PCP is diagnosed by identifying P. jiroveci in respiratory specimens, but it is often difficult to obtain BAL samples in non-HIV patients with severe and rapidly progressive disease [3, 26]. Highly sensitive PCR allowed us to detect P. jirovecii in samples with negative direct examination tests and in non-invasive samples, such as induced sputum  or oral wash . Although PCR for P. jiroveci can be positive in asymptomatic colonized patients, PCP therapy is recommended in patients with clinical suspicion of PCP and positive PCR results . According to these backgrounds, our diagnostic criteria included presumptive treatment due to strong suspicion by a clinician for PCP, in addition to the PCR and imaging studies. We excluded 2 patients for whom treatment was initiated after the PCR result. Because it usually takes 3 to 10 days to get PCR results in our practice, a delay in treatment would have some effect over outcomes. PCR-positive patients without treatment had clinical diagnoses other than PCP. PCR-negative patients irrespective of receipt of treatment were also excluded, because PCR in non-HIV patients has a quite high negative predictive value .
The main risk factors for PCP are deficiencies in cellular immunity and the use of immunosuppressive agents, especially corticosteroids. Similar to previous reports [28, 29], even with a low dose or short duration of corticosteroids, our patients developed PCP. In addition to corticosteroids, the introduction of high-intensity immunosuppressive regimens such as anti-tumor necrosis factor-α drugs has increased in inflammatory diseases. Post-marketing surveillance of anti-tumor necrosis factor-α drugs revealed a much higher incidence of PCP in Japan than in the US . Six PCP patients (7.3%) in this study received anti-tumor necrosis factor-α drugs. Reported attack rates of PCP in non-HIV patients not receiving prophylaxis are highest in hematological malignancies and transplantation [28, 31]. In our study, almost two-thirds of PCP patients had inflammatory diseases and fewer patients had hematological malignancies or transplantation. This may be because PCP prophylaxis is common in hematological malignancies or transplantation, but not in inflammatory diseases, and the number of patients with inflammatory diseases in the general population is larger.
Elevated serum LDH levels have been noted in patients with PCP  but were not specific. β-D-glucan was reported to be elevated in 96% of 111 HIV patients with microscopic diagnosis of PCP  and 78% of 18 non-HIV patients with a PCR-based diagnosis . The latter value is almost the same as our result (80%). Some reports have found high β-D-glucan to be a predictor of death [32, 33], while others have not [34, 35]. The β-D-glucan value did not differ between survivors and non-survivors in our study (Table 2). In a study of 21 non-HIV PCP patients of whom 2 died within 30 days, it was reported that β-D-glucan a median of 3 days after treatment had significantly decreased . In our study, β-D-glucan in survivors and non-survivors was separately analyzed, and in both groups it was significantly reduced after treatment. However, normalization occurred only in 33% of survivors, and in some patients from both groups, β-D-glucan increased after treatment. In a report of 42 HIV PCP patients, normalization after 3 weeks of successful treatment occurred in only 17% of patients, and elevated levels were found in 21% of patients . These results suggest that a decrease of β-D-glucan simply corresponds with treatment, but does not predict treatment response. Since β-D-glucan is a cell wall component of P. jirovecii, the level of β-D-glucan may reflect the organism burden directly, not the severity of PCP. Limper et al. demonstrated that mortality was not associated with organism burden, but rather the number of neutrophils in the lower respiratory tract . When β-D-glucan increases compared to the level before treatment, it may not indicate deterioration of PCP.
Non-HIV immunocompromised patients often develop invasive aspergillosis . We found 5 patients who developed probable aspergillosis. Although Aspergillus has not been generally recognized as a co-pathogen of P. jirovecii, it is possible that it is because both PCP and aspergillosis develop with the use of immunosuppressive agents [28, 29, 36]. Aspergillosis was found to be a significant predictor of death in univariate analysis and almost reached significance in multivariate analysis. Aspergillosis should be considered when there is a poor treatment response to PCP.
Twenty-four percent of our patients died, which is similar to the reported mortality rate of 19 to 40% [1, 29, 37]. We found that albumin and mechanical ventilation were independent predictors. These factors have been reported previously [29, 37].
The epidemiology of PCP in large numbers of non-HIV patients separate to that of HIV patients has never been reported. From a previous small Japanese study, 2 of 8 non-HIV patients had the DHPS mutation . Our patients had a very low rate of DHPS mutations, which cannot be generalized for non-HIV patients, because mutant rates in HIV patients vary widely (3% to 69%) . In addition, the prophylactic use of TMP-SMX may be associated with the emergence of DHPS mutant strains in HIV patients . One possible explanation of the low rate of DHPS mutants in our patients is the rarity of prophylaxis.
There is little information about the clinical background of the mt LSU rRNA genotypes. The only study describing the clinical relationships in 53 HIV and 7 non-HIV patients found no association . We also found no association in our patients. The largest study that has investigated the genotypes of non-HIV PCP included 17 patients from 3 PCP clusters in Sweden . Genotypes 1, 2, 3, and mixed-types were found in 5, 2, 3, and 7 patients, respectively. Other studies for HIV PCP, non-HIV PCP, and non-HIV patients without pneumonia found that genotypes 1 or 2 were the most prevalent, and genotype 4 was the least prevalent [20, 40, 42–45]. Genotype 1 was the most prevalent in our study, which is consistent with those reports. However, it is inconsistent with regard to genotype 4, which was the second prevalent type in this study.
The clinical correlation with the ITS genotypes has been conflicting [14, 46]. We did not find a correlation. Some studies have focused on the epidemiology of the ITS genotype in non-HIV PCP [40, 47], but the number of patients was less than 10. We investigated 31 patients and found Eb, Eg, and Bi from multiple patients as a single haplotype. Mixed infections were detected in 16%. Genotypes Eb, Eg, Bi, and Ne are reported to be common in HIV PCP, and mixed infections are reported in 5-79% [20, 22–24, 40, 48]. Those previous studies overestimated the mixed infections, because many of the less common haplotypes resulted from in vitro recombinations of globally common ITS haplotypes present in the same sample .
Our study is retrospective, and it has several limitations. One limitation is the heterogeneity of the respiratory sample and underlying clinical conditions. More than half of the samples obtained for PCR were not BAL fluid, and not all of the samples underwent microscopic analysis. Of the PCR-diagnosed PCP, microscopic examinations were positive only in 24% of BAL fluid and in 6% of sputum. One possible reason for the low positive rate is that Gomori methenamine silver stain has less sensitivity than an immunofluorescent antibody stain . The underlying clinical conditions and stage of disease might have varied, which could be uncontrolled confounders of the cause of death. Another limitation is that our study may still lack the statistical power to determine a clinical relationship between genotypes and outcome, although it is the largest to date on genotypes of non-HIV PCP.