Introduction

Inhaled corticosteroids (ICS) are the cornerstone of asthma pharmacotherapy and a common first-line treatment.1 Though considered generally safe, clinical studies suggest that ICS increase the risk of pneumonia in adults with COPD.2 The safety evidence for children with asthma is limited.

ICS could increase the risk of pneumonia by immunosuppression and studies have shown increased carriage of Streptococcus pneumoniae in asthma patients using ICS in comparison with no ICS use.3 However, meta-analyses of randomised controlled trials (RCTs) show evidence of a protective or null effect of ICS on the risk of pneumonia in asthma patients.4–6 A meta-analysis of RCTs on children (31 trials; 11 615 patients; age<18 years) found no difference in the risk of pneumonia between the ICS and placebo group.4 A meta-analysis of 10 RCTs of mostly adults (19 098 patients; age>12 years) found that ICS use decreased the risk of pneumonia.5 Although patients in the RCTs were 12 years or older, no children-specific results were presented.5 Another meta-analysis (26 RCTs; 14 993 patients; age>4 years) also found that ICS use decreased the risk of pneumonia events.6 In contrast, two case–control studies in adults and patients aged 12–35 years with asthma, respectively, found that ICS use was associated with an increased risk of pneumonia.7 8 No paediatric-specific results based on real-world data have been presented to our knowledge.

Asthma is a chronic disease that requires long-term pharmacotherapy to prevent harmful inflammation, loss of lung function and long-term adverse effects. Most asthma patients develop symptoms and initiate ICS during childhood when the lungs and the immune system are still developing and the risk for respiratory infections is high.9 It is hence key to investigate potential adverse effects of ICS in children, such as the risk for pneumonia, and evidence from large real-world samples is needed. In this study, we assessed the potential association between ICS use and the risk of hospitalisation for pneumonia in Swedish paediatric asthma patients (age 2–17 years), as compared with no ICS use.

MethodsSource population and data sources

This was a Swedish nationwide cohort study using linked data from multiple Swedish national registers. The data sources are further described in online supplemental appendix etable 1. The study period was from 1 January 2007 to 30 November 2021. The source population consisted of children aged 2–17 years residing in Sweden and with a confirmed asthma diagnosis before or during the study period. Patients under the age of 2 years were not included in the primary analysis due to the difficulties in ascertaining both asthma and pneumonia in these patients.10 11 Confirmed asthma was defined based on a previously validated algorithm (online supplemental appendix etable 2) using physician-assigned diagnosis records (international classification of diseases, 10th revision (ICD-10) codes) from the Swedish National Patient Register and records of asthma drug dispensing (anatomical therapeutic chemical (ATC) codes) from the Swedish Prescribed Drug Register.12

Procedures

The Swedish Prescribed Drug Register that contains details on dispensed prescription medication for all individuals residing in Sweden was used to identify exposure. From the children with confirmed asthma, mutually exclusive episodes of ICS use and no use during the study period were identified (see an overview of the design in online supplemental appendix efigure 1).13 The index date for an ICS episode was the day of a new dispensing of an ICS drug. Potential index dates for no-use episodes were set with 12-month intervals during the study period. Eligible episodes, both ICS and no-use episodes, were preceded by an 18-month period of no ICS dispensing. No-use episodes were required to have at least one dispensing of another respiratory medication in the last 12 months to verify that patients during comparator episodes were currently or recently treated. Thus, one asthma patient could contribute with multiple ICS and no-use episodes during the study period. Episodes did not overlap, and each outcome event was only accounted for once.

The following exclusion criteria were applied to all episodes based on their status at the index date: previous hospitalisation within the last 30 days, previous pneumonia within the last 60 days, lived outside Sweden within the last 5 years, history of cancer, immune suppression including HIV, severe chronic lung disease or other life-limiting conditions.14 The exclusion criteria are further described in online supplemental appendix etable 2.

Outcome

The primary outcome was hospitalisation for pneumonia, defined as an inpatient contact with a physician-assigned primary diagnosis of pneumonia (ICD-10 codes J12-18), based on data from the Swedish National Patient Register. The date of admission was used as the event date. The secondary outcome was hospitalisation or emergency outpatient contact for pneumonia, which was defined as the primary outcome but extended to also include pneumonia in the emergency outpatient setting.

