Background
Pneumocystis pneumonia (PCP) is a potentially life-threatening infectious disease that mainly occurs in immunocompromised hosts [
1]. PCP was first recognized as the most common opportunistic infection in patients infected with human immunodeficiency virus (HIV), but its incidence has fallen dramatically since the development of effective anti-retroviral treatment [
2]. However, PCP is still an important cause of atypical pneumonia in patients without HIV who have immunosuppressed conditions [
3,
4].
Incidence of PCP in patients with rheumatic diseases remains uncertain. Previous study reported that its prevalence in autoimmune diseases was 1~7% in systemic lupus erythematosus (SLE), 2~37.5% in inflammatory myositis, and 6~17% in Wegener’s granulomatosis [
5]. The broad range of prevalence among studies suggests that incidence of PCP could be different according to immunosuppressant use. However, all of these studies consistently reported high PCP-related mortality, which spanned between 30 and 80%. Considering that PCP in patients without HIV usually follows a fulminant course and has higher mortality than PCP in HIV-infected patients, evidence on incidence and efficacy of primary prophylaxis of PCP is required for better treatment outcome of rheumatic diseases [
6‐
9].
The most significant risk factor for PCP in patients without HIV is treatment with a moderate-to-high dose of glucocorticoid, which is the principal therapeutic agent for many rheumatic diseases [
6,
10]. We previously demonstrated that prophylactic trimethoprim–sulfamethoxazole (TMP-SMX) significantly decreases the incidence of PCP in patients with rheumatic diseases who receive prolonged (≥ 4 weeks), high-dose (≥ 30 mg/day prednisone) steroid treatment, with an acceptable safety profile [
11]. However, it is still uncertain whether prophylaxis is justified in patients who receive lower doses of glucocorticoid treatment. The incidence of PCP and its potential risk factors have not been adequately characterized in this group of patients, and reports have been limited to small case series and observational studies [
12‐
15].
The main objective of this study was to identify the risk factors which primary PCP prophylaxis is justified despite relatively lower dose of steroid treatment. We determined the incidence of PCP in patients with rheumatic diseases receiving various doses of steroids for ≥ 4 weeks and investigated the clinical features that affected the incidence of PCP.
Methods
Patients and treatment episodes
The electronic medical database at Seoul National University Hospital was analyzed for data collection and capturing study population. The institute is one of the largest tertiary referral centers in South Korea and covers patients living in all geographic areas of the country. The database was reviewed for the period between January 2004 and December 2017 to capture the “treatment episode,” which was defined as a clinical situation where a patient received steroid treatment within a specific dose range for ≥ 4 consecutive weeks. First, we collected the entire prescription data of oral/IV steroids. Then, according to the initial 4-week dose of steroid and prednisone-equivalent dose conversions, ≥ 2.5 to < 15 mg/day was defined as low dose, ≥ 15 to < 30 mg/day as medium dose, and ≥ 30 mg/day as high dose. The definition was based on the degree of steroid receptor saturation and previous studies reporting PCP cases in patients with steroid treatment [
6,
10,
12,
16,
17]. All captured episodes were then linked to the database of registered ICD-10 code of all patients in our institution. If a patient had more than one ICD-10 codes of rheumatic diseases, the patient’s medical records were reviewed to confirm the diagnostic code.
The starting date of each treatment episode was defined as the first day of particular dose of steroid treatment. The observation period following each treatment episode was 1 year, unless PCP or a censoring event (death or loss to follow-up) occurred. To enable precise determination of the effect of initial steroid treatment, multiple episodes in a single patient had to be spaced ≥ 1 year apart to be included in the study. Defining process of treatment episodes is summarized in Additional file
1: Figure S1. Patients with a history of PCP, HIV infection, current cancer, or a solid organ transplant, and those < 18 years old, were excluded. The ICD-10 codes for case identification are presented as Additional file
1.
Incidence of PCP in each dose group was determined, and the values were compared. If the incidence rate (IR) was < 0.1 per 100 person-years, the risk from PCP prophylaxis (previously calculated as the incidence of serious adverse drug reaction (ADR) related to TMP-SMX) [
11] was considered to clearly outweigh any potential benefit. Therefore, the efficacy of primary PCP prophylaxis was only assessed for PCP IR ≥ 0.1 per 100 person-years. PCP IR in the high-dose group was used for only comparison with that in each non-high-dose group.
Treatment episodes in the eligible dose group were classified into two groups (control or prophylaxis group) according to whether a patient received primary PCP prophylaxis concomitant with initiation of glucocorticoid treatment. The baseline date was the first day of specific dose of glucocorticoid treatment (control group) or TMP-SMX (prophylaxis group). Each patient received the corresponding dose of steroid for ≥ 4 weeks from the baseline date.
