Predictive model for inhaled corticosteroid response in hospitalized asthma: implications for precision medicine

Introduction

Asthma is a chronic inflammatory disorder of the airways characterized by marked heterogeneity, and it involves multiple cell types and inflammatory pathways, leading to diverse clinical phenotypes (1). Epidemiological data indicate that approximately 300 million individuals worldwide are affected by asthma each year, with a continuous increase in its prevalence (2). In western China—particularly in regions such as Xinjiang—asthma burden is even greater, likely promoted by environmental factors such as air pollution, wind-blown sand, and region-specific allergens (3). Consequently, asthma patients represent a substantial proportion of those individuals attending respiratory departments in Xinjiang hospitals (4), underscoring the urgent need for clinical decision-support tools that can assist physicians in optimizing patient management.

Over the past three decades, corticosteroids have played a critical role in treating asthma and have markedly reduced asthma-related mortality (5). However, the prevailing treatment paradigm presents certain limitations. Inhaled corticosteroids (ICS) are typically administered as the first-line therapy, with escalation to systemic corticosteroids (SCS) reserved for patients showing inadequate response to ICS, such as those with severe or uncontrolled asthma or those experiencing acute exacerbations (6). SCS use, however, carries a considerable risk for the development of comorbidities, including osteoporosis, hypertension, gastrointestinal bleeding, and even neuropsychiatric complications (7). Notably, Roger et al. suggests that even local administration of corticosteroids in the middle ear does not mitigate these systemic side effects, emphasizing the importance of minimizing unnecessary systemic exposure to corticosteroids (8). Therefore, precision medicine strategies that inform timely and appropriate treatment escalation are essential but remain underdeveloped.

To date, candidate biomarkers, including blood eosinophil count, fractional exhaled nitric oxide (FENO), total IgE, and type 2-related cytokines, have been proposed to guide treatment escalation from ICS to SCS (9). Simultaneously, the emergence of biological therapies targeting type 2 inflammation, such as anti-IL-5 and anti-IgE agents, has transformed the management of severe eosinophilic asthma (10). However, selecting appropriate candidates for such therapies remains a challenging issue. Recent insights highlight the need for precision medicine approaches that integrate clinical, inflammatory, and molecular phenotyping to guide both biologic use and corticosteroid escalation (11). A systematic review by Breiteneder et al. (2020) concluded that while these markers can facilitate the phenotyping of patients, their individual predictive value for treatment escalation is modest, particularly in non-type 2 asthma (12). Similarly, Kroes et al. (2020) emphasized that existing algorithms lack longitudinal biomarker trajectories and do not account for the heterogeneity of severe asthma phenotypes, resulting in sub-optimal discrimination (13). Earlier studies on prediction tools have been further constrained by a small sample size, short follow-up duration, or reliance on pulmonary function tests (PFTs) that are difficult to standardize in large, real-world populations (14). Although Ong et al. (2024) developed genomic machine learning models for determining ICS responsiveness, their study was restricted to European ancestry populations and did not address progression to SCS, thus limiting external validity to multiethnic, hospitalized cohorts (15). By leveraging the CHRONICLE study, Trevor et al. (2021) demonstrated that real-world, specialist-verified data can capture treatment escalation events with greater granularity than claim-based analyses; however, they also emphasized the absence of validated decision rules to distinguish necessary from excessive SCS use (16).

These gaps highlight several methodological challenges. First, although large, long-term cohort studies provide more robust evidence, ICS responsiveness assessment traditionally relies on PFTs—a gold standard with difficulty for uniform application in such settings (17). Additionally, PFTs may be contraindicated in patients with severe or acute exacerbations. Second, inconsistent outpatient medication adherence affects data integrity. Finally, randomized controlled trials (RCTs) to investigate SCS overuse raise ethical concerns, as randomizing patients to escalate therapy without clinical indication would be inappropriate.

