Our data suggest that ICS use is associated with an increased risk of developing AI in a nationwide population-based study. Although we observed a moderate dose-dependent relationship between the ICS dosage and AI risk, this association was particularly pronounced above a certain threshold. Furthermore, this association remained consistent across various subgroups, including age, smoking status, income level, and underlying airway diseases. However, it exhibited a more significant association in males, individuals with systemic steroid use, and those with high CCI scores.
We found that ICS use increased the risk of AI in patients with chronic airway disease, consistent with previous findings [10–12]. Lapi et al. revealed that while the overall rate of AI was not significantly higher in current ICS users, those receiving the highest dosages had a greater risk (odds ratio [OR]: 1.84, 95% CI: 1.16–2.90) among patients treated with ICS for respiratory conditions in a Canadian nested case-control study conducted from 1990 to 2005 (cases = 397, controls = 3,920) [11]. In another UK case-control study (cases = 154, controls = 870), AI was linked to a prescription for ICS within 90 days of diagnosis (OR: 3.4, 95% CI: 1.9–5.9), but there was no association with a prescription for ICS before this time point [10]. A meta-analysis on steroid use and AI risk revealed that while ICS had a lower absolute risk (7.8, 95% CI: 4.2–13.9) than oral (48.7, 95% CI: 36.9–60.6) or intraarticular forms (52.2, 95% CI: 40.5–63.6), they were still associated with an increased risk of AI [12]. However, conflicting findings have been reported [14–16]. In a 52-week randomised controlled trial involving 187 children aged 4–9 years with asthma, ICS did not significantly affect serum or urinary cortisol levels [15]. Lipworth et al. also found that ciclesonide did not affect adrenal function markers, while the fluticasone group showed significantly reduced cortisol levels in response to cosyntropin stimulation and 24-hour urinary free cortisol levels in a 12-week study of adult asthma (n = 248) [14]. In a meta-analysis of asthma patients, beclomethasone had the most significant dose-dependent urinary cortisol suppression (8.4% per 100 µg; p = 0.029), followed by fluticasone (3.2% per 100 µg; p < 0.001) and budesonide (3.1% per 100 µg; p = 0.001). In contrast, ciclesonide did not significantly suppress urinary cortisol (1.8% per 100 µg; p = 0.267) due to its unique pharmacokinetics [16]. Our data suggest that regardless of the ingredient, all components of the ICS appeared to increase the risk of AI. Collectively, the use of ICS, regardless of their specific ingredients, may be associated with a slightly varying increased risk of AI.
The mechanism of AI with ICS involves minimal systemic absorption, conversion to systemic steroids, and the HPA axis through negative feedback [28]. Prolonged and high-dose usage can lead to HPA deficiency, adrenal gland atrophy, and persistent adrenal suppression even after treatment cessation, lasting up to several years [29, 30]. Kachroo et al. found that ICS use in asthma patients (n = 661) significantly reduced levels of key steroid metabolites, including dehydroepiandrosterone sulphate and cortisol, suggesting a potential association between ICS use and AI, with a global reduction in cortisol levels observed over a 24-hour period [31]. The clinical presentation of AI with non-specific symptoms, including fatigue, headaches, weakness, poor growth, syncope, hypotension, hyponatraemia, and hypoglycaemia, can complicate the diagnosis. In addition, several paediatric asthma clinical guidelines recommend conducting AI screening tests when using ICS for six months or longer or at medium to high doses [32, 33]. While the potential for AI associated with inhaled ICS use in adults has not received significant attention, our study indicates that ICS use increases the risk of AI, even in adult patients. Notably, ICS remains the first-line therapy for adult patients with asthma [34] and are recommended for use in patients with COPD accompanied by type 2 inflammation [35, 36]. Goldbloom et al. demonstrated that among 46 paediatric patients with symptomatic AI, 80% were treated with ICS for asthma, with 32% of them receiving ICS therapy alone [37]. Furthermore, in our study, the intermittent use of ICS appeared to increase the risk of AI to a similar extent as that in continuous use. A study by Schütz et al. found that a 14-day regimen of systemic corticosteroids significantly suppressed the HPA axis in 8 out of 9 COPD patients not previously on systemic steroids, with effects lingering in 3 patients three weeks post-therapy [38]. This highlights the importance of monitoring for AI in adults using ICS or systemic steroids, even with intermittent use, not just prolonged continuous use.
The association between ICS use and AI was more pronounced in individuals with systemic steroid use, and those with a high CCI, consistent with previous findings [39]. However, no clear dose-dependent increase in AI risk was seen at higher ICS doses. However, continuous dose analysis showed a significant rise in AI occurrence with increasing ICS doses, particularly at medium doses or with systemic steroid use. Despite the low overall incidence of AI (1.69 per 1000 ICS users, 0.54 per 1000 overall), the statistical significance in categorical analysis may have been obscured due to low-frequency high doses. Our study also found a significantly lower AI risk among those not using systemic steroids, highlighting the need for ongoing monitoring of AI in ICS users, especially those on oral systemic steroids.
Our study had several limitations. First, as it was a retrospective analysis of claims databases, it is susceptible to inherent bias, including selection bias, information bias, or confounding factors. To address this, we employed a propensity score analysis with IPTW to mitigate selection bias and control for potential confounders. IPTW minimizes group differences while preserving our full sample size, a key advantage over methods like propensity matching that restrict data to matched cases [41], thereby enhancing the strength and credibility of our results. Additionally, our study, conducted with a randomly sampled population from the Korean NHI, which covers the entire nation, likely has a lower possibility of selection bias. Second, owing to the nature of our claims data, we were unable to obtain information regarding the adherence to and the duration of ICS use. Therefore, we conducted multiple subgroup analyses that yielded consistent results. Third, our findings may be specific to the Korean population, which limits their generalisability. Forth, we used diagnostic codes from claims databases, which may introduce systematic biases despite careful definitions of diseases and outcomes. To enhance validity, further large-scale studies using more accurate medical records are recommended. Lastly, the low incidence of AI, especially in smaller groups where HR exceeded 1 without statistical significance, may be due to limited statistical power from the small sample size. This necessitates cautious interpretation and indicates a need for further research with larger datasets. Despite these limitations, our comprehensive approach using a large claims database provides valuable insights into the association between ICS use and AI risk.