The primary outcome was the 1-year incidence of PCP in each group. Secondary outcome was the incidence of ADRs related to the prophylaxis. All suspected ADRs were reviewed by two authors (JWP and MJK) and assigned a probability of causation based on their timing compared with established patterns of specific ADRs. Among them, only those with probable/likely or certain causality were finally selected [
18].
The study was carried out in accordance with the Declaration of Helsinki and was approved by the institutional review board (IRB) of Seoul National University Hospital [IRB No 0907-062-287]. Consent was waived by the IRB because of the retrospective nature of the study.
Detection of clinically relevant PCP
An algorithm was constructed to enable detection of symptomatic PCP occurring during the observation period in patients with rheumatic disease (Additional file
1: Figure S2). All data for confirmatory microbiological tests for PCP (polymerase chain reaction and direct fluorescent antibody staining of induced sputum or bronchoalveolar lavage fluid) conducted from January 2004 to December 2017 were obtained from medical records. Medical records of patients with positive test results were further evaluated to ascertain whether they had clinical features of PCP infection. All diagnoses of PCP were confirmed based on (1) positive PCR and/or immunofluorescence result of induced sputum or bronchoalveolar lavage fluid, (2) whether the patient showed appropriate clinical features of PCP (acute onset of dyspnea, cough, or fever, along with characteristic radiographic findings), and (3) exclusion of other pulmonary infections. Characteristic radiographic findings of PCP were bilateral interstitial infiltrates, predominant in the perihilar regions and the apices [
19]. If a patient showed typical manifestations suggesting PCP but chest radiographs were inconclusive, high-resolution computerized tomography (HRCT) was evaluated to confirm the diagnosis. Positive microbiological tests without clinical manifestations consistent with respiratory infection were not considered to indicate PCP.
PCP prophylaxis
TMP-SMX was the only agent used for PCP prophylaxis in this study. Because there was no consensus regarding PCP prophylaxis in patients with rheumatic disease receiving steroid treatment, initiation and duration of prophylaxis was mainly determined by physicians’ judgment. In the prophylaxis group, patients started TMP-SMX concurrently with steroid treatment unless they had contraindications for use of TMP-SMX at this time. TMP-SMX was given as one single-strength tablet (400/80 mg) per day, and the dose was adjusted according to the patient’s creatinine clearance (one tablet three times weekly when creatinine clearance 15~30 mL/min). Second-line prophylactic agents such as dapsone, atovaquone, and aerosolized pentamidine were not used for primary prophylaxis against PCP during the observation period.
Statistical analysis
Continuous and dichotomous variables were analyzed using Student’s
t test and the chi-square test, respectively. The Cox proportional-hazards regression model was used for comparison of the incidence of PCP between the groups and for estimation of effects of clinical factors on the 1-year PCP incidence. The hazard ratio (HR) was adjusted for baseline clinical factors that showed a relevant association (
p < 0.1) with outcome. If an outcome variable showed a complete separation, Firth’s penalized maximum likelihood was used for reduction of statistical bias [
20]. In addition, the final model was adjusted for intra-cluster correlation using grouped sandwich variance estimates, as some patients may have undergone multiple treatment episodes. To define the high-risk subgroup for PCP, a penalized regression method with the least absolute shrinkage and selection operator (LASSO) was used to select variables that predict the outcome most precisely. Briefly, among the nine variables (age, sex, disease duration, initial steroid dose, concomitant cyclophosphamide treatment, high cumulative steroid use, interstitial lung disease, lymphopenia, and concomitant steroid pulse) potentially associated with PCP, cross-validation was used to find the model with the minimal value for tuning parameter lambda. Because all clinical variables used for analysis had less than 1% of missing data, all statistical analyses were performed without imputation of them.
All statistical analyses were performed using R 3.3.1 software (package glmnet for Cox regression with LASSO), and a p value < 0.05 was considered statistically significant.
Discussion
Glucocorticoid treatment is an essential part of the therapy for many rheumatic diseases, but is also an important risk factor for PCP. The association of high mortality with PCP means that a universal recommendation regarding PCP prophylaxis is important for the improvement of treatment outcomes, and this goal requires the establishment of an evidence-based threshold for the commencement of prophylaxis. To the best of our knowledge, this is the first study to investigate the incidence of PCP and its risk factors in patients with rheumatic diseases who received prolonged, medium-dose steroid treatment (≥ 15 to < 30 mg/day prednisone or equivalent).