To overcome these limitations, our study implemented the following approaches: (I) we defined ICS non-responsiveness as treatment escalation to SCS and limited the study population to hospitalized patients to ensure medication compliance; and (II) we utilized causal analysis methods within a real-world observational dataset to approximate RCT conditions and evaluate potential SCS overuse. Specifically, we initially developed a predictive model using a large, long-term cohort to identify patients likely to escalate from ICS to SCS and then applied this model to examine patterns indicative of SCS overuse. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2260/rc).

Methods

Study cohort

In this retrospective cohort study, patients diagnosed to have asthma at Xinjiang Uygur Autonomous Region People’s Hospital between January 1, 2006 and December 31, 2022 were consecutively recruited. All patients were diagnosed to have asthma on the basis of Global Initiative for Asthma (GINA) diagnostic criteria (18). The inclusion criteria were as follows: (I) age ≥12 years; and (II) ICS prescribed as the initial treatment/first-line medication. No exclusion criteria were used for this study. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics board of the Xinjiang Uygur Autonomous Region People’s Hospital (No. KJ2024-252-01). Informed consent was waived for this retrospective study. Although this study was a retrospective investigation, we performed a post-hoc power analysis to confirm that the final sample size was adequate for both risk factor comparison and subsequent predictive model development. By using an expected escalation rate of 20% for ICS-to-SCS escalation (based on our pilot data), a two-sided α-level of 0.05, and an absolute between-group difference of 10% (e.g., 15% vs. 25%) as the smallest clinically relevant effect, PASS 2021 software indicated that 1,050 patients per arm were required to achieve 80% power. The final analytical cohort comprised 5,463 patients (1,088 events), yielding an actual power of >90%. In addition, for the machine learning component, we adhered to the “events-per-variable” (EPV) rule; for 41 predictors and 1,088 events, the EPV was approximately 26.5, which exceeded the commonly used minimum threshold of 20.

Data collection and outcome

The demographic and clinical data potentially predictive of patients’ initial ICS treatment response were collected. The primary outcome of interest was defined as treatment escalation from the initial treatment (i.e., ICS) to SCS. The secondary outcomes of interest included total hospitalization cost, hospitalization length, and death or transfer to intensive care unit (ICU).

Data from patients admitted before and after January 1, 2019, were assigned to derivation and temporal validation sets, respectively. The derivation set was randomly divided into training and hold-out internal validation sets at the ratio of 7:3.

Model development

Variables with a missing rate of 0.25 or higher were excluded, along with any observations containing one or more missing values in the remaining predictors. Potential predictor variables were converted to numeric or binary formats (e.g., 1 indicates a history of coronary disease in patients, while 0 indicates the absence of such a history). The primary outcome variable was recoded to 0 (indicating a negative outcome) and 1 (indicating a positive outcome), which represented successful ICS treatment and treatment escalation from initial ICS to later SCS, respectively.

Machine learning algorithms were used to develop a prediction model. Multivariate logistic regression, least absolute shrinkage and selection operator (LASSO), random forest, and eXtreme Gradient Boosting were developed for the dataset. Subsequently, the optimal parameters of the machine learning algorithms were obtained through cross-validation in the training set. Finally, the performance of the models was evaluated using the hold-out internal validation set.

Model validation

First, we validated the performance metrics of the models in the temporal validation cohort. Model performance was evaluated using the receiver operating characteristic (ROC) curve. The area under the ROC curve (AUC) value was calculated. The accuracy of the optimal cutoff value was assessed using sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).

Second, we compared the outcomes of patients whose predicted probability was negative. The outcomes between patients with treatment escalation and those without treatment escalation were compared on a matched set. For patients with false-negative prediction (i.e., those who were predicted negative but actually required treatment escalation), we paired them with patients with true-negative prediction (i.e., those who were predicted negative and did not require treatment escalation) and those with close predicted probabilities. By using this approach, we simulated a prospective study wherein we could evaluate the effects of treatment escalation/no treatment escalation on the outcomes of patients while controlling for other variables.

