Clinical burden of community-associated infections caused by multidrug-resistant Pseudomonas aeruginosa: a propensity-matched longitudinal cohort study in Southern China
Our latest research article was just published in GMS Hygiene and Infection Control (published by the German Society of General and Hospital Hygiene), October 23, 2024.
Clinical burden of community-associated infections caused by multidrug-resistant Pseudomonas aeruginosa: a propensity-matched longitudinal cohort study in Southern China
by
Mouqing Zhou, Baohua Xu, Zhusheng Guo, Yongfeng Zeng, Jiayao Lei, Evangelos I. Kritsotakis, and Jiancong Wang
Open Access to the full paper is available via:
Abstract
Background: Limited research has been conducted on the burden of community-associated infections caused by multidrug-resistant Pseudomonas aeruginosa (CA-MDRPa). We quantitatively modeled the incidence rate and clinical factors associated with CA-MDRPa among hospitalized patients in Southern China.
Methods: Data were obtained from the local nosocomial surveillance system. Poisson regression was applied to estimate annual incidence rate ratios (IRRs) from 2018 to 2021. After propensity-score 1:2 matching, multivariable conditional logistic regression was used to identify factors for CA-MDRPa upon admission and adverse clinical outcomes during hospitalization.
Results: 278 patients were clinically and microbiologically diagnosed with CA-MDRPa and 647 with CA-non-MDRPa. CA-MDRPa rate exhibited a slight, non-significant, increase during the research period (IRR=1.03; 95% confidence interval [CI], 0.93–1.15). Neurological conditions, cardiovascular diseases, respiratory disorders, urinary tract infections, and use of cefoperazone/sulbactam prior to admission were identified as risk factors for CA-MDRPa upon admission. CA-MDRPa upon admission was associated with ESBL-producing P. aeruginosa acquisition during hospitalization (odds ratio [OR], 2.70; 95% CI, 1.53–4.77) and increased in-hospital mortality (OR, 2.24; 95% CI, 1.17–4.28).
Conclusions: The findings emphasize the importance of regular targeted screening for CA-MDRPa upon hospital admission and offer valuable insights for strengthening infection control and antimicrobial stewardship programs.
Keywords: community-associated infections, Pseudomonas aeruginosa, multidrug-resistant pathogens, incidence density, age- and sex-specific, China, Dongguan
Statistical methods
Poisson regression model for the incidence rate ratio
The absolute numbers of newly diagnosed CA-MDRPa and CA-non-MDRPa cases were aggregated monthly from January 2018 to December 2021. We calculated the incidence densities at monthly intervals by dividing the total numbers of CA-MDRPa and CA-non-MDRPa cases each month by the total number of hospital patient admissions for that month. These rates were then expressed per 100 patient admissions. To model the incidence trends of CA-MDRPa and CA-non-MDRPa, we used a Poisson regression model, which included the natural logarithm of the number of hospitalized patient admissions as an offset and estimated the average annual relative changes in incidence from 2018 to 2021 expressed in terms of incidence rate ratios. The model's effectiveness was evaluated by examining its summary, including the dispersion parameter and 95% confidence intervals (CIs). These metrics allowed statistical significance and the impact of time on incidence rates to be assessed. We also separately modeled the incidence trends for two major types of community-associated infections – lower respiratory tract infections and urinary tract infections – caused by MDRPa and by non-MDRPa.
Propensity-score matching
To ensure balanced comparisons between the CA-MDRPa and CA-non-MDRPa groups, we employed propensity-score matching using a logistic regression model to estimate propensity scores, treating age, sex, and admitting departments as covariates. We used the MatchIt R package for 1:2 matching of the CA-MDRPa and CA-non-MDRPa groups using nearest-neighbor matching with a maximum caliper width of 0.2 times the standardized difference of the logit of the propensity score (Supplementary Material 1 in Attachment 1). Subclass values were created to identify matched pairs or groups from the propensity-score analysis, balancing the distribution of covariates between the CA-MDRPa treatment group and the CA-non-MDRPa control group. We also generated propensity-score density plots to graphically evaluate the matching quality by comparing the overlaps and similarities in the distributions of the propensity scores between the two groups. Unmatched data were not considered for further analysis.
Variable selection for multivariable analyses
The basis for variable selection was not only subject-matter information, as presented in the literature and systematic reviews, but also the availability of data from the Dongguan Nosocomial Infection Surveillance System. To ensure reliable estimation in the logistic regression analysis, we followed the criteria suggested by Peduzzi et al. who recommended that the number of events per variable be ten or greater to avoid biased regression coefficients in both the positive and negative directions.
Multivariable analysis using a conditional logistic regression model
To account for potential confounding covariates and ensure reliable estimation, we first used the clogit function from the Survival R package to conduct a multivariable analysis on the propensity-score-matched dataset with a conditional logistic regression model (Supplementary Material 2 in Attachment 1). In the model, CA-MDRPa was treated as the dependent variable, while age, sex, admitting department, admission diagnoses, infection sites, comorbidities, antibiotic use prior to admission, and admission during the COVID pandemic were treated as independent variables. The subclass variable was used as a stratification factor to account for matched pairs or groups. Statistical significance was defined as a two-tailed p-value of less than 0.05. Results from the conditional logistic regression model are presented as odds ratios (ORs) with corresponding 95% CIs. Furthermore, a mixed-effects logistic regression model was fitted using the glmer function from the lme4 R package. We finally compared the conditional logistic regression model with the mixed-effects logistic regression model using evaluation metrics such as the Akaike information criterion, Bayesian information criterion, and Log-likelihood value to assess the robustness and fitness of the models.
We also used multivariable analysis, again with a conditional logistic regression model, to assess the impact of CA-MDRPa and other clinical factors – treated as independent variables – on the probabilities of two important clinical adverse events: the development of ESBL-producing P. aeruginosa during hospitalization and in-hospital mortality, treated as dependent variables.
None of the study variables had missing data, and all the analyses were performed and graphs prepared using the statistical software R, version 4.3.2 (The R Foundation for Statistical Computing Platform). We also reported the age- and sex-specific distributions of absolute cases of CA-MDRPa and CA-non-MDRPa, consistent with the EARS-Net surveillance report.
Figure 2: Distributions of CA-MDRPa and CA-non-MDRPa cases stratified by age and sex
Table 1: Multivariable analysis of clinical factors associated with CA-MDRPa upon admission using conditional and mixed effects logistic regression models in the propensity-score-matched dataset (n=834 observations)