In this study, the 1-year IR of PCP in the medium-dose group was 0.5 per 100 person-years, which was prominently lower than that in the high-dose group. However, the IR of PCP was highly variable and was affected by whether a patient had lymphopenia and/or had received concomitant steroid-pulse treatment at baseline. No PCP cases were associated with treatment episodes without these risk factors, suggesting that prolonged, medium-dose steroid treatment alone does not lead to sufficient immunosuppression for the occurrence of PCP. Previous experimental data have also demonstrated that medium-dose steroid treatment (unlike high-dose treatment) does not fully saturate the glucocorticoid receptor which leads to immunosuppression by genomic effect [
16,
21,
22]. In addition, the most previously reported cases of PCP in patients without high doses of steroids were associated with other immunosuppressive conditions, such as lymphopenia, or with the concomitant use of other immunosuppressants [
12,
23‐
25]. Taken together, our results suggest that a prolonged (≥ 4 weeks), 30-mg/day dosage of prednisone (or equivalent) is a relevant threshold for justification of PCP prophylaxis in patients with rheumatic disease who do not have any other risk factors.
Notably, our results demonstrated that treatment episodes for patients receiving prolonged medium-dose glucocorticoids with at least one risk factor (in particular, concomitant steroid pulse) were associated with a comparable PCP incidence to that in the high-dose group. This result was contrary to that of our previous study, in which steroid-pulse treatment did not increase the risk of PCP in patients receiving prolonged high-dose steroids [
11]. Results from previous studies of PCP in patients with rheumatic diseases showed that it usually occurs after > 4 weeks of steroid treatment [
6,
25,
26], suggesting that both duration and dose of the steroid treatment are important predisposing factors for PCP occurrence. Therefore, pulse treatment could increase the risk of PCP by a different mechanism from that of prolonged high-dose steroid treatment. This mechanism might involve acute immunosuppression and subsequent rapid immune reconstitution after the pulse treatment [
27]. It is also supported by some previous reports showing that rapid reduction of immunosuppressive agents is significantly associated with PCP in patients without HIV [
9,
28,
29].
In the current study, PCP incidence in the high-risk subgroup was comparable to that in the high-dose group, but TMP-SMX prophylaxis did not significantly reduce the 1-year PCP incidence. The number of treatment episodes with prophylaxis in the high-risk subgroup was small (n = 18), so this study did not have adequate power to demonstrate efficacy of prophylaxis. However, in a risk–benefit assessment, the NNT to prevent one case of PCP in the high-risk group was numerically lower than the NNH for serious ADR. Although the NNH was calculated for the entire medium-dose group (because the only serious ADR occurred in the non-high-risk subgroup), this result suggests that the clinical benefit of prophylactic TMP-SMX could outweigh the risk for serious ADR in high-risk patients. However, considering the wide ranges of 95% CI of NNT and NNH, this result should be replicated in the future studies.
It is also interesting that most ADRs related to TMP-SMX had mild-to-moderate severity, but prophylaxis was withdrawn in 75% of the episodes in which these ADRs occurred. It could be attributed to the physicians’ concern for possible progression to serious ADR such as Stevens–Johns syndrome. Results from previous studies also have reported a high incidence of ADR related to TMP-SMX, especially in patients with SLE [
30,
31]. However, considering the high mortality associated with PCP in patients with rheumatic diseases, risk–benefit analysis of prophylaxis should be based not only on the incidence of ADRs, but also on their severity.
Some limitations of this study should be considered. First, the number of treatment episodes with prophylaxis was too small to precisely estimate the prophylactic efficacy of TMP-SMX. It also led to wide 95% confidence interval of both NNT and NNH, which make the result less powered. Therefore, it should be confirmed in the future studies including greater number of patients with prophylaxis. However, our data clearly showed that PCP did not occur despite large-scale observation in the treatment episodes where patients did not have risk factors. Second, although we defined the high-risk subgroups based on the LASSO selection method, it is uncertain whether these criteria can be generalized in other populations. Although all PCP cases occurred in the high-risk subgroup, the predictive performance should be validated in other populations. Third, because this study was retrospective, patients in the prophylaxis groups were more likely to have clinical factors that increase the risk for PCP (steroid pulse, interstitial lung disease, and numerically high cumulative steroid use before the baseline), which could lead to a biased result. However, this ‘confounding by indication’ biased away the result from showing a protective effect so true prophylaxis effect might be even better than estimated. In addition, change in the immunosuppressant use during the observation could be also different between the two groups. Although we showed that cumulative dose of steroid treatment between the two groups was comparable, difference in other immunosuppressive agent uses was not considered in the analysis. Finally, unmeasured confounders such as patients’ compliance cannot be completely adjusted by statistical manipulation.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.