Statistical analysis

Statistical analyses were performed using R software (version 4.4.1; R Foundation for Statistical Computing, Vienna, Austria). Continuous variables were compared using the independent samples t-test for normally distributed data and the Wilcoxon rank sum test for non-normally distributed data. Categorical variables were compared using the Chi-squared test or Fisher’s exact test, as appropriate. All statistical tests were two-sided, and a P value of <0.05 was considered statistically significant. Following the current reporting guidelines (TRIPOD and PROBAST), we pre-specified that the model would be considered clinically useful if it achieved an AUC of ≥0.70, a sensitivity of ≥60%, and a specificity of ≥60% in the hold-out (internal) validation set.

Results

Patient characteristics

A total of 10,530 eligible patients (6,166 females and 4,364 males) who met the inclusion criteria were initially recruited. After excluding patients with missing values ≥0.25 and left-out variables, 5,463 patients (3,225 females and 2,238 males) were ultimately included. All selected patients received an initial treatment with ICS, and 1,088 (19.9%) patients required treatment escalation during hospitalization. The escalated treatments consisted of SCS, which included a glucocorticoid saline infusion (regardless of dose and duration). The clinical variables of patients with and without treatment escalation are shown in Table 1 and Figure S1. Additionally, the clinical variables of the derivation and temporal validation cohorts are presented in Table S1 and Figure S2.

After excluding variables with a missing rate of ≥0.25 and observations with missing values for the left-out variables, 41 variables were finally obtained and selected for model development. The models were developed using multivariate logistic regression, LASSO, random forest, and eXtreme Gradient Boosting. The random forest model demonstrated superior performance compared to other models in the hold-out internal validation set, with an AUC value of 0.7483 [95% confidence interval (CI), 0.709–0.7876] (Figure 1A). The sensitivity, specificity, PPV, and NPV of the random forest model in identifying unresponsiveness to the initial ICS treatment were 68.50%, 66.23%, 34.51%, and 89.01%, respectively, in the hold-out internal validation cohort (Table 2).

Figure 1 ROC curve analysis. (A) ROC curves for evaluating the model’s discrimination ability in the internal hold-out validation set. (B) ROC curves for evaluating the model’s discrimination ability in the temporal validation set. FPR, false positive rate; ROC, receiver operating characteristic; TPR, true positive rate.

The random forest model was subsequently applied to make predictions in the temporal validation set. The model exhibited effective performance in predicting treatment escalation, with an AUC value of 0.6941 (95% CI, 0.6651–0.7231) in the temporal validation set (Figure 1B). The sensitivity, specificity, PPV, and NPV of the random forest model for identifying unresponsiveness to the initial ICS treatment were 63.36%, 61.53%, 27.80%, and 87.78%, respectively, in the temporal validation cohort (Table 2).

Outcome comparison

In the temporal validation set (N=2,232), 741 (33% of 2,232) patients had a predicted probability of <0.135 (the lower quartile of all predicted probabilities in the temporal validation cohort). Among these 741 patients, 71 (9.58%) patients actually received SCS treatment. However, their predicted probability was not significantly higher than that of the remaining 670 patients (P=0.05). For these 71 patients, we identified their matched patients (with lower predicted probability and those without treatment escalation) among the 670 patients. This resulted in a matched set of 62 pairs of patients. In each pair, the predicted probability of patients, which indicated the likelihood of treatment escalation and was determined by all predictors, was similar.

The secondary outcomes of interest were compared between each patient pair, including treatment escalation from the initial ICS treatment to SCS treatment, total hospitalization cost, hospitalization length, and death or ICU transfer (Table 3 and Figure 2). The analysis revealed that only the total hospitalization cost differed significantly between patient pairs. This finding suggests potential overuse of corticosteroids, as these patients (predicted by the model to not require treatment escalation) showed no improvement in major outcomes even after treatment escalation to SCS.

Figure 2 Boxplot showing the results of outcome comparison. (A) Comparison of total hospitalization cost. (B) Comparison of hospitalization length. 0, predicted negative patients without treatment escalation; 1, predicted negative patients with treatment escalation.

