Pneumonia due to infection is the leading cause of hospitalization in the US, excluding childbirth and psychosis (1). It is also one of the major causes of mortality in patients with rheumatoid arthritis (RA) (2). However, most of the interest in the pulmonary diseases associated with RA has been directed toward uncommon entities, including interstitial lung disease secondary to RA (3) and adverse pulmonary effects of specific treatments. Prior reports have putatively implicated injectable gold (4), penicillamine (5), sulfasalazine (6, 7), methotrexate (8–10), infliximab (11), and leflunomide (12, 13). Thus, while bacterial and viral causes of pneumonia predominate, they are rarely reported in research studies.
The use of immunomodulatory drugs has sparked an interest in the infections that might result from treatment interventions. Methotrexate is suspected of conferring susceptibility to infectious pathogens (14–20), although most reports concern methotrexate pneumonitis, a hypersensitivity reaction (21). Low-dose prednisone is another commonly used treatment for RA, but there are no reported studies determining whether it is a risk factor for pneumonia in RA. Anti–tumor necrosis factor (anti-TNF) therapy has raised concern about a general risk of infection (22, 23), but there is little evidence from clinical trials suggesting a real or substantial risk (24, 25).
Surprisingly, there have been no previous published studies of RA and pneumonia. We therefore undertook this study to 1) determine the rate of pneumonia in RA, 2) determine whether RA treatments increase the risk of hospitalization for infection-associated pneumonia, and 3) estimate the degree of risk associated with each treatment.
- Top of page
- PATIENTS AND METHODS
At the time of the last assessment, prednisone and methotrexate were the most common treatments in this cohort during the previous 6 months (38.1% and 54.5%, respectively), followed by infliximab (36.9%), hydroxychloroquine (17.7%), and etanercept (12.8%) (Table 1). Among prednisone users, 66.9% received ≤5 mg/day, 23.4% received >5–10 mg/day, and 9.8% received >10 mg/day. The mean and median daily dosage of prednisone in the study cohort were 7.4 mg and 5.0 mg. On average, patients had been exposed to 3.3 disease-modifying antirheumatic drugs (DMARDs) or biologic agents, and the HAQ score was 1.1. Preexisting (pre-pneumonia) pulmonary disease had occurred in 17.0% of subjects. Diabetes was noted in 10.1%, and myocardial infarction in 8.2%.
Table 1. Characteristics of the 16,788 RA patients*
| Age, mean ± SD years||62.0 ± 13.3|
| Male sex||22.8|
| Ethnic origin|| |
| White, not of Hispanic origin||89.7|
| Black, not of Hispanic origin||4.8|
| Asian or Pacific Islander||1.0|
| American Indian or Alaskan native||1.1|
| Education, years|| |
|RA characteristics|| |
| Disease duration, mean ± SD years||16.3 ± 11.3|
| Lifetime no. of DMARDs or biologic agents, mean ± SD||3.3 ± 2.1|
| HAQ (0–3), mean ± SD||1.1 ± 0.7|
| Semiannual direct medical costs, median dollars||7,024|
| Prednisone, all dosages||38.1|
| Daily dosage among prednisone users, mg|| |
| Smoking (ever)||54.2|
| Pulmonary disease (ever)||17.0|
| Myocardial infarction (ever)||8.2|
| Comorbidity index (0–11), mean ± SD||2.5 ± 2.0|
| Pneumonia-associated hospitalization||3.7|
| Pneumonia-associated death||0.4|
There were 749 hospitalizations for pneumonia that occurred in 644 patients in the full study cohort. The incidence density of pneumonia was 17 per 1,000 patient-years (95% CI 16.4–19.1) for all patients, 19.2 per 1,000 patient-years for men (95% CI 16.3–22.5), and 17.3 per 1,000 patient-years for women (95% CI 15.8–18.9). For greater generalizability, we also conducted detailed incidence rate analyses of the 9,619 patients who were not part of the safety registries (Table 2). The incidence rate for pneumonia in this group was slightly lower than in the full cohort (14.7 per 1,000 patient-years [95% CI 13.1–16.4]). The rate was greater in men (17.9 per 1,000 patient-years [95% CI 14.2–22.1]) than in women (13.8 per 1,000 patient-years [95% CI 12.0–15.7]). The rate increased with age (Table 2). In the age group 75–84 years, it reached its maximum (21.0 per 1,000 patient-years). Only 3% of pneumonia hospitalizations were attributed directly to nonviral, nonbacterial opportunistic infections.
