¹ However, long-term hospitalized patients often have multiple underlying diseases and complex medical conditions, which may also have a confounding effect on the results.²⁹ Future studies need to further control these confounding factors to more accurately assess the relationship between prolonged hospitalization and MDR-PA pneumonia.

Invasive procedures also impact the occurrence of MDR-PA pneumonia, with ventilator support and tracheostomy being significant risk factors.³² Studies have shown that P. aeruginosa is one of the main pathogens of ventilator-associated pneumonia (VAP), and the emergence of MDR-PA has made the treatment of VAP more complex and difficult.³³ In patients with mechanical ventilation, intubation or tracheostomy disrupts the natural defense barrier of the respiratory tract, increasing the chances of P. aeruginosa entering the lower respiratory tract.³⁴ Meanwhile, the surfaces of medical devices such as ventilator tubing are prone to biofilm formation, in which drug-resistant bacteria can survive and reproduce, further increasing the risk of infection.³⁴ MDR-PA infection is associated with ventilator treatment failure,³³ and the crude mortality rate of VAP patients caused by MDR-PA is as high as 42.1%–87%, which remains high even with appropriate antibiotic therapy.³⁴ Notably, in this study, nasogastric tube placement was one of the 13 independent predictors identified, indicating its clinical predictive value. Nasogastric tube placement directly damages the neuromuscular coordination of the pharynx, weakening the swallowing reflex and increasing the risk of aspiration.³⁵ Aspirated gastric acid and bile can damage the mucosal barrier of the lower respiratory tract, creating conditions for MDR-PA colonization. According to research results in the literature, the use of nasogastric tubes (enteral/nasogastric tubes) is one of the independent risk factors for carbapenem-resistant P. aeruginosa.³⁶ In a study of 157 patients with P. aeruginosa, 78.26% had used enteral/nasogastric tubes.³⁶ Previous studies reported a significantly higher frequency of nasogastric tube use among patients infected with non-mucoid P. aeruginosa, with 18.5% of these non-mucoid isolates being carbapenem-resistant P. aeruginosa, suggesting that nasogastric tube placement may serve as a risk factor for non-mucoid strain acquisition.³⁷

Although carbapenem resistance remains a major concern in high-income countries, our cohort study showed that, compared with resistance to cephalosporins/aminoglycosides, carbapenem resistance has relatively lower predictive value for resistance phenotypes. This difference may reflect the unique antimicrobial management landscape in China. Under the national antibiotic restriction policy, the use of carbapenems is strictly regulated, while third-generation cephalosporins (eg, cefoperazone-sulbactam) are still widely used for respiratory infections. This regional prescription pattern may drive different resistance trajectories.^8,38

Notably, hemoglobin decline emerged as an independent predictor of MDR-PA infection, corroborating the emerging “host iron metabolism-bacterial virulence” axis proposed in recent studies.³⁹ This finding resonates with Peng et al’s 2024 report linking reduced admission hemoglobin levels to increased 30-day mortality in septic patients.⁴⁰ Physiologically, hemoglobin reduction often reflects chronic disease burden or immunocompromised status. While host iron restriction mechanisms typically inhibit pathogen proliferation through free iron sequestration, chronic anemia may compromise this defensive strategy.⁴¹ Existing MDR-PA prediction models mainly focus on inflammatory biomarkers (CRP/PCT) or antibiotic exposure history. By incorporating hemoglobin levels, we expanded the predictive framework, which may enhance risk stratification for multidrug-resistant infections.¹⁹

The limitations of this study should be acknowledged. First, the single center retrospective design introduced potential selection bias, especially regarding the exclusion of non-BALF specimens. Second, the relatively small sample size (n = 209) and the geographical scope limited to Southwest China may restrict the generalizability. Third, although internal validation supported the robustness of the methodology, external validation through multicenter prospective cohorts is still crucial for confirming clinical applicability. Lastly, this study was limited to MDR-PA pulmonary infection, and the applicability of similar models to other respiratory pathogens, such as Streptococcus pneumoniae and Klebsiella pneumoniae, needs to be explored.^42,43 Future research should address these limitations while investigating the mechanistic relationships between iron homeostasis, chronic cerebrovascular pathology, and the pathogenesis of MDR-PA.

Conclusion

The nomogram developed and validated in this study integrates host comorbidities, iatrogenic interventions, and antimicrobial exposure to predict the risk of MDR-PA pneumonia, which is of importance in clinical practice. The nomogram enhances the early identification of high-risk patients and optimizes antimicrobial stewardship in regions with cephalosporin/aminoglycoside-dominated resistance patterns, thereby reducing the incidence of MDR-PA. Additionally, the nomogram can serve as a reference for formulating public health policies and infection control guidelines. Future research should focus on validating the nomogram in multicenter cohorts and exploring its integration into clinical decision-making systems.