Statistical analyses

We used propensity score overlap weighting to control for confounding.15 16 With this method, we accounted for confounding by balancing the distribution of covariates at baseline between the treatment groups (ICS use and no-use episodes).17 With overlap weights, we estimated the average treatment effect in the overlap population. The propensity scores were estimated with logistic regression. The potential confounders (baseline covariates) included in the propensity score were age, sex, season, socioeconomic status of parents (disposable income, education level), history of lower respiratory tract infections (LRTI), trisomy 21 and congenital heart disease, previous use of asthma medication, and asthma-related and general healthcare utilisation (online supplemental appendix etable 3). Potential confounders were defined based on status at the index date of each episode and balance was assessed with standardised mean differences. Missing values (less than 0.1%) were observed for parental income and education. The missing values were imputed with median and mode, respectively. Episodes were censored at the first outcome event, maximum follow-up (12 months), end of study period (30 November 2021), age>17 years, initiation of an ICS drug (for no-use episodes), death or emigration.

Cox proportional hazards regression was used to estimate the association between ICS use and the risk of the outcome, as compared with no use. The rates of outcome events during episodes of ICS use and no use were estimated. Cumulative incidence was estimated with the Kaplan-Meier function. Poisson regression was used to obtain absolute rate differences; events per 10 000 patient-years. The time scale was time since the index date. Robust sandwich estimators were used to estimate standard errors that account for the weighted nature of the cohort.

Subgroup analyses were performed based on sex, age group (2–5, 6–11, 12–17 years) and type of ICS initiated (ICS as monotherapy, combination therapy of ICS and long-acting beta-agonists (LABA), any budesonide and any fluticasone). These subgroup analyses were conducted to explore potential effect modification, as the association between ICS use and the risk of pneumonia may vary between demographic groups and types of ICS therapy. A few sensitivity analyses were performed. First, children younger than 2 years were not included in the primary analysis, but in a sensitivity analysis all children of age 0–17 years were included. Second, we adjusted for additional potential confounders that required full register coverage since birth, which was only available for a subset of patients born in Sweden in 2006 or later. We adjusted for prematurity, small for gestational age, parity, LRTI before and after age 2 years in inpatient and outpatient care, respectively, and sibling with confirmed asthma (online supplemental appendix etable 4). The literature suggests that LRTI early in life may be associated with an increased risk for developing asthma and recurrent LRTIs.18–20 Third, we used a less restrictive definition of the outcome, which included inpatient contacts with pneumonia as secondary diagnosis. Fourth, we performed an analysis where we excluded the years of the COVID-19 pandemic (2020–2021) from the study period, when the background incidence of hospitalisation for pneumonia in children might have been different. All analyses were performed in SAS Enterprise Guide V.7.1

Study population

We identified 425 965 children with confirmed asthma during the study period. From these patients, we identified 249 351 ICS and 214 840 no-use episodes that met the inclusion criteria (figure 1). The mean follow-up time was 0.9 and 0.7 years in the ICS and no-use episodes, respectively (the median was 1.0 year in both groups). In the unweighted cohort, the most imbalanced covariates between ICS and no-use episodes included age, previous use of leucotriene receptor antagonists, number of respiratory drugs used and number of previous outpatient hospital contact for asthma (table 1). In the weighted cohort, nearly exact balance between ICS and no-use episodes was achieved for all covariates. In the weighted cohort, the mean age at baseline was 10.5 years for both ICS use and no-use episodes, with females accounting for 46% of ICS use and no-use episodes.

Figure 1

Selection flow of ICS and no-use episodes. Exclusion criteria are not mutually exclusive. *Episodes of patients with confirmed asthma at age 2–17 years, with recent respiratory pharmacotherapy (in last 12 months, including the index day) and with no recent ICS use (in last 18 months). ICS, inhaled corticosteroids.

Table 1

Baseline characteristics of episodes included in the unweighted and weighted cohorts

Results

Primary analysis

In the ICS and no-use episodes, 369 and 181 incident hospitalisations for pneumonia were observed during follow-up, respectively (table 2). This translated to incidence rates of 15.8 and 11.5 events per 10 000 patient-years, respectively, and the unweighted HR associated with ICS use was 1.44 (95% CI 1.21 to 1.72; table 2). In the weighted cohort, the incidence rates were similar between the groups, 14.5 events per 10 000 patient-years (cumulative incidence 0.14%; figure 2) in the ICS episodes and 14.6 (cumulative incidence 0.13%) in the no-use episodes. We found no material difference between the groups, in relative or absolute terms. The weighted HR of hospitalisation for pneumonia associated with ICS use was 1.06 (95% CI 0.88 to 1.28; table 2 and figure 3), in comparison with no use. The absolute rate difference was −0.06 (95% CI −2.83 to 2.72) events per 10 000 patient-years.

Figure 2

Cumulative incidence of hospitalisation for pneumonia in unweighted and weighted cohorts of ICS and no-use episodes. ICS, inhaled corticosteroids.