Discussion

Predicting the response to ICS treatment has high clinical relevance. Traditional prediction methods for ICS treatment response primarily relied on multivariable logistic regression models alone, and most studies were conducted in children with asthma, with limited predictive factors (19,20). In contrast, machine learning techniques can identify complex associations among multiple features through large-scale data analysis and enhance prediction model accuracy by addressing potential errors or false information (21). An AUC value of 0.7 indicates a moderate level of predictive accuracy. A comparison of our present study to similar investigations revealed no directly comparable existing models. While Chung et al. employed machine learning to predict severe asthma response to corticosteroids (22), their study focused solely on identifying variables related to corticosteroid treatment response, and they did not investigate whether the ICS treatment of asthma patients should be elevated to systemic treatment such as oral or intravenous administration.

ICS demonstrates significant efficacy in most asthma patients by improving lung function and preventing acute exacerbation (23). However, approximately 10% of asthma patients exhibit steroid resistance, necessitating the administration of SCS to control asthma attacks (24). Steroid resistance may originate from an increased risk of adverse reactions after long-term use of high-dose corticosteroids or insufficient patient compliance with corticosteroid treatment (25,26). In the present study, 1,088 (19.9%) required treatment escalation to SCS during hospitalization. This notably high percentage suggests potential corticosteroid overuse. In clinical practice, the requirement for SCS indicates poor asthma control, commonly observed in patients with severe asthma or acute exacerbations (27). Similar potential corticosteroid overuse was observed in the temporal validation cohort. Bhattacharya et al. also reported that approximately 30% of patients with severe or uncontrolled asthma received oral corticosteroids (OCS) with a high cumulative dose (≥420 mg/year) (28). Their findings indicated that high-dose OCS treatment correlated with poor compliance to corticosteroid treatment and suboptimal use of inhaled ICS controllers, suggesting possible OCS overuse (28). Hale et al. observed that in asthma patients receiving OCS, high-dose OCS exposure (≥1,600 mg of prednisolone) accounted for 11.3% of the cases (29). However, the efficacy of maintenance treatment remained poor in these patients, impeding proper symptom control. Higher OCS doses may be required for asthma control, while enhanced maintenance treatment might reduce the excessive use of OCS. Most asthma patients who use OCS as a prescription drug require higher treatment steps for symptom control (30). OCS is more frequently utilized for asthma control and treatment compared to controller drugs (such as ICS), leading to OCS overuse (31).

Biologically active corticosteroids exert their effects mainly by binding to their receptors (32). However, multiple factors, including DNA, post-translational modifications (such as phosphorylation, ubiquitination, and acetylation), and cofactor complexes, may influence corticosteroid receptor activity and physiological effects of corticosteroids. Corticosteroid resistance is affected by the HSD3B1 (1,245) genotype (adrenal-restrictive) (32). Abnormal G protein-coupled receptor-related calcium homeostasis and airway smooth muscle shortening may affect the lung function of asthma patients. In the present study, the prediction of treatment escalation primarily utilized variables potentially predictive of corticosteroid treatment response, including demographic and clinical characteristics. These variables demonstrated correlation with corticosteroid response. According to previous research, FENO, total eosinophil count, and IgE levels may influence corticosteroid response, resulting in varying therapeutic effects among asthma patients (33). Recurrent bronchitis, obesity, an elevated FENO level, and the baseline forced expiratory volume in one second (FEV₁) value are documented risk factors for chronic asthma patients requiring SCS treatment, with these patients showing significantly increased probability of developing osteoporosis. Age, sex, environmental exposure, comorbidities, and molecular abnormalities influence corticosteroid treatment response and efficacy. Helper T cell inflammation and delayed diseases are associated with an optimal response of corticosteroids. Patients requiring continuous SCS treatment may have a more complex form of asthma or additional complications, including allergic pneumonia, immunodeficiency, and autoimmune airway diseases. Based on these previous findings, 41 characteristic variables were identified for model development and prediction of treatment escalation probability.