Table 2. Incidence rates of hospitalization for pneumonia among 9,619 nonregistry patients with rheumatoid arthritis
| ||Infections||Exposure, years||Incidence rate per 1,000 patient-years||95% confidence interval|
| Sex|| || || || |
| Age, years|| || || || |
A number of demographic and clinical variables were associated with the risk of pneumonia, as seen in the covariate-specific disease associations presented in Table 3. For example, each 10-year increase in age was associated with a 30% increase in pneumonia risk. Pneumonia was also more common in those with less education. Comorbidity status predicted future pneumonia among those who had ever smoked (hazard ratio [HR] 1.3 [95% CI 1.1–1.5]), were diabetic (HR 2.0 [95% CI 1.6–2.5]), had a past myocardial infarction (HR 2.1 [95% CI 1.7–2.6]), or had prior pulmonary disease (HR 3.8 [95% CI 3.2–4.4]). Certain RA features were also associated with the risk of pneumonia. Risk was increased with increasing duration of RA (HR 1.1 per 10 years increase in RA duration [95% CI 1.0–1.2]) and with each additional prior DMARD or biologic agent (HR 1.1 [95% CI 1.1–1.2]). Among the most powerful predictors of hospitalization for pneumonia was functional status: a 1-unit increase in HAQ score had an HR of 2.0 (95% CI 1.8–2.2).
Table 3. Univariable nontreatment predictors of pneumonia hospitalization*
|Variable||Hazard ratio||P||95% CI|
|Demographics|| || || |
| Age (per 10 years)||1.3||<0.001||1.3–1.4|
| Sex (male)||1.1||0.253||0.9–1.3|
| Education level (years)||0.9||0.001||0.9–1.0|
| Non-Hispanic white†||1.1||0.357||0.9–1.5|
|Comorbidity|| || || |
| Smoking (ever)||1.3||0.001||1.1–1.5|
| Pulmonary disease (ever)||3.8||<0.001||3.2–4.4|
| Myocardial infarction (ever)||2.1||<0.001||1.7–2.6|
| Comorbidity score (0–11)||1.3||<0.001||1.2–1.3|
|RA characteristics|| || || |
| RA duration (per 10 years)||1.1||0.009||1.0–1.2|
| No. of previous DMARDs or biologic agents||1.1||<0.001||1.1–1.2|
| HAQ (0–3)||2.0||<0.001||1.8–2.2|
The effect of treatment variables on the risk of pneumonia varied according to the covariates in the model. Table 4 presents univariable associations unadjusted for covariates and associations adjusted for covariates. The addition of covariates mitigated the risk associated with treatment variables. In the adjusted analyses, use of prednisone increased the risk of pneumonia by 70% (HR 1.7 [95% CI 1.5–2.1]) and leflunomide increased the risk by 30% (HR 1.3 [95% CI 1.0–1.5], P = 0.036), while persons receiving sulfasalazine had a reduced HR (0.7 [95% CI 0.4–1.0], P = 0.053). Etanercept use in this model was also associated with a marginally reduced HR (0.8 [95% CI 0.6–1.0], P = 0.051). In addition to a simple association of prednisone use with pneumonia, we also found a dose-related increase. Even dosages of ≤5mg/day were associated with pneumonia risk, in both the unadjusted model (HR 1.7 [95% CI 1.4–2.1]) and the adjusted model (HR 1.4 [95% CI 1.1–1.6]).