Figure 3

Subgroup and sensitivity analyses of association between ICS use and the risk of hospitalisation for pneumonia in comparison with no use. Incidence rates and rate differences are presented as events per 10 000 person years. Individual propensity score models and related weights were estimated for each sensitivity and subgroup analysis. aP values for test of differences between subgroups were 0.95 and 0.35 for sex and age, respectively. bEnd of study period was set before the COVID-19 pandemic. cSubset 2006 consisted of patients born in 2006 or later and with available data on additional potential confounders (prematurity, small for gestational age, parity, sibling with confirmed asthma and lower respiratory tract infections before the age of 2 years); the analysis was performed with the same adjustment as in the primary analysis and with the additional variables. ICS, inhaled corticosteroids; LABA, long-acting beta-agonists.

Table 2

Association between ICS use and the risk of hospitalisation for pneumonia (primary analysis) and hospitalisation or emergency outpatient contact for pneumonia (secondary analysis) in comparison with no use

Secondary and subgroup analyses

In the secondary analysis where both hospitalisations and emergency outpatient visits for pneumonia were considered, 1104 and 504 events in the ICS and no-use episodes were observed, respectively. The weighted HR associated with ICS use was higher than the primary analysis, 1.10 (95% CI 0.98 to 1.23; table 2). The weighted HRs by age groups were 0.88 (95% CI 0.68 to 1.13) among children of age 2–5 years, 0.99 (95% CI 0.68 to 1.45) for age 6–11 years and 1.28 (95% CI 0.82 to 2.02) for age 12–17 years (figure 3). The results were similar between boys and girls, HRs of 1.12 (95% CI 0.85 to 1.46) and 1.06 (95% CI 0.82 to 1.38), respectively.

We performed analyses stratified on the type of ICS that was initiated. In the analysis of ICS as monotherapy (222 872 episodes), the HR of hospitalisation for pneumonia was 1.04 (95% CI 0.86 to 1.26) and it was higher in the analysis of ICS and LABA as combination therapy (26 692 episodes), HR 1.18 (95% CI 0.68 to 2.07). The results were similar in the analyses of ICS types, budesonide (167 109 episodes) and fluticasone (80 699 episodes); we observed HRs of 1.03 (95% CI 0.81 to 1.30) and 1.05 (95% CI 0.83 to 1.34), respectively (figure 3).

Sensitivity analyses

In the sensitivity analysis where children under the age of 2 years were also included, higher weighted incidence rates of hospitalisation for pneumonia were observed; 36.6 events per 10 000 patient-years in the ICS episodes and 32.6 in the no-use episodes. The HR associated with ICS use was higher in this cohort than in the primary analysis, 1.23 (95% CI 1.09 to 1.38). In the analysis where the outcome definition also included secondary diagnoses, we observed a similar result as in the primary analysis, HR 1.03 (95% CI 0.87 to 1.23). The sensitivity analysis where we adjusted for additional potential confounders was restricted to patients with available data since birth (44% of the ICS episodes; max age 16 years). The unweighted cohort was fairly balanced on the additional potential confounders (online supplemental appendix etable 5) and the HRs with the same adjustment as in the primary analysis and with the additional adjustment were the same, 0.99 (95% CI 0.77 to 1.26) (figure 3). Finally, we found no difference in relation to the primary analysis when the study period ended before the COVID-19 pandemic, HR 1.01 (95% CI 0.83 to 1.22).

Discussion

In this Swedish nationwide cohort study, we assessed the potential association between ICS use and the risk of hospitalisation for pneumonia in children with asthma. The study was based on a sample of more than 425 965 patients from routine clinical practice during a study period of 15 years. Our results indicated no evidence of an association between ICS use and the risk of hospitalisation for pneumonia, as compared with no use. Based on the upper limit of the CI, our results were not compatible with an increased rate greater than 2.7 events per 10 000 patient-years.

The results obtained from this study of patients in routine clinical practice are in agreement with those of previous meta-analyses of RCTs providing children-specific results, which show a null or protective effect of ICS use on the risk of pneumonia.4 6 Cazeiro et al found no difference in risk of pneumonia between ICS and placebo groups (risk difference –0.1%; 95% CI –0.3% to 0.2%) among patients younger than 18 years.4 O’Byrne et al found a protective effect of ICS on the risk of pneumonia events in the age group 4–11 years, HR 0.54 (95% CI 0.33 to 0.90).6 For serious pneumonia events, the results were imprecise, and a null effect was observed, HR 0.93 (95% CI 0.31 to 2.78). The corresponding result was similar for pneumonia events in the analysis that included adults (age 4 to 78 years) HR 0.51 (95% CI 0.35 to 0.74); while it was higher but still imprecise for serious pneumonia events, HR 1.30 (95% CI 0.53 to 3.14).

