In-hospital mortality from healthcare-associated infection by multidrug-resistant Pseudomonas aeruginosa: a competing risks analysis of a 4-year propensity-matched cohort study in southern China

Statistical analysis

Temporal changes in MDRPa and non-MDRPa HAI incidence were examined with a Poisson regression model to describe the variation of monthly incidence rates (number of infections per 1,000 hospitalisation days), for each major site of infection, with the time (in months) elapsed since the start of the study. The monthly series of hospitalisation days was used as an offset variable (log transformed) to account for the size of the hospital population and the length of hospital stay. This approach allowed the estimation of the mean monthly incidence rate ratio (IRR) and its 95% confidence interval (CI). IRR>1 indicates an upward trend and IRR<1 a downward trend of infection incidence over time. The mean monthly percentage change in infection incidence was calculated as (IRR–1)x100%.

A propensity score-matched sample was constructed to minimise bias when comparing in-hospital mortality rates between patients with MDRPa HAI and those with non-MDRPa HAI. Propensity scores were estimated by logistic regression, accounting for age, sex, department at hospital admission, site of infection, diabetes, immunocompromised status, and COVID-19 period. Matching 1:1 was achieved by applying a nearest-neighbour method with a calliper width of 0.2 times the standard deviation of the logit of the propensity score using the MatchIt R package [18], [19]. Standardised mean differences less than 10% were considered to indicate an acceptable balance of covariates between groups [18], [19].

Cumulative probabilities of a patient dying in the hospital before any given day were calculated using the Aalen-Johansen method [20]. In this analysis, being discharged alive was treated as a competing event to in-hospital death [21]. The results were illustrated by cumulative incidence function (CIF) plots produced with the cuminc function of the cmprsk package in R. When comparing mortality and discharge-alive rates between MDRPa and non-MDRPa HAIs, the effect sizes were expressed as subdistribution hazard ratios (sHRs) with 95% confidence intervals from the Fine–Gray model, through the FGR function of the risk Regression R package [22]. The sHRs described the relative effect of MDR status and other covariates on the subdistribution hazard functions for (thereby, the probabilities of) in-hospital death and discharge-alive [22]. A low sHR for discharge-alive (<1) indicates a reduced daily discharge rate, leading to prolonged hospitalisation.

The results from bivariable and multivariable Fine–Gray models were shown for both the original unmatched cohort and the propensity-matched sample. The multivariable regression analysis of the matched data, incorporating the variables used in the propensity-score model, can be regarded as a doubly robust adjustment [18]. Nine baseline covariates were included: age over 65 years, sex, admission diagnosis, department of admission, site of infection, receipt of empiric therapy, presence of diabetes, immunocompromised status, and year of infection occurrence. Multicollinearity of the covariates was ruled out by examining variance inflation factors (see Supplementary Table S1 in Attachment 1 [Att. 1]).

For all survival analyses, time zero was defined as the time of infection onset. For the 14-day, 30-day and overall hospitalisation outcomes, event-free time was administratively censored at 14, 30 and 120 days, respectively, for patients who remained hospitalised for longer periods. There were no missing data for any study variable. Statistical significance was considered when two-sided P < 0.05. R code for the main analyses is included in Supplementary Material for R code (Attachment 1 [Att. 1]).

Table 1: Baseline characteristics of the original and propensity-matched groups of patients infected with P. aeruginosa by multidrug resistance status

Figure 1: Incidence rate trends of healthcare-associated infection by multidrug-resistant P. aeruginosa, 2018–2021

Figure 2: Univariate cumulative function curves showing the probability of each event (in-hospital mortality and discharge alive) over time for MDRPa HAI and non-MDRPa HAI, in both original (unmatched) and propensity-matched groups.

Table 2: Estimated effects of the multidrug resistance status in P. aeruginosa infection on patient outcomes