Machine learning algorithms may exhibit over-fitting, demonstrating strong performance with high AUC values in the training set or the validation set, but relatively poor performance in the testing set (34). To address this limitation, in the present study, the derivation set was randomly divided into training (70%) and hold-out internal validation sets (30%). Cross-validation of the training set was used to optimize machine learning algorithm parameters and prevent over-fitting. The random forest model demonstrated superior performance compared to other models and was selected as the final model. This model exhibited good accuracy in predicting treatment escalation in both temporal and internal validation sets. The cutoff value was determined as 0.2. However, current evaluation systems cannot precisely determine the time point for treatment escalation, which is crucial to prevent corticosteroid overuse (35). While our model effectively predicted treatment escalation probability, its limitation was the modest odds ratio, potentially related to overuse. Consequently, a “prospective study” simulation using 62 pairs of patients with false-negative and true-negative predictions was conducted. Outcome comparison revealed increased total hospitalization cost solely in false-negative patients with treatment escalation, with no differences in other outcomes. This finding suggests potential corticosteroid overuse, as false-negative patients showed no benefits from treatment escalation to SCS in terms of outcomes.

There are some limitations in this study. First, as a single-center retrospective study, the findings may be subject to selection bias and limited generalizability; hence, validation through multicenter, prospective cohorts with diverse demographic and environmental backgrounds is necessary. Second, lung function indices such as FEV₁% predicted were not measured. Current guidelines, including GINA 2024, explicitly advise against lung function testing during acute asthma exacerbations (18), as forced expiratory maneuvers may trigger or exacerbate bronchospasm, increase patient distress, and precipitate respiratory failure. Additionally, the retrospective design limited access to reliable pre-admission lung function data. Future research should therefore obtain comprehensive lung function measurements after patient stabilization or during outpatient follow-up to enhance risk stratification and model validation. Third, although the model showed a moderate AUC value, the high NPV (87–89%) supports its utility as a screening tool to exclude unnecessary treatment escalation. Fourth, data regarding medication adherence, inhaler technique, and specific asthma phenotypes (e.g., T2-high vs. T2-low) were unavailable; these factors could independently influence corticosteroid response and represent important targets for future model refinement. Fifth, outcomes were limited to hospitalization indices; hence, prospective follow-up studies are required to evaluate long-term exacerbation rates, corticosteroid-related adverse events, and quality-of-life measures. Finally, the model was developed, trained, and validated entirely on retrospective data; therefore, additional multicenter prospective studies incorporating complete lung function data are warranted to confirm the reliability of the model.

Conclusions

In summary, by utilizing real-world data from 5,463 hospitalized asthma patients, we developed a 41-variable-based random forest model that effectively identifies individuals likely to fail initial ICS therapy and require treatment escalation to SCS. The model demonstrated robust discrimination (AUC =0.75) and a high NPV (88%). A simulated comparison indicated that false-negative patients—those who were predicted to not require treatment escalation but received SCS—showed no clinical benefits while incurring a higher hospitalization cost, suggesting corticosteroid overuse. By incorporating readily available clinical variables, this low-cost tool facilitates precision medicine-driven, individualized decision-making to optimize corticosteroid use in real-world asthma management.

Acknowledgments

We thank the People’s Hospital of the Autonomous Region and the First Affiliated Hospital of Shihezi University for their help in this study. We would like to thank TopEdit (www.topeditsci.com) for its linguistic assistance during the preparation of this manuscript. We would like to thank Yulu Ma and Panpan Wang for their contributions to data collection.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2260/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2260/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2260/prf

