The appropriate ethics committee approved this study; informed consent was not required because of the observational design. For the study, the head mycologist at each of 17 university-affiliated hospitals in France prospectively included consecutive patients with confirmed PCP who were admitted during January 1, 2007–December 31, 2010. We defined confirmed PCP as a positive result for Pneumocystis jirovecii by Gomori-Grocott or toluidine blue stain or positive immunofluorescence test results (4) for a bronchoalveolar lavage (BAL) fluid or induced sputum specimen. Induced sputum testing or bronchoscopy with BAL was performed at the discretion of the clinicians by using previously described procedures (13,14). We did not include patients for whom only PCR results were positive.

For all included patients, chest radiographs were obtained; computed tomography (CT) scans were performed when deemed necessary by clinicians. Bilateral interstitial or alveolointerstitial opacities on chest radiographs and diffuse ground-glass opacities on CT images were considered typical findings for PCP. Septal lines and centrolobular nodules were also interpreted to support a diagnosis of PCP. Focal consolidation, pleural effusion, subpleural nodules, and cavitation were considered atypical findings. Criteria for microbiologically documented pneumonia were as follows: clinical symptoms of pneumonia; pulmonary infiltrates; and >1 positive noncolonizing microbiological sample (i.e., blood culture, tracheal aspirate, sputa examination, BAL, protected sample, or pleural fluid). Urine antigens (Streptococcus pneumoniae and Legionella pneumophila) were included in routine testing for the microbiological documentation of pneumonia. When the microbiological data were negative but the patient had symptoms of pneumonia and pulmonary infiltrates, the case was classified as clinically documented pneumonia (13).

Data were collected prospectively. Steroid treatment was either high dose (>1 mg/kg for >1 mo) or long term (>3 mo at any dose) (3,14). Mycologists and study investigators obtained missing and follow-up data by reviewing patients’ medical charts and by interviewing the specialists who provided usual care to the patients. Bacterial, viral, fungal, and parasitic infections were diagnosed on the basis of criteria reported elsewhere (13). Information on PCP prophylaxis was recorded; trimethoprim-sulfamethoxazole (TMP-SMX), aerosolized pentamidine (1×/mo), dapsone, and atovaquone were classified as effective prophylaxis options (3,4). Information on treatments used for PCP and the time from admission to treatment initiation were recorded; TMP-SMX, pentamidine, atovaquone, dapsone, and clindamycin-primaquine were considered acceptable options for PCP treatment (4). Shock was defined as persistent hypotension despite appropriate fluid load, requiring treatment with a vasopressive drug.

Steroids as adjunctive therapy were used on the basis of standard protocol recommendations for patients with AIDS at a dose that depended on patient location (e.g., medical unit). In deeply hypoxemic, critically ill patients, steroids were implemented at the dosage of 240 mg/day for 3 days, then at 1 mg/kg/day for 7 days, followed by tapering doses to be stopped before day 21 (15,16). In patients who were less critically ill, the dose was 1 mg/kg/day followed by a tapering dose after day 7 to be stopped before day 21. Similar protocols were used for AIDS and non-AIDS patients.

The variables in the dataset are described or summarized by using either median and interquartile range or number and proportion of the total (%). Categorical variables were compared by using the Fisher exact test and continuous variables by using the nonparametric Wilcoxon test or Mann-Whitney test for pairwise comparisons. All tests were 2-sided, and p<0.05 was considered statistically significant. Kaplan-Meier curves and log-rank tests were used to compare hospital death rates between groups of patients. Logistic regression analysis was used to identify variables significantly associated with hospital deaths by estimating the odds ratios (OR) with 95% CIs. Variables yielding p values <0.20 in the univariable analyses were entered into a multivariable logistic regression model with stepwise variable selection using an automatic procedure based on the Akaike Information Criterion, with hospital deaths as the variable of interest. The co-variates were entered into the model with a p value cutoff for removal of 0.1. Co-linearity and interactions were tested. For the multivariable analysis, missing data were handled by using multiple imputation with chained equations. The imputed missing data focused on 3 variables: oxygen saturation on admission (35 patients), time to treatment (17 patients), and time from respiratory symptom onset to diagnosis (11 patients). The Hosmer-Lemeshow test was used to check goodness-of-fit of the logistic regression model.

We included 544 patients in the study. Median age was 51 (interquartile range 40–62) years, and 370 (68%) were men A total of 223 (41%) patients had AIDS and 321 (59%) had other immunosuppressive conditions (Figure 1). Among patients with AIDS, PCP was the first manifestation of AIDS for 105 (44.8%); only 4 (5.0%) had CD4 lymphocyte counts >200 cells/mm³. As shown in Table 1, the main causes of immunodeficiency in the non-AIDS patients were SOT (n = 99, 30.8%), chiefly of a kidney (80/99); and hematologic malignancies (n = 84, 26.2%), chiefly lymphoproliferative diseases (72/84). Twenty-seven (8.4%) patients were HSCT recipients (14 allogeneic, 13 autologous); 65 (20.2%) had autoimmune or chronic inflammatory disease; and 46 (14.3%) had solid-organ malignancies.

