Study design and population
A multi-center retrospective cohort study was conducted during November 10th 2016, and December 20th 2023 in National Taiwan University Hospital and three affiliated branches in Taiwan. We considered the inclusion of consecutive patients who (1) had probable PJP based on EORTC/MSGERC definition11 (presence of appropriate host factors, clinical, and radiologic criteria, in addition to amplification of Pneumocystis jirovecii deoxyribonucleic acid by quantitative real-time polymerase chain reaction in respiratory specimen), (2) deemed pathogenic instead of colonization and thus received anti-Pneumocystis jirovecii agent (trimethoprim/sulfamethoxazole or anidulafungin) at a therapeutic dose, and (3) were followed till death or for 60 days regardless remaining hospitalized or discharged, whichever occurred first. Exclusion criteria were patients with HIV, pediatric patients aged less than 18 years, and individuals who were initially treated with anidulafungin as the primary agent for Pneumocystis jirovecii treatment. Each patient was included only once, even if recurrent Pneumocystis jirovecii infections were present.
Ethical issues
The study was approved by the National Taiwan University Hospital Research Ethics Committee (202002106RIND). The study was conducted in accordance with the Declaration of Helsinki, and the need for informed consent was waived by the institutional review board of National Taiwan University Hospital Research Ethics Committee owing to the retrospective nature of the study.
Data extraction
Patient-level characteristics were obtained from a linked population-based administrative electronic medical records system. We extracted the demographic parameters (age, sex, body height, body weight, body mass index), disease severity and corresponding treatment (initial cyclic threshold of Pneumocystis jirovecii polymerase chain reaction, hospital length of stay, shock, arterial oxygen tension/inspiratory oxygen fraction ratio (PaO2/FiO2), respiratory failure, mechanical ventilation, intensive care unit admission, in-hospital mortality, trimethoprim dose per body weight), chronic underlying conditions (hypertension, diabetes mellitus, stroke, heart failure, chronic kidney disease, dialysis, cirrhosis, asthma, chronic obstructive lung disease (COPD), intestinal lung disease, hematological malignancy, peripheral blood stem cell transplantation, solid cancer, solid organ transplantation, autoimmune disease, HIV), and selected laboratory data (absolute leukocyte count, hemoglobin, platelet, C-reactive protein, lactate dehydrogenase) based on the planning case record form. The type, dose, and time of initiation of adjunctive steroid as well as exposure of steroid with its type and dosage prior to PJP diagnosis were documented. The average steroid dosage was recorded as the methylprednisolone equivalent after conversion of anti-inflammatory effects.
PJP is diagnosed based on clinical (fever, cough, dyspnea, or hypoxia), imaging (diffuse interstitial or alveolar pattern bilaterally on chest plain film; nodular or patchy ground-glass opacities on computed tomography)12, and microbiological criteria (positive automated real-time quantitative polymerase chain reaction assay (Becton Dickinson, Franklin Lakes, New Jersey, USA) of sputum, endotracheal aspirates, or bronchoalveolar lavage samples for Pneumocystis jirovecii detection)13,14. The qualitative polymerase chain reaction for Pneumocystis jirovecii was available throughout the study period.
Enrolled patients were defined as steroid users if any kind of steroid had been introduced adjunctively after the initiation of Pneumocystis jirovecii treatment, regardless of baseline steroid administration. Shock is defined pragmatically as systemic arterial hypotension (mean arterial pressure < 70 mmHg, systolic blood pressure < 90 mmHg, or use of vasopressors/inotropic agents necessary for circulatory support). Respiratory failure was defined as tachypnea (respiratory rate exceeding 30 breaths per minute) and pronounced hypoxemia (PaO2/FiO2 ≤ 200)15. SpO2/FiO2 was recorded as a surrogate of PaO2/FiO2 based on the established formula if PaO2 is not available16.
The primary outcome was all-cause mortality at 60 days, which has been validated as a widely used timeframe in previous randomized controlled trials on steroid use in patients with severe community-acquired pneumonia17 and ARDS18. As other clinical endpoints, including respiratory failure and shock, were parameters for propensity score matching (PSM), they were not included as study outcomes.
Statistical analysis
Statistical analyses were conducted using R software, version 4.3.3 (R Core Team, 2024). Continuous variables are expressed as means with standard deviations, whereas categorical variables are expressed as absolute numbers with percentages. Variables with a P value < 0.15 in the univariable logistic or Cox regression analysis were included in the multivariable model using backward conditional selection and expressed as odds ratio (OR) or hazard ratio (HR) with 95% confidence interval (CI). Subgroup analyses were conducted based on prespecified baseline conditions, which have been reported to alter the prognosis of PJP. The cohort was stratified according to age, sex, shock, respiratory failure, fungal load, lymphocyte count, solid cancers, and autoimmune diseases. Cutoffs of lymphocyte count and cycle threshold of Pneumocystis jirovecii polymerase chain reaction were determined by respective median. Relative risk (RR) with corresponding 95% CI were reported. The Kaplan-Meier method was used to illustrate the accumulation of mortality events to assess the effect of adjunctive corticosteroids, and the performance of subgroups stratified by PaO2/FiO2 and presence of shock were assessed. Besides, propensity score matching was applied to minimize confounding by indication when unbiased causal therapeutic effects were determined. The propensity score has been well established for quantifying the probability of a certain treatment being assigned according to the observed baseline characteristics19. A total of 23 potential confounders were used a priori to estimate the propensity scores (Supplementary Table S1). Calibre was designated as 0.2 standard deviation of the logit propensity score for 1:1 matching20. The standardized mean differences before and after matching were depicted by Love plot. Inverse probability of treatment weighting was performed as a sensitivity analysis to the primary analysis method21. All comparisons with two-tailed P values less than 0.05 indicated statistical significance.