Funding: This work was supported by grants from the Corps Guiding Science and Technology Program Projects (No. 2023ZD019) and the Tianshan Innovation Team Program (No. 2022D14006).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2260/coif). C.L. and Y.W. are employed by Yidu Cloud Technology Inc. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics board of the Xinjiang Uygur Autonomous Region People’s Hospital (No. KJ2024-252-01). Informed consent was waived for this retrospective study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Franz T, Stegemann-Koniszewski S, Schreiber J, et al. Metabolic and ionic control of T cells in asthma endotypes. Am J Physiol Cell Physiol 2024;327:C1300-7. [Crossref] [PubMed]
2. Merhej T, Zein JG. Epidemiology of Asthma: Prevalence and Burden of Disease. Adv Exp Med Biol 2023;1426:3-23. [Crossref] [PubMed]
3. Pang K, Li G, Li M, et al. Prevalence and Risk Factors for Allergic Rhinitis in China: A Systematic Review and Meta-Analysis. Evid Based Complement Alternat Med 2022;2022:7165627. [Crossref] [PubMed]
4. Lin JT, Xing B, Tang HP, et al. Seasonal distribution of patient hospitalization due to asthma exacerbation in 7 geographic areas in China. Zhonghua Liu Xing Bing Xue Za Zhi 2018;39:1477-81. [Crossref] [PubMed]
5. Guan WC, Chai RN, Xu C, et al. Analysis on the management of type 2 inflammatory asthma from the guideline of Global Strategy for Asthma Management and Prevention. Zhonghua Yu Fang Yi Xue Za Zhi 2023;57:1964-71. [Crossref] [PubMed]
6. Bleecker ER, Al-Ahmad M, Bjermer L, et al. Systemic corticosteroids in asthma: A call to action from World Allergy Organization and Respiratory Effectiveness Group. World Allergy Organ J 2022;15:100726. [Crossref] [PubMed]
7. Ji H, Han HJ, Choi IY, et al. Association between corticosteroid use and fracture risk in children with asthma: A nationwide cohort study. Pediatr Allergy Immunol 2025;36:e70143. [Crossref] [PubMed]
8. Roger SD, Kertesz T. Administration of corticosteroids into the middle ear does not limit steroid-induced side effects. Aust J Otolaryngol 2025;8:21.
9. Guida G, Bagnasco D, Carriero V, et al. Critical evaluation of asthma biomarkers in clinical practice. Front Med (Lausanne) 2022;9:969243. [Crossref] [PubMed]
10. Menzella F, Ruggiero P, Ghidoni G, et al. Anti-IL5 Therapies for Severe Eosinophilic Asthma: Literature Review and Practical Insights. J Asthma Allergy 2020;13:301-13. [Crossref] [PubMed]
11. Kim SH, Kim Y, Moon JY, et al. Insights and considerations in the use of biologics for managing eosinophilic exacerbations in asthma and chronic obstructive pulmonary disease. J Thorac Dis 2025;17:1795-7. [Crossref] [PubMed]
12. Breiteneder H, Peng YQ, Agache I, et al. Biomarkers for diagnosis and prediction of therapy responses in allergic diseases and asthma. Allergy 2020;75:3039-3068. [Crossref] [PubMed]
13. Kroes JA, Zielhuis SW, van Roon EN, et al. Prediction of response to biological treatment with monoclonal antibodies in severe asthma. Biochem Pharmacol 2020;179:113978. [Crossref] [PubMed]
14. Balasubramanian A, Cervantes C, Gearhart AS, et al. Estimating rate of lung function change using clinical spirometry data. BMJ Open Respir Res 2024;11:e001896. [Crossref] [PubMed]
15. Ong MS, Sordillo JE, Dahlin A, et al. Machine Learning Prediction of Treatment Response to Inhaled Corticosteroids in Asthma. J Pers Med 2024;14:246. [Crossref] [PubMed]
16. Trevor J, Lugogo N, Carr W, et al. Severe asthma exacerbations in the United States:: Incidence, characteristics, predictors, and effects of biologic treatments. Ann Allergy Asthma Immunol 2021;127:579-587.e1. [Crossref] [PubMed]