Table 4. Univariable treatment predictors of pneumonia hospitalization*
|Hazard ratio||P||95% CI||Hazard ratio||P||95% CI|
|Prednisone, all dosages||2.3||<0.001||1.9–2.7||1.7||<0.001||1.5–2.1|
|No prednisone||1.0|| || ||1.0|| || |
|Prednisone ≤5 mg/day||1.7||<0.001||1.4–2.1||1.4||<0.001||1.1–1.6|
|Prednisone >5–10 mg/day||2.9||<0.001||2.3–2.7||2.1||<0.001||1.7–2.7|
|Prednisone >10 mg/day||3.1||<0.001||2.2–4.3||2.3||<0.001||1.6–3.2|
Because drugs are often used concurrently, Table 5 recapitulates the adjusted analyses shown in Table 4, but includes all of the treatment variables simultaneously. In this model, treatment effects were weakened for all drugs except prednisone (HR 1.7 [95% CI 1.5–2.0]). Of interest, most demographic, comorbidity, and RA-related variables remained significant in this multivariable model.
Table 5. Multivariable predictors of pneumonia hospitalization*
|Variable||Hazard ratio||P||95% CI|
|Treatment|| || || |
|Demographics†|| || || |
| Age (per 10 years)||1.3||<0.001||1.2–1.4|
| Sex (male = 1)||1.1||0.425||0.9–1.1|
| Smoking (ever)||1.1||0.161||1.0–1.3|
|Comorbidity|| || || |
| Pulmonary disease (ever)||2.9||<0.001||2.3–3.4|
| Myocardial infarction (ever)||1.4||0.092||1.1–1.8|
|RA characteristics|| || || |
| HAQ (0–3)||1.5||<0.001||1.3–1.7|
| No. of previous DMARDs or biologic agents||1.1||0.020||1.0–1.1|
| Duration of RA (years)||1.0||0.687||1.0–1.0|
- Top of page
- PATIENTS AND METHODS
Despite the fact that infectious pneumonia ranks among the most common causes of death in RA, it has not yet been studied specifically in RA patients (2). Instead, most studies address noninfectious pulmonary complications of RA (such as interstitial lung disease), drug-related complications (such as hypersensitivity pneumonitis), or opportunistic pulmonary infections caused by unusual organisms. In our analyses, we confirmed that these types of infections were indeed rare, contributing only 3% of cases.
Many of the pneumonia risk factors described in our model are consistent with findings of previous investigations conducted in non-RA populations and may reflect underlying biologic phenomena. Among these factors is preexisting pulmonary disease, which has been shown in more than 20 studies to greatly increase the risk of pneumonia (38–41). The pathophysiologic explanation for this finding is, however, incompletely understood. A number of studies have also established smoking history as a risk factor for pneumonia (42–45).The relevant mechanisms that have been proposed include decreased ciliary and respiratory epithelial function, as well as defects in cellular and humoral immunity (44, 46). Interestingly, smoking entered the univariate model (Table 3), but was not predictive by multivariate analysis (Table 5) because of its correlation with preexisting pulmonary disease (which it may have caused).
The proinflammatory cytokine TNFα mediates the early response of mononuclear phagocytes to bacterial infections and may play an important role in lung disease. In particular, in endotoxin-reliant animal models of pneumonia, TNFα production has been shown to be stimulated, which in turn contributes to inflammatory cell recruitment (47). Furthermore, bronchoalveolar fluid TNFα levels may be higher in infected pulmonary lobes compared with the uninvolved lobes of patients with community-acquired pneumonia (48) (or may show a trend toward increased levels ). But despite the theoretical role of TNFα in pneumonia, our study failed to demonstrate an increase in risk associated with any of the anti-TNFα therapies (Table 5).
Steroid exposure has rarely been addressed as a potential risk factor for pneumonia in the general population (38–41, 44). To our knowledge, glucocorticoid dosage has not been demonstrated to be predictive of pneumonia risk in any previous study. However, we found a dose-related association between prednisone and pneumonia hospitalization in patients with RA (Tables 4 and 5). This relationship was evident even with average daily dosages of ≤5mg, and the association was robust to covariate control.