At the time of PCP diagnosis, high-dose or long-term steroid therapy was the most common immunosuppressive treatment for non-AIDS patients: 88% (n = 63) for those with SOTs, 85% (n = 71) for those with hematologic malignancies, 100% (n = 13) for those with HSCTs, 91% (n= 41) for those with solid-organ malignancies, and 89% (n = 52) for those with autoimmune or chronic inflammatory disease. SOT recipients were also receiving anticalcineurin agents (n = 59, 82%), purine inhibitors (n = 56, 77%), rituximab (n = 1, 6%), mechanistic target of rapamycin inhibitors (n = 6, 8%), tumor necrosis factor–α antagonists (n = 3, 4%), or intravenous immunoglobulins (n = 1, 1%). Patients with hematological malignancies also had received rituximab (n = 29, 35%) and fludarabin (n = 8, 9.5%). All HSCT patients were receiving anticalcineurin agents. Patients with autoimmune or chronic inflammatory diseases were receiving methotrexate (n = 14, 21.5%), rituximab (n = 6, 9.2%), tumor necrosis factor–α antagonists (n = 6, 9.2%), anticalcineurin agents (n = 5, 7.7%), or purine inhibitors (n = 4, 6.1%). (Note that values are approximate; data are missing among each subpopulation.)

The clinical manifestations of PCP in AIDS and non-AIDS patients are shown in Table 2 and Figure 1. Overall, 96% of patients were not receiving PCP prophylaxis. The median time from the onset of respiratory symptoms to PCP diagnosis was significantly shorter for non-AIDS patients (5 [range 1–15] days) than for AIDS patients (21 [7–30] days) (p<0.0001). However, as shown in Tables 1 and 2, hypoxemia was more severe in non-AIDS patients, who required higher oxygen flow rates (4 [range 3–5] L/min) than did AIDS patients (2 [1.3–3] L/min). Non-AIDS patients also more often required intensive care and noninvasive or invasive ventilation. Shock was more common in non-AIDS patients and was significantly associated with pulmonary microbial co-infection (OR 3.09 [95% CI 1.44–6.68]; p = 0.004), in agreement with the presence of microbiologically documented pneumonia (13) in half of the patients with shock. CT scans were obtained for 279 patients; typical findings were seen in 37 patients, of whom 29 were believed to have pulmonary bacterial microbial co-infection and 8 microbial co-infections with a second opportunistic microorganism.

A BAL sample was diagnostic for 87% and 97% of AIDS and non-AIDS patients, respectively (p = 0.0003). Overall, microbial co-infection as previously defined (13) was suspected or confirmed for 169 (31%) patients, including 92 (16.9%) with bacterial infection, 65 (6.6%) with viral infection, and 38 (6.9%) with fungal infection. For 32 (5.8%) patients, >2 microbial co-infections were identified; the most frequent site of microbial co-infection was the lung (Technical Appendix Table). Univariable analysis showed an association between microbial co-infection and death (OR 2.29, 95% CI 1.41–3.71) (Technical Appendix Table), but multivariable analysis did not confirm this finding (OR 1.99, 95% CI 0.91–4.33; p = 0.09).

Time from hospital admission to the start of treatment for PCP was longer for non-AIDS patients than for AIDS patients (2 [range 0–6]) days vs. 1 [0–2] days; p<0.0001). TMP-SMX was the first-line anti-PCP agent used for 97% of patients. However, patients with AIDS more often required a switch to second-line therapy than did non-AIDS patients (25.5% vs. 7%; p<0.0001); this change was most often made because of allergic reactions or hepatic or renal toxicity. Adjunctive steroid therapy was used in 40.4% of patients overall and in a significantly higher proportion of AIDS than in non-AIDS patients (55% vs. 43%; p = 0.01).

Hospital death data were available for 478 (88%) patients, and the percentage of hospital deaths was significantly higher for non-AIDS than AIDS patients (27% vs. 4%; p<0.0001). For non-AIDS patients, death rates varied by cause of immunodeficiency, from a low of 3.75% for kidney transplant recipients to a high of 43% for allogeneic HSCT recipients. Invasive mechanical ventilation was used in 28% and 43% of patients in these 2 groups, respectively (p<0.0001).

Of 7 variables independently associated with hospital death by multivariable analysis, 2 were associated with lower death rates: AIDS diagnosis (OR 0.33, 95% CI 0.12–0.92) and receipt of a SOT (OR 0.08, 95% CI 0.02–0.31) (Table 3). Five variables were associated with increased death rates: older age (OR 1.04/additional year, 95% CI 1.02–1.06); receipt of a HSCT (OR 8.6, 95% CI 1.40–53.02); need for oxygen on admission (OR 4.06, 95% CI 1.44–11.5); need for invasive mechanical ventilation (OR 16.70, 95% CI 7.25–38.47); and longer time from admission to initiation of PCP treatment (OR 1.11/additional day, 95% CI 1.04–1.18).

Improved cumulative survival was significantly associated with underlying condition (p<0.0001 for AIDS vs. non-AIDS comparison; Figure 2). Shorter time from admission to treatment initiation was also associated with improved cumulative survival (Figure 2).