17. Stanojevic S, Kaminsky DA, Miller MR, et al. ERS/ATS technical standard on interpretive strategies for routine lung function tests. Eur Respir J 2022;60:2101499. [Crossref] [PubMed]
18. Rajvanshi N, Kumar P, Goyal JP. Global Initiative for Asthma Guidelines 2024: An Update. Indian Pediatr 2024;61:781-6.
19. Reinhardt D. ICS/LABA für ehemalige Frühgeborene? MMW Fortschr Med 2022;164:31. [Crossref] [PubMed]
20. Herrera-Luis E, Forno E, Celedón JC, et al. Asthma Exacerbations: The Genes Behind the Scenes. J Investig Allergol Clin Immunol 2023;33:76-94. [Crossref] [PubMed]
21. Busse WW. Improving systemic corticosteroid stewardship in asthma. Eur Respir J 2022;60:2201440. [Crossref] [PubMed]
22. Chung LP, Upham JW, Bardin PG, et al. Rational oral corticosteroid use in adult severe asthma: A narrative review. Respirology 2020;25:161-72. [Crossref] [PubMed]
23. Zheng Z, Li J, Liu Y, et al. Polymorphisms in the FCER2 gene have associations with asthma and chronic obstructive pulmonary disease. J Thorac Dis 2023;15:589-99. [Crossref] [PubMed]
24. Oba Y, Anwer S, Patel T, et al. Addition of long-acting beta2 agonists or long-acting muscarinic antagonists versus doubling the dose of inhaled corticosteroids (ICS) in adolescents and adults with uncontrolled asthma with medium dose ICS: a systematic review and network meta-analysis. Cochrane Database Syst Rev 2023;8:CD013797. [Crossref] [PubMed]
25. Ocejo A, Correa R. Methylprednisolone. Treasure Island (FL): StatPearls Publishing; 2025.
26. Agarwal V, Nangia S, Prasenan S, et al. Drug-Induced Acute Eosinophilic Pneumonia With Pneumomediastinum: An Unusual Presentation. Cureus 2024;16:e64708. [Crossref] [PubMed]
27. Wang YN, Lu SQ, Chen H, et al. Efficacy and safety of dust mite subcutaneous immunotherapy in children with allergic asthma: a prospective randomized controlled study. Zhongguo Dang Dai Er Ke Za Zhi 2024;26:559-66. [Crossref] [PubMed]
28. Bhattacharya A, Syamlal G, Dodd KE. Medical costs and incremental medical costs of asthma among workers in the United States. Am J Ind Med 2024;67:834-43. [Crossref] [PubMed]
29. Hale EM, Greene G, Mulvey C, et al. Use of digital measurement of medication adherence and lung function to guide the management of uncontrolled asthma (INCA Sun): a multicentre, single-blinded, randomised clinical trial. Lancet Respir Med 2023;11:591-601. [Crossref] [PubMed]
30. Ray A, Das J, Wenzel SE. Determining asthma endotypes and outcomes: Complementing existing clinical practice with modern machine learning. Cell Rep Med 2022;3:100857. [Crossref] [PubMed]
31. Versi A, Azim A, Ivan FX, et al. A severe asthma phenotype of excessive airway Haemophilus influenzae relative abundance associated with sputum neutrophilia. Clin Transl Med 2024;14:e70007. [Crossref] [PubMed]
32. Niu H, Li C, Zhang H, et al. Androgen synthesis cell-specific CREBZF deficiency alters adrenal cortex steroid secretion and develops behavioral abnormalities in adult male mice. FASEB J 2024;38:e23650. [Crossref] [PubMed]
33. van Boven JFM, Chapman KR. Triple Therapy of Asthma-Permutations and Combinations. J Allergy Clin Immunol Pract 2022;10:2914-5. [Crossref] [PubMed]
34. Fang C, Ji X, Pan Y, et al. Combining Clinical-Radiomics Features With Machine Learning Methods for Building Models to Predict Postoperative Recurrence in Patients With Chronic Subdural Hematoma: Retrospective Cohort Study. J Med Internet Res 2024;26:e54944. [Crossref] [PubMed]
35. Bruse N, Motos A, van Amstel R, et al. Clinical phenotyping uncovers heterogeneous associations between corticosteroid treatment and survival in critically ill COVID-19 patients. Intensive Care Med 2024;50:1884-96. [Crossref] [PubMed]