The immunomodulatory effect of prednisone is facilitated through both genomic and nongenomic pathways. These actions have been recently summarized and involve an extensive cascade of transcriptional/translational events, along with more rapid glucocorticoid receptor–dependent and –independent means (50). Given the broad spectrum of cellular mechanisms provoked by glucocorticoids (in relative contrast to the relatively specific actions of other DMARDs), it is perhaps not surprising that prednisone exhibited the strongest association with subsequent pneumonia.
Prednisone use is common in RA and is therefore a potentially important health risk. In our study cohort, a substantial minority (38.1%) of patients were receiving prednisone. This prevalence would be reduced to 30.9% if the infliximab and leflunomide safety registry patients were excluded. Likewise, the removal of subjects from these 2 registries would reduce the prevalence of anti-TNF exposure to 32.3%.
If the results of this study are correct, they may undermine the current belief that low-dose prednisone is safe. Given the prevalence of prednisone use, the findings of this investigation suggest a potentially important public health problem. Our data do not, and cannot, address the issue of net benefit. It is possible that discontinuing prednisone or not using prednisone in the first place might provoke equally undesirable adverse effects.
It is noteworthy that we found no evidence of increased rates of pneumonia associated with methotrexate (Tables 4 and 5), which challenges the perception from earlier reports (14, 15). We found a very slight increased risk with leflunomide (HR 1.2 [95% CI 1.0–1.5]) (Table 5). This compound, an inhibitor of de novo pyrimidine synthesis, retards the enzymatic activity of dihydroorotate dehydrogenase, thereby blocking T cell expansion. In addition, other pathways have recently been described, including inhibition of TNF-dependent NF-κB activation of T cells (51, 52). The relevance of these actions on the development of pneumonia has yet to be articulated. Finally, our results rule out the notion of an increased risk of pneumonia associated with sulfasalazine (HR 0.7 [95% CI 0.5–1.0]). This marginally significant effect may or may not relate to sulfasalazine's known mechanisms of action (53, 54).
Although observational studies reflect actual clinical practice in the community, nonrandom assignment to therapy can confound the association between treatment and pneumonia unless all important covariates are controlled for. In the current study, we adjusted for differences in severity by the use of lagged covariates (Tables 4 and 5). The HAQ is usually thought of as the best predictor of hospitalization and long-term outcomes (28, 55), and we included this variable in the model. We also included a count of the lifetime number of DMARDs or biologic agents used by patient, since the number of drugs is a measure of the lack of control of RA; furthermore, we included the duration of RA. HAQ disability and number of therapeutic agents were both found to be predictive of pneumonia risk. It is interesting to speculate whether known defects in immunity that parallel RA disease severity (such as mannose-binding lectin concentration) (56) might mediate this elevated risk.
Although not included in the models whose results are shown in Tables 4 and 5, we examined a series of other covariates, including the RADAI, pain scores, and global severity. In the presence of the study covariates, these factors were not statistically significant and did not change the results of the analyses; therefore they were excluded from the analyses presented in Tables 4 and 5. In addition to direct estimates of arthritis severity, the inclusion of biologic agents in the analysis in Table 5 serves as a further adjustment for arthritis severity, since anti-TNF therapy is prescribed to patients whose arthritis is more severe. As demonstrated in Tables 3 and 5, demographic factors and the presence of pulmonary disease, cardiovascular disease, and diabetes also contribute to the risk of pneumonia, and we controlled for these factors. Even so, we may not have accounted for all covariate effects and we suggest that, ideally, a randomized controlled trial should be undertaken to confirm the findings of this study, particularly given the public health implications of our findings.
In summary, our findings demonstrate a dose-related relationship between prednisone treatment and the risk of pneumonia in RA patients in the community. No increased risk was found with anti-TNF therapy or methotrexate. A slight increase in risk was found with leflunomide. Diabetes, prior myocardial infarction, and prior pulmonary disease also increase the risk of pneumonia. Because corticosteroid use is common in RA, the results of this study suggest that prednisone use may have important public health consequences.