Our results differed from those of previous observational studies, which showed an association between ICS use and increased risk for pneumonia: OR of 1.20 (95% CI 1.12 to 1.27) in adults7 and a rate ratio of 1.83 (95% CI 1.57 to 2.14) in patients aged 12–35 years.8 The observed differences between results could be due to important differences in the design (both previous studies used case–control designs), definition of the population (less strict definitions of asthma and no required use of respiratory medication in the no-use group in the previous studies), and in the confounding adjustment. Further, the previous studies analysed only or mostly adult populations. The impact of some design choices can be observed in the study by Qian et al.8 For example, the results of the primary analysis, a rate ratio of 1.83 (95% CI 1.57 to 2.14), was attenuated when the non-ICS users were restricted to current users of other respiratory medication, 1.61 (95% CI 1.38 to 1.89), and increased when the non-ICS users instead did not use any other respiratory medication, 4.81 (95% CI 3.56 to 6.51).

Asthma is a heterogenous disease and manifestation, and severity varies greatly between age groups.21–23 Children in general and those under the age of 5 years in particular are prone to developing symptoms that may be misclassified as asthma when infected with respiratory viruses.10 11 In our analysis, children aged 2–5 years had the highest unweighted rates of hospitalisation for pneumonia, especially ICS users. Our sensitivity analysis that included patients under the age of 2 years showed an increased risk of hospitalisation for pneumonia associated with ICS use. However, this increased risk should be interpreted with caution given the difficulties in ascertaining both asthma and pneumonia in the youngest children.9–11 Further, we observed a higher but less precise HR in the age group 12–17 years. A similar imprecise (only 18 events in the ICS episodes) result was observed for the subgroup of patients initiating a combination therapy of ICS and LABA. The combination therapy is indicated for children older than 6 years and generally targets patients with more severe disease.24 Given the no-use design, this analysis might be subject to more extensive confounding by indication.

The study had limitations. As in all observational drug safety studies, the analysis was subject to potential unmeasured confounding and several measures were taken to limit this risk. In the study design, we included only patients with confirmed asthmatic disease based on a validated algorithm.12 The definition of asthma was comparably strict and to avoid including patients with very mild or transient disease we additionally required that patients received respiratory pharmacotherapy in the last 12 months. Due to the difficulties in distinguishing asthma from obstructive bronchitis or viral-induced wheeze in very young children, those under the age of 2 years were excluded from the analysis.

In the statistical analysis, we adjusted for a large number of potential confounders, including comorbidities, asthma drug history and proxies of disease severity. We used propensity score overlap weighting to achieve nearly exact balance on all covariates. With this method, we efficiently estimated the HR in the patients with the most propensity score overlap.15 Further, we performed a sensitivity analysis with adjustment for additional potential confounders, including LRTIs at age below and above 2 years, prematurity, parity, small for gestational age, and having a sibling with asthma, in patients with available data since birth. The additional adjustment did not change the results. Finally, the potential unmeasured confounding in the primary analysis was expected to bias the results away from the null, that is, unmeasured risk factors would be more prevalent in the ICS episodes than the no-use episodes. Hence, it is unlikely that the close-to null result of the study is explained by this bias. Nonetheless, the presence and impact of unmeasured confounding cannot be ruled out.

Outcome and exposure misclassification are other potential biases. Outcome events were ascertained based on physician-assigned diagnosis codes in the Swedish National Patient Register. Validation studies have shown that the ICD-10 codes for pneumonia have high positive predictive values (85%–95%).25 26 To increase specificity of the definition, we restricted our primary outcome to hospitalisations and required that pneumonia was the primary diagnosis, that is, the reason for admission. In a sensitivity analysis where secondary diagnosis codes were also included, the result did not change materially. A similar result was observed in the secondary analysis where we analysed the incidence of pneumonia in inpatient or emergency outpatient care. We did not analyse the risk of bacterial and viral pneumonia separately since the aetiology is difficult to distinguish in children due to inaccurate diagnostic methods and the specific diagnosis codes have low validity.

Our exposure definition was based on data of dispensed prescriptions of ICS as mono or combination therapy and low adherence would bias the results toward the null. Due to the great variability in dosing and general use patterns of ICS, we used a comparably short maximum follow-up of 12 months and to make sure that no patients used ICS recently we used a wash-out period of 18 months. It was not possible to investigate a potential association between the prescribed daily dose of ICS and the risk of pneumonia, since these data cannot be derived from the registers and the dose can vary over time within ICS dispensing. However, such an analysis would be of interest as higher ICS doses may be associated with greater immunosuppression.

In summary, based on nationwide data from routine clinical practice, we found no evidence of an association between ICS use and the risk of hospitalisation for pneumonia in children with asthma. The results were in accordance with results of previous RCTs on children, but not aligned with the previous observational data, which mainly included adults. To our knowledge, this is the first large study that provides paediatric-specific evidence on the potential association between ICS use and the risk of hospitalisation for pneumonia. Our findings further support the safety of ICS use for